Add files using upload-large-folder tool

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2023-2025 Songlin Yang, Yu Zhang
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -0,0 +1,471 @@
+<div align="center">
+
+# 🔥 Flame: Flash Linear Attention Made Easy
+
+</div>
+
+Welcome to 🔥 `flame`, a minimal and efficient framework built on `torchtitan` for training Flash Linear Attention (FLA) models (and more broadly, arbitrary autoregressive language models) with blazing efficiency.
+
+**Feature Highlights:**
+
+- 🚀 Minimal, easy-to-use, extensible training framework
+- 🤗 Seamless integration with `fla` and `transformers`
+- 🔄 Zero-cost data preprocessing: online tokenization, dataset shuffling, and multiple datasets support
+- 🔮 4D parallelism (coming soon)
+
+## Setup
+
+To get started, clone the `flame` repository and install the required dependencies:
+
+```bash
+git clone https://github.com/fla-org/flame.git
+cd flame
+pip install .
+```
+
+`flame` manages minimal dependencies, only including `fla` and `torchtitan` as submodules.
+After installation, initialize and update the submodules:
+```sh
+git submodule update --init --recursive
+```
+
+## Dataset Preparation
+To download the dataset to your local disk, create a new Python file with the following content and execute it:
+
+```py
+from datasets import load_dataset
+
+# load fineweb-edu with parallel processing
+dataset = load_dataset("HuggingFaceFW/fineweb-edu", name="default", num_proc=64, cache_dir="/your/cache/path")
+
+# or load a subset with roughly 100B tokens, suitable for small- or medium-sized experiments
+dataset = load_dataset("HuggingFaceFW/fineweb-edu", name="sample-100BT", num_proc=64, cache_dir="/your/cache/path")
+```
+
+## Training Recipes
+
+Here's an example of training a 340M FLA Transformer model with a LLaMA-like architecture from scratch on a 100BT subset of the Fineweb-edu corpus in streaming mode.
+
+> [!WARNING]
+> If the dataset is not downloaded beforehand, the streaming mode will attempt to fetch it from a remote server and download it on-the-fly, which can be highly unstable during training due to network issues.  
+> For stable training, ensure the dataset is downloaded locally (see [**Dataset Preparation**](#dataset-preparation)). Otherwise, we assume you are only testing the new corpus.
+
+```sh
+bash train.sh \
+  --job.config_file flame/models/fla.toml \
+  --job.dump_folder exp/transformer-340M-4K-10B/batch1.seqlen65536.context4096.warmup1024.update1.steps20480.lr3e-4.cosine \
+  --model.config configs/transformer_340M.json \
+  --model.tokenizer_path fla-hub/transformer-1.3B-100B \
+  --optimizer.name AdamW \
+  --optimizer.eps 1e-15 \
+  --optimizer.lr 3e-4 \
+  --lr_scheduler.warmup_steps 1024 \
+  --lr_scheduler.lr_min 0.1 \
+  --lr_scheduler.decay_type cosine \
+  --training.batch_size 1 \
+  --training.seq_len 65536 \
+  --training.context_len 4096 \
+  --training.varlen \
+  --training.gradient_accumulation_steps 1 \
+  --training.steps 20480 \
+  --training.max_norm 1.0 \
+  --training.skip_nan_inf \
+  --training.dataset HuggingFaceFW/fineweb-edu \
+  --training.dataset_name sample-100BT \
+  --training.dataset_split train \
+  --training.streaming \
+  --training.num_workers 32 \
+  --training.prefetch_factor 2 \
+  --training.seed 42 \
+  --training.compile \
+  --checkpoint.interval 2048 \
+  --checkpoint.load_step -1 \
+  --checkpoint.keep_latest_k 2 \
+  --metrics.log_freq 1
+```
+
+You can specify the number of GPUs by setting the environment variable `NGPU`, which defaults to 8.  
+**For single-GPU debugging, set `NGPU=1`.**
+
+We provide several [config files](https://github.com/fla-org/flame/tree/main/configs) for different models.
+By default, the learning rate is set to 3e-4 with a cosine scheduler. Other schedulers, such as WSD (wsd), are also supported.
+
+**Key parameters:**
+- `--lr_scheduler.decay_ratio`: The proportion of the steps allocated to the decay phase. The learning rate will remain stable after the warmup period and only start decaying during the last `decay_ratio` portion of the total training steps, which is known as the Warmup-Stable-Decay (WSD) schedule.
+- `--lr_scheduler.warmup_steps`: The number of steps for the learning rate warmup phase.
+- `--training.steps`: Total number of training steps.
+- `--training.batch_size`: Batch size per device, must be 1 if `--training.varlen` is set.
+- `--training.seq_len`: The length of each sequence in the batch, which is concatenated from multiple samples.
+- `--training.context_len`: The max allowed length of a sample. For non-varlen mode, this is equivalent to `seq_len`.
+- `--training.varlen`: Whether to conduct variable-length sequence training.
+- `--training.gradient_accumulation_steps`: Number of gradient accumulation steps.
+
+> [!WARNING]
+> The total number of tokens processed per batch, referred to as `global_batch_size`, is calculated as batch_size × gradient_accumulation_steps × num_gpus.
+> Each step processes `global_batch_size * seq_len` tokens.  
+> Monitor the value of `global_batch_size`, `warmup_steps`, and `steps` carefully when modifying any of the hyperparameters!
+
+For a detailed explanation of all parameters, run:
+
+```sh
+bash train.sh -h
+```
+
+<details>
+<summary>Usage</summary>
+
+```py
+options:
+  -h, --help            show this help message and exit
+  --job.config_file JOB.CONFIG_FILE
+                        Job config file
+  --job.dump_folder JOB.DUMP_FOLDER
+                        Folder to dump job outputs
+  --job.description JOB.DESCRIPTION
+                        Description of the job
+  --job.use_for_integration_test
+                        Add this config to the integration test suite
+  --job.print_args      Print the args to terminal
+  --model.config MODEL.CONFIG
+                        Path to the model config
+  --model.norm_type MODEL.NORM_TYPE
+                        Type of layer normalization to use [layernorm,
+                        np_layernorm, rmsnorm, fused_rmsnorm]
+  --model.tokenizer_path MODEL.TOKENIZER_PATH
+                        Tokenizer path
+  --profiling.enable_profiling
+                        Whether to enable pytorch profiler
+  --profiling.save_traces_folder PROFILING.SAVE_TRACES_FOLDER
+                        Trace files location
+  --profiling.profile_freq PROFILING.PROFILE_FREQ
+                        How often to collect profiler traces, in iterations
+  --profiling.enable_memory_snapshot
+                        Whether to dump memory snapshot
+  --profiling.save_memory_snapshot_folder PROFILING.SAVE_MEMORY_SNAPSHOT_FOLDER
+                        Memeory snapshot files location
+  --optimizer.name OPTIMIZER.NAME
+                        Optimizer to use
+  --optimizer.eps OPTIMIZER.EPS
+                        Epsilon value for the optimizer.
+  --optimizer.fused     Whether the fused implementation(CUDA only) is used.
+  --optimizer.scheduler {wsd,cosine,linear}
+                        Scheduler to use. Currently supported: wsd, cosine,
+                        and linear.
+  --optimizer.lr OPTIMIZER.LR
+                        Learning rate to use
+  --optimizer.min_lr_ratio OPTIMIZER.MIN_LR_RATIO
+                        Min lr ratio for lr scheduler
+  --optimizer.early_step_in_backward
+                        Whether to apply optimizer in the backward. Caution,
+                        optimizer_in_backward is not compatible with gradients
+                        clipping, users should not call
+                        register_post_accumulate_grad_hook after the optimizer
+                        is built.
+  --training.batch_size TRAINING.BATCH_SIZE
+                        Batch size
+  --training.seq_len TRAINING.SEQ_LEN
+                        Sequence length
+  --training.context_len TRAINING.CONTEXT_LEN
+                        Max length allowed for each sequence
+  --training.varlen     Whether to take sequences of variable length as input
+  --training.warmup_steps TRAINING.WARMUP_STEPS
+                        Steps for lr scheduler warmup, normally 1/5 of
+                        --training.steps
+  --training.gradient_accumulation_steps TRAINING.GRADIENT_ACCUMULATION_STEPS
+                        Number of steps to accumulate gradients before
+                        updating parameters
+  --training.steps TRAINING.STEPS
+                        How many train steps to run
+  --training.max_norm TRAINING.MAX_NORM
+                        Max norm for gradient clipping
+  --training.skip_nan_inf
+                        Skip batch updates when NaN or INF gradients are
+                        encountered during training
+  --training.dataset TRAINING.DATASET
+                        Dataset to use, with comma separated values
+  --training.dataset_name TRAINING.DATASET_NAME
+                        The name of the dataset config, with comma separated
+                        values if provided
+  --training.dataset_split TRAINING.DATASET_SPLIT
+                        Dataset split to use, with comma separated values if
+                        provided
+  --training.data_dir TRAINING.DATA_DIR
+                        Data dirs to use, with comma separated values if
+                        provided
+  --training.data_files TRAINING.DATA_FILES
+                        Data files to use, with comma separated values if
+                        provided
+  --training.data_probs TRAINING.DATA_PROBS
+                        Data sampling probabilities, with comma separated
+                        values if provided
+  --training.streaming  Whether to load dataset in streaming mode, used for
+                        huge dataset
+  --training.num_workers TRAINING.NUM_WORKERS
+                        Number of subprocesses to use for data loading. 0
+                        means that the data will be loaded in the main
+                        process.
+  --training.prefetch_factor TRAINING.PREFETCH_FACTOR
+                        Number of batches loaded in advance by each worker.2
+                        means there will be a total of 2 * num_workers batches
+                        prefetched across all workers.
+  --training.data_parallel_replicate_degree TRAINING.DATA_PARALLEL_REPLICATE_DEGREE
+                        The `data_parallel_replicate_degree` argument
+                        specifies the degree of data parallelism for weight
+                        replication. When this value is greater than 1,
+                        weights will be replicated across
+                        `data_parallel_replicate_degree` ranks. If
+                        `data_parallel_shard_degree` is also greater than 1,
+                        the parallelism method used is HSDP (Hybrid Sharded
+                        Data Parallelism). Otherwise, the parallelism method
+                        used is DDP (Distributed Data Parallelism). 1 means
+                        disabled.
+  --training.data_parallel_shard_degree TRAINING.DATA_PARALLEL_SHARD_DEGREE
+                        The `data_parallel_shard_degree` argument specifies
+                        the degree of data parallelism for weight sharding.
+                        When this value is greater than 1, weights will be
+                        sharded across `data_parallel_shard_degree` ranks. If
+                        `data_parallel_replicate_degree` is also greater than
+                        1, the parallelism method used is HSDP (Hybrid Sharded
+                        Data Parallelism). Otherwise, the parallelism method
+                        used is FSDP (Fully Sharded Data Parallelism). -1
+                        means leftover ranks will be used (After
+                        DP_REPLICATE/SP/PP). Note that only
+                        `data_parallel_shard_degree` can be negative. 1 means
+                        disabled.
+  --training.enable_cpu_offload
+                        Whether to apply CPU offloading of parameters,
+                        gradients, and optimizer states in FSDP
+  --training.tensor_parallel_degree TRAINING.TENSOR_PARALLEL_DEGREE
+                        Tensor Parallelism degree. 1 means disabled.
+  --training.disable_loss_parallel
+                        Whether to apply loss parallel when sequence parallel
+                        is enabled
+  --training.mixed_precision_param {bfloat16,float32}
+                        torch dtype to use for parameters when applying mixed
+                        precision via FSDP. This feature only takes effect
+                        when data_parallel_shard_degree > 1
+  --training.mixed_precision_reduce {float32}
+                        torch dtype to use for reductions when applying mixed
+                        precision via FSDP. This feature only takes effect
+                        when data_parallel_shard_degree > 1
+  --training.compile    Whether to compile the model
+  --training.gc_freq TRAINING.GC_FREQ
+                        Python garbage control scheduling interval, in steps
+  --training.seed TRAINING.SEED
+                        Choose the base RNG seed used for training
+  --training.deterministic
+                        Use deterministic algorithms wherever possible, may be
+                        slower
+  --metrics.log_freq METRICS.LOG_FREQ
+                        How often to log metrics to TensorBoard, in iterations
+  --metrics.enable_tensorboard
+                        Whether to log metrics to TensorBoard
+  --metrics.disable_color_printing
+                        Whether to disable color printing in logs
+  --metrics.save_tb_folder METRICS.SAVE_TB_FOLDER
+                        Folder to dump TensorBoard states
+  --metrics.rank_0_only
+                        Whether to save TensorBoard metrics only for rank 0 or
+                        for all ranks. When pipeline_parallel_degree is > 1,
+                        this option uses the 0th rank of the last stage
+                        pipeline group, which is the only stage that computes
+                        loss metrics.
+  --metrics.enable_wandb
+                        Whether to log metrics to Weights & Biases
+  --experimental.enable_async_tensor_parallel
+                        Whether to apply async tensor parallel (currently only
+                        effective when compile is enabled)
+  --experimental.pipeline_parallel_degree EXPERIMENTAL.PIPELINE_PARALLEL_DEGREE
+                        Pipeline Parallelism degree, or number of ranks. 1
+                        means disabled. If using looped schedules, this still
+                        specifies the number of physical ranks, not the number
+                        of stages. Stages per rank are inferred from split
+                        points degree, and schedule.
+  --experimental.pipeline_parallel_split_points EXPERIMENTAL.PIPELINE_PARALLEL_SPLIT_POINTS [EXPERIMENTAL.PIPELINE_PARALLEL_SPLIT_POINTS ...]
+                        Specify comma-separated names of modules to use as the
+                        beginning of a split point. e.g. "layers.0,layers.2"
+                        will cause the model to be split into 3 stages, the
+                        first containing all the layers up to layers.0, the
+                        second containing layers.0 and up to layers.2, the
+                        third containing layers.2 and all the remaining
+                        layers. Note: fully-automated splitting may be enabled
+                        in the future, but currently the split points must be
+                        specified manually.
+  --experimental.pipeline_parallel_schedule EXPERIMENTAL.PIPELINE_PARALLEL_SCHEDULE
+                        Specify the Pipeline Parallel schedule to use. The
+                        supported schedules are: https://github.com/pytorch/py
+                        torch/blob/de4c2a3b4e89d96334dc678d1c3f2ae51a6630a0/to
+                        rch/distributed/pipelining/schedules.py#L2161. The
+                        schedule must be compatible with the split points and
+                        stages_per_rank. Looped schedules (e.g.
+                        Interleaved1F1B) require specifying
+                        pipeline_parallel_degree = number of ranks, and
+                        split_points = number of stages - 1
+  --experimental.pipeline_parallel_schedule_csv EXPERIMENTAL.PIPELINE_PARALLEL_SCHEDULE_CSV
+                        Specify the path to the pipeline parallel schedule csv
+                        file to use. The pipeline_parallel_schedule argument
+                        must be either PipelineScheduleSingle,
+                        PipelineScheduleMulti, or _PipelineScheduleRuntime.
+  --experimental.pipeline_parallel_microbatches EXPERIMENTAL.PIPELINE_PARALLEL_MICROBATCHES
+                        How many microbatches to split the global training
+                        batch into when using pipeline parallelism. The global
+                        training batch size must be evenly divisible by the
+                        number of microbatches. The default value will be the
+                        number of pipeline stages, if unspecified.
+  --experimental.enable_compiled_autograd
+                        Enable CompiledAutograd to compile the backward.
+  --experimental.context_parallel_degree EXPERIMENTAL.CONTEXT_PARALLEL_DEGREE
+                        Context parallelism degree. 1 means disabled.
+  --experimental.context_parallel_rotate_method EXPERIMENTAL.CONTEXT_PARALLEL_ROTATE_METHOD
+                        The collective to use in context parallel SDPA for kv
+                        shards exchange. 'allgather' means to all-gather all
+                        kv shards on ranks after the first sub-SDPA
+                        computation, 'alltoall' means to all-to-all shuffle
+                        the kv shards. The default value is 'allgather'.
+  --checkpoint.enable_checkpoint
+                        Whether to enable checkpoint
+  --checkpoint.folder CHECKPOINT.FOLDER
+                        The folder to store the checkpoints. When
+                        enable_checkpoint is set to true, checkpoints will be
+                        in {--job.dump_folder}/{--checkpoint.folder}.
+  --checkpoint.interval_type CHECKPOINT.INTERVAL_TYPE
+                        Checkpointing interval unit of measurement ['step',
+                        'seconds']
+  --checkpoint.interval CHECKPOINT.INTERVAL
+                        Checkpointing interval, in steps or seconds depending
+                        on --checkpoint.interval_type
+  --checkpoint.model_weights_only
+                        When model_weights_only=True, only model weights will
+                        be saved at the end of training. With this,
+                        checkpoints can be loaded using `torch.load(...,
+                        weights_only=True)` after conversion. When
+                        model_weights_only=False, the full checkpoint will be
+                        saved. A full checkpoint includes model, optimizer and
+                        train_state, which can be used to resume training. The
+                        default value is false.
+  --checkpoint.export_dtype {float16,bfloat16,float32}
+                        Converts to the specified precision when training
+                        completes and model_weights_only=true. Currently
+                        supports float32, float16, and bfloat16. The default
+                        value is float32.
+  --checkpoint.create_seed_checkpoint
+                        Initializes the full model without applying
+                        parallelisms, and then saves it as a seed checkpoint.
+                        Note: requires user to call train.py without
+                        specifying any parallelisms, e.g. NGPU=1. Could be
+                        implemented as a separate script, but this way shares
+                        more code.
+  --checkpoint.async_mode CHECKPOINT.ASYNC_MODE
+                        Which async checkpoint mode to use. Currently there
+                        are 3 different modes. 1. "disabled": synchronized
+                        checkpointing will be used. 2. "async":
+                        torch.distributed.checkpoint.async_save will be used.
+                        1. "async_with_pinned_mem": this option utilizes a
+                        dedicated pinned memory space and creates a separate
+                        process for faster GPU->CPU transfer performance and
+                        eliminating GIL contention. The cost is increased CPU
+                        memory usage. If insufficient CPU memory is available,
+                        performance may degrade due to memory paging. For most
+                        users, "async" should suffice as the performance
+                        overhead is typically small (on the order of tens of
+                        seconds) compared to checkpointing frequency. This
+                        mode can be employed to pursue near-zero checkpointing
+                        times (e.g., < 1 second) given appropriate hardware
+                        support such as ample CPU memory and fast PCIe.
+                        "disabled" is the default mode.
+  --checkpoint.keep_latest_k CHECKPOINT.KEEP_LATEST_K
+                        Keeps only the latest k checkpoints, and purging older
+                        ones. If 0, keep all checkpoints. 0 is the default
+                        value.
+  --checkpoint.load_step CHECKPOINT.LOAD_STEP
+                        Load the checkpoint at the specified step. If -1, load
+                        the latest checkpoint.
+  --float8.enable_float8_linear
+                        If true, swaps `torch.nn.Linear` with `Float8Linear`.
+                        This feature requires you to install 'torchao' which
+                        can be found here: https://github.com/pytorch/ao
+  --float8.enable_fsdp_float8_all_gather
+                        Whether enable float8 all-gather in FSDP
+  --float8.precompute_float8_dynamic_scale_for_fsdp
+                        Whether precompute float8 scales dynamically for FSDP
+  --float8.scaling_type_input {dynamic,delayed}
+                        float8 scaling for input, dynamic (default) or delayed
+  --float8.scaling_type_weight FLOAT8.SCALING_TYPE_WEIGHT
+                        float8 scaling for input, dynamic (default) or delayed
+  --float8.scaling_type_grad_output FLOAT8.SCALING_TYPE_GRAD_OUTPUT
+                        float8 scaling for input, dynamic (default) or delayed
+  --comm.init_timeout_seconds COMM.INIT_TIMEOUT_SECONDS
+                        Timeout for communication operations, during
+                        initialization and first train step.
+  --comm.train_timeout_seconds COMM.TRAIN_TIMEOUT_SECONDS
+                        Timeout for communication operations after the first
+                        train step -- usually a tighter bound than during
+                        initialization.
+  --comm.trace_buf_size COMM.TRACE_BUF_SIZE
+                        Flight recorder ring buffer size, >0 means recording
+                        by default, 0 means disabled
+  --memory_estimation.enabled
+                        Whether to estimate memory usage for FSDP
+  --memory_estimation.disable_fake_mode
+                        Whether to estimate memory under FakeTensorMode
+```
+</details>
+
+### Training with `torch.compile`
+
+Starting from `torch 2.0`, `torch.compile` has been introduced as a new feature to seamlessly accelerate training processes.
+In `flame`, one can simply enable `torch.compile` by adding `--training.compile` flag to your training script.
+
+However, `fla` has integrated numerous fused kernels for acceleration, which may potentially conflict with `torch.compile`.
+We are actively working on resolving these issues to make compilation transparent to users.
+In the meantime, please ensure you are using the latest dependencies.
+
+Specifically, **we recommend using `torch>=2.6` and `triton>=3.0`**.
+
+### Training with multiple datasets
+
+If you wish to train a model with all-round capabilities (e.g., code, math, and multilingual ability), it's necessary to train on multiple datasets.
+`flame` allows training with multiple datasets easily.
+For example, you can specify the following arguments to train on 6 datasets with different proportions:
+
+```sh
+  --training.dataset HuggingFaceFW/fineweb-edu,opencsg/Fineweb-Edu-Chinese-V2.1,OpenCoder-LLM/opc-fineweb-code-corpus,math-ai/AutoMathText,EleutherAI/proof-pile-2,OpenCoder-LLM/opc-fineweb-math-corpus   \
+  --training.data_probs 0.6,0.15,0.15,0.014,0.058,0.028     \
+```
+
+### ~Finalizing training~
+
+> [!NOTE]  
+> We have done this conversion automatically in the training script since our latest updates.
+
+Once training is complete, you may want to convert the distributed checkpoints (DCPs) into the 🤗 format for broader use.
+To facilitate this, we provide a straightforward conversion script:
+
+```sh
+python -m flame.utils.convert_dcp_to_hf --path <path_to_model> --step <step> --config <path_to_config> --tokenizer <path_to_tokenizer>
+```
+After this, your model will be in the 🤗 format, ready to be shared or deployed.
+You can then easily publish your model using the `huggingface_hub` for wider accessibility.
+
+### Continual training
+
+If you wish to build upon a strong pre-trained model (in 🤗 format) and continue training, we also offer a script to convert the 🤗 format model back into DCP format.
+This allows you to seamlessly resume training with `flame`.
+```sh
+python -m flame.utils.convert_hf_to_dcp --model <path_to_hf> --checkpoint <path_to_dcp/checkpoint/step-0>
+```
+Here, `<path_to_dcp>` is the directory where your distributed checkpoints will be stored.
+The checkpoint is intentionally saved at `<step-0>` within the checkpoint folder to ensure it is loadable by `flame` during the initial training step, similar to how a seed checkpoint is handled.
+
+Once the conversion is complete, you can proceed with training using `flame` as usual, continuing from where the pretrained model left off.
+
+## Multi-node training
+
+If you have access to multi-node GPUs, consider leveraging them for optimal performance.
+This process is straightforward and well-documented in the PyTorch [docs](https://pytorch.org/docs/stable/elastic/run.html).
+
+To set up multi-node training:
+* Set the environment variables `MASTER_ADDR=<ip>` and `MASTER_PORT=<port>` before running the training script across all nodes.
+* If you're using a job scheduler like Slurm, it will handle these variables for you.
+
+`torchtitan` provides a [Slurm script](https://github.com/pytorch/torchtitan/blob/main/multinode_trainer.slurm) for multi-node training, which you can use as a reference or starting point.

config.json

@@ -0,0 +1,34 @@
+{
+  "architectures": [
+    "MTPTransformerForCausalLM"
+  ],
+  "attention_bias": false,
+  "bos_token_id": 1,
+  "elementwise_affine": true,
+  "eos_token_id": 2,
+  "fuse_cross_entropy": true,
+  "fuse_norm": false,
+  "fuse_swiglu": true,
+  "hidden_act": "swish",
+  "hidden_ratio": 4,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": null,
+  "max_position_embeddings": 4096,
+  "model_type": "mtp_transformer",
+  "n_future_tokens": 4,
+  "norm_eps": 1e-06,
+  "num_heads": 12,
+  "num_hidden_layers": 14,
+  "num_kv_heads": null,
+  "qk_norm": false,
+  "qkv_bias": false,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": true,
+  "torch_dtype": "float32",
+  "transformers_version": "4.51.3",
+  "use_cache": true,
+  "use_custom_backward": false,
+  "vocab_size": 32000,
+  "window_size": null
+}

download_checkpoint.py

@@ -0,0 +1,35 @@
+import argparse
+import os
+from huggingface_hub import HfApi, HfFolder, snapshot_download
+
+def main(args):
+    api = HfApi()
+    token = HfFolder.get_token()
+    experiment_checkpoint_folder = os.path.join(args.experiment_checkpoint_folder, "checkpoint")
+    os.makedirs(
+        experiment_checkpoint_folder,
+        exist_ok=True
+    )
+
+    snapshot_download(
+        repo_id=args.repo_id, 
+        token=token,
+        local_dir=experiment_checkpoint_folder,
+    )
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Download a checkpoint from Hugging Face Hub.")
+    parser.add_argument(
+        "--repo_id",
+        type=str,
+        required=True,
+        help="The repository ID on Hugging Face Hub.",
+    )
+    parser.add_argument(
+        "--experiment_checkpoint_folder",
+        type=str,
+        required=True,
+        help="The local directory to save the downloaded checkpoint.",
+    )
+    args = parser.parse_args()
+    main(args)

fla/layers/bitattn.py

@@ -0,0 +1,192 @@
+# -*- coding: utf-8 -*-
+# Copyright (c) 2023-2025, Songlin Yang, Yu Zhang
+
+from __future__ import annotations
+
+import warnings
+from typing import TYPE_CHECKING, Optional, Tuple
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from einops import rearrange
+from transformers.utils import logging
+
+from fla.modules import RotaryEmbedding
+from fla.modules.fused_bitlinear import FusedBitLinear
+
+if TYPE_CHECKING:
+    from fla.models.utils import Cache
+
+try:
+    from flash_attn import flash_attn_func, flash_attn_varlen_func
+    from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input
+except ImportError:
+    warnings.warn(
+        "Flash Attention is not installed. Please install it via `pip install flash-attn --no-build-isolation`",
+        category=ImportWarning
+    )
+    flash_attn_func = None
+
+logger = logging.get_logger(__name__)
+
+
+class BitAttention(nn.Module):
+
+    def __init__(
+        self,
+        hidden_size: int = 2048,
+        num_heads: int = 32,
+        num_kv_heads: Optional[int] = None,
+        window_size: Optional[int] = None,
+        rope_theta: Optional[float] = 10000.,
+        max_position_embeddings: Optional[int] = None,
+        norm_eps: float = 1e-5,
+        layer_idx: int = None
+    ):
+        super().__init__()
+
+        self.num_heads = num_heads
+        if num_kv_heads is None:
+            self.num_kv_heads = self.num_heads
+        else:
+            self.num_kv_heads = num_kv_heads
+        self.num_kv_groups = num_heads // self.num_kv_heads
+        self.hidden_size = hidden_size
+        self.head_dim = self.hidden_size // self.num_heads
+        self.kv_dim = self.num_kv_heads * self.head_dim
+        self.kv_dim = self.num_kv_heads * self.head_dim
+        self.window_size = window_size
+        self.rope_theta = rope_theta
+        self.max_position_embeddings = max_position_embeddings
+        self.layer_idx = layer_idx
+
+        self.q_proj = FusedBitLinear(self.hidden_size, self.hidden_size, bias=False)
+        self.k_proj = FusedBitLinear(self.hidden_size, self.kv_dim, bias=False)
+        self.v_proj = FusedBitLinear(self.hidden_size, self.kv_dim, bias=False)
+        self.o_proj = FusedBitLinear(self.hidden_size, self.hidden_size, bias=False)
+
+        self.rotary = RotaryEmbedding(dim=self.head_dim, base=self.rope_theta)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        **kwargs,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        if attention_mask is not None:
+            assert len(attention_mask.shape) == 2, (
+                "Expected attention_mask as a 0-1 matrix with shape [batch_size, seq_len] "
+                "for padding purposes (0 indicating padding). "
+                "Arbitrary attention masks of shape [batch_size, seq_len, seq_len] are not allowed."
+            )
+
+        batch_size, q_len, _ = hidden_states.size()
+
+        q = rearrange(self.q_proj(hidden_states), '... (h d) -> ... h d', d=self.head_dim)
+        k = rearrange(self.k_proj(hidden_states), '... (h d) -> ... h d', d=self.head_dim)
+        v = rearrange(self.v_proj(hidden_states), '... (h d) -> ... h d', d=self.head_dim)
+
+        # equivalent to cu_seqlens in `flash_attn`
+        cu_seqlens = kwargs.get('cu_seqlens', None)
+
+        seqlen_offset, max_seqlen = 0, q_len
+        if past_key_values is not None:
+            seqlen_offset = past_key_values.get_seq_length(self.layer_idx)
+            max_seqlen = q.shape[1] + seqlen_offset
+
+            if attention_mask is not None:
+                # to deliminate the offsets of padding tokens
+                seqlen_offset = seqlen_offset + attention_mask.sum(-1) - attention_mask.shape[-1]
+                max_seqlen = q.shape[1] + max(seqlen_offset)
+
+        if self.max_position_embeddings is not None:
+            max_seqlen = max(max_seqlen, self.max_position_embeddings)
+        q, k = self.rotary(q, k, seqlen_offset=seqlen_offset, max_seqlen=max_seqlen, cu_seqlens=cu_seqlens)
+
+        if past_key_values is not None:
+            cache_has_content = past_key_values.get_seq_length(self.layer_idx) > 0
+            k_cached, v_cached = past_key_values.update(
+                attn_state=(k.flatten(-2, -1), v.flatten(-2, -1)),
+                layer_idx=self.layer_idx,
+                offset=q_len,
+                cache_kwargs=dict(window_size=self.window_size)
+            )['attn_state']
+            if cache_has_content:
+                k, v = k_cached, v_cached
+                k = rearrange(k, '... (h d) -> ... h d', d=self.head_dim)
+                v = rearrange(v, '... (h d) -> ... h d', d=self.head_dim)
+
+        if flash_attn_func is None:
+            raise ImportError("Please install Flash Attention via `pip install flash-attn --no-build-isolation` first")
+
+        # Contains at least one padding token in the sequence
+        if attention_mask is not None:
+            q, k, v, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(q, k, v, attention_mask, q_len)
+            cu_seqlens_q, cu_seqlens_k = cu_seq_lens
+            max_seqlen_q, max_seqlen_k = max_seq_lens
+            o = flash_attn_varlen_func(
+                q, k, v,
+                cu_seqlens_q=cu_seqlens_q,
+                cu_seqlens_k=cu_seqlens_k,
+                max_seqlen_q=max_seqlen_q,
+                max_seqlen_k=max_seqlen_k,
+                causal=True,
+                window_size=(-1, -1) if self.window_size is None else (self.window_size-1, 0)
+            )
+            o = pad_input(o, indices_q, batch_size, q_len)
+        elif cu_seqlens is not None:
+            o = flash_attn_varlen_func(
+                q.squeeze(0), k.squeeze(0), v.squeeze(0),
+                cu_seqlens_q=cu_seqlens,
+                cu_seqlens_k=cu_seqlens,
+                max_seqlen_q=max_seqlen,
+                max_seqlen_k=max_seqlen,
+                causal=True,
+                window_size=(-1, -1) if self.window_size is None else (self.window_size-1, 0)
+            ).unsqueeze(0)
+        else:
+            o = flash_attn_func(
+                q, k, v,
+                causal=True,
+                window_size=(-1, -1) if self.window_size is None else (self.window_size-1, 0)
+            )
+        o = o.reshape(batch_size, q_len, -1)
+        o = self.o_proj(o)
+
+        if not output_attentions:
+            attentions = None
+
+        return o, attentions, past_key_values
+
+    def _upad_input(self, q, k, v, attention_mask, q_len):
+        batch_size, seq_len, num_key_value_heads, head_dim = k.shape
+        cache_mask = attention_mask[:, -seq_len:]
+        seqlens = cache_mask.sum(-1, dtype=torch.int32)
+        indices_k = torch.nonzero(cache_mask.flatten(), as_tuple=False).flatten()
+        max_seqlen_k = seqlens.max().item()
+        cu_seqlens_k = F.pad(torch.cumsum(seqlens, dim=0, dtype=torch.int32), (1, 0))
+
+        k = index_first_axis(k.reshape(batch_size * seq_len, num_key_value_heads, head_dim), indices_k)
+        v = index_first_axis(v.reshape(batch_size * seq_len, num_key_value_heads, head_dim), indices_k)
+        if q_len == seq_len:
+            q = index_first_axis(q.reshape(batch_size * seq_len, self.num_heads, head_dim), indices_k)
+            cu_seqlens_q = cu_seqlens_k
+            max_seqlen_q = max_seqlen_k
+            indices_q = indices_k
+        elif q_len == 1:
+            max_seqlen_q = 1
+            # There is a memcpy here, that is very bad.
+            cu_seqlens_q = torch.arange(batch_size + 1, dtype=torch.int32, device=q.device)
+            indices_q = cu_seqlens_q[:-1]
+            q = q.squeeze(1)
+        else:
+            # The -q_len: slice assumes left padding.
+            attention_mask = attention_mask[:, -q_len:]
+            q, indices_q, cu_seqlens_q, max_seqlen_q = unpad_input(q, attention_mask)
+
+        return q, k, v, indices_q, (cu_seqlens_q, cu_seqlens_k), (max_seqlen_q, max_seqlen_k)

fla/layers/lightnet.py

@@ -0,0 +1,210 @@
+# -*- coding: utf-8 -*-
+# Copyright (c) 2023-2025, Songlin Yang, Yu Zhang
+
+# ["You Only Scan Once: Efficient Multi-dimension Sequential Modeling with LightNet"](https://arxiv.org/abs/2405.21022)
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Dict, Optional, Tuple
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from einops import rearrange
+
+from fla.modules import FusedRMSNormGated, ShortConvolution
+from fla.modules.fused_norm_gate import rms_norm_swish_gate_linear
+from fla.ops.gla import chunk_gla, fused_recurrent_gla
+
+if TYPE_CHECKING:
+    from transformers.processing_utils import Unpack
+
+    from fla.models.utils import Cache
+
+
+class LightNetAttention(nn.Module):
+
+    def __init__(
+        self,
+        mode: str = 'chunk',
+        hidden_size: int = 1024,
+        num_heads: Optional[int] = None,
+        expand_ratio: Optional[int] = 128,
+        use_short_conv: bool = False,
+        conv_size: int = 4,
+        conv_bias: bool = False,
+        gate_low_rank_dim: int = 128,
+        elementwise_affine: Optional[bool] = True,
+        norm_eps: float = 1e-5,
+        layer_idx: int = None
+    ) -> LightNetAttention:
+        super().__init__()
+
+        self.mode = mode
+        self.hidden_size = hidden_size
+
+        if expand_ratio is None and num_heads is not None:
+            expand_ratio = hidden_size // num_heads
+        elif expand_ratio is not None and num_heads is None:
+            num_heads = hidden_size // expand_ratio
+        elif expand_ratio is None and num_heads is None:
+            raise RuntimeError("One of `expand_ratio` or `num_heads` should be provided.")
+        self.num_heads = num_heads
+        self.expand_ratio = expand_ratio
+
+        self.use_short_conv = use_short_conv
+        self.conv_size = conv_size
+        self.conv_bias = conv_bias
+
+        self.key_dim = int(self.num_heads * self.expand_ratio)
+        self.value_dim = hidden_size
+        self.gate_low_rank_dim = gate_low_rank_dim
+        self.layer_idx = layer_idx
+
+        assert mode in ['chunk', 'fused_chunk'], f"Not suppoerted mode `{mode}`."
+        assert self.key_dim % num_heads == 0, f"key dim must be divisible by num_heads of {num_heads}"
+        assert self.value_dim % num_heads == 0, f"value dim must be divisible by num_heads of {num_heads}"
+
+        self.head_f_dim = self.expand_ratio
+        self.head_i_dim = self.hidden_size // num_heads
+
+        self.q_proj = nn.Linear(hidden_size, self.key_dim, bias=False)
+        self.k_proj = nn.Linear(hidden_size, self.key_dim, bias=False)
+        self.v_proj = nn.Linear(hidden_size, self.value_dim, bias=False)
+
+        if use_short_conv:
+            self.conv_size = conv_size
+            self.q_conv1d = ShortConvolution(self.key_dim, conv_size, activation=None)
+            self.k_conv1d = ShortConvolution(self.key_dim, conv_size, activation=None)
+            self.v_conv1d = ShortConvolution(self.value_dim, conv_size, activation=None)
+
+        self.g_proj = nn.Sequential(
+            nn.Linear(hidden_size, gate_low_rank_dim, bias=False),
+            nn.Linear(gate_low_rank_dim, hidden_size, bias=False)
+        )
+        self.g_norm = FusedRMSNormGated(
+            hidden_size=hidden_size,
+            elementwise_affine=elementwise_affine,
+            eps=norm_eps
+        )
+        self.o_proj = nn.Linear(self.value_dim, hidden_size, bias=False)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Cache] = None,
+        use_cache: Optional[bool] = False,
+        output_attentions: Optional[bool] = False,
+        **kwargs: Unpack[Dict]
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Cache]]:
+        if attention_mask is not None:
+            assert len(attention_mask.shape) == 2, (
+                "Expected attention_mask as a 0-1 matrix with shape [batch_size, seq_len] "
+                "for padding purposes (0 indicating padding). "
+                "Arbitrary attention masks of shape [batch_size, seq_len, seq_len] are not allowed."
+            )
+
+        # launching the triton kernel for just one token will actually be slower
+        mode = 'fused_recurrent' if hidden_states.shape[1] <= 64 else self.mode
+
+        last_state = None
+        if past_key_values is not None and len(past_key_values) > self.layer_idx:
+            last_state = past_key_values[self.layer_idx]
+
+        cu_seqlens = kwargs.get('cu_seqlens', None)
+        if self.use_short_conv:
+            conv_state_q, conv_state_k, conv_state_v = None, None, None
+            if last_state is not None:
+                conv_state_q, conv_state_k, conv_state_v = last_state['conv_state']
+            conv_mask = attention_mask[:, -hidden_states.shape[1]:] if attention_mask is not None else None
+            q, conv_state_q = self.q_conv1d(
+                x=self.q_proj(hidden_states),
+                mask=conv_mask,
+                cache=conv_state_q,
+                output_final_state=use_cache,
+                cu_seqlens=cu_seqlens
+            )
+            k, conv_state_k = self.k_conv1d(
+                x=self.k_proj(hidden_states),
+                mask=conv_mask,
+                cache=conv_state_k,
+                output_final_state=use_cache,
+                cu_seqlens=cu_seqlens
+            )
+            v, conv_state_v = self.v_conv1d(
+                x=self.v_proj(hidden_states),
+                mask=conv_mask,
+                cache=conv_state_v,
+                output_final_state=use_cache,
+                cu_seqlens=cu_seqlens
+            )
+        else:
+            q = self.q_proj(hidden_states)
+            k = self.k_proj(hidden_states)
+            v = self.v_proj(hidden_states)
+
+        # dealing with left-padding
+        if attention_mask is not None:
+            v = v.mul_(attention_mask[:, -v.shape[-2]:, None])
+
+        q = F.silu(q)
+        q, k = map(lambda x: rearrange(x, '... (h d) -> ... h d', d=self.head_f_dim), (q, k))
+        v = rearrange(v, '... (h d) -> ... h d', d=self.head_i_dim)
+        # TODO: this 2 steps took huge amount of time, which should be optimized
+        z = k.float().logcumsumexp(1)
+
+        if cu_seqlens is not None:
+            raise NotImplementedError("LightNet does not support variable-length sequences for now.")
+        k, g = torch.exp(k - z).to(k.dtype), (torch.cat((z[:, :1], z[:, :-1]), 1) - z).to(k.dtype)
+
+        recurrent_state = last_state['recurrent_state'] if last_state is not None else None
+        if mode == 'fused_recurrent':
+            o, recurrent_state = fused_recurrent_gla(
+                q=q,
+                k=k,
+                v=v,
+                gk=g,
+                initial_state=recurrent_state,
+                output_final_state=use_cache,
+                cu_seqlens=cu_seqlens,
+                head_first=False
+            )
+        elif mode == 'chunk':
+            o, recurrent_state = chunk_gla(
+                q=q,
+                k=k,
+                v=v,
+                g=g,
+                initial_state=recurrent_state,
+                output_final_state=use_cache,
+                cu_seqlens=cu_seqlens,
+                head_first=False
+            )
+        else:
+            raise NotImplementedError(f"Not supported mode `{mode}`.")
+
+        if past_key_values is not None:
+            past_key_values.update(
+                recurrent_state=recurrent_state,
+                conv_state=(conv_state_q, conv_state_k, conv_state_v) if self.use_short_conv else None,
+                layer_idx=self.layer_idx,
+                offset=q.shape[1]
+            )
+
+        o = rms_norm_swish_gate_linear(
+            rearrange(o, 'b t h d -> b t (h d)'),
+            self.g_proj(hidden_states),
+            self.g_norm.weight,
+            self.g_norm.bias,
+            self.o_proj.weight,
+            self.o_proj.bias
+        )
+        return o, None, past_key_values
+
+    def state_size(self, **kwargs) -> int:
+        state_size = self.key_dim * self.head_i_dim
+        for module in self.children():
+            if isinstance(module, ShortConvolution):
+                state_size += module.state_size
+        return state_size

fla/layers/nsa.py

@@ -0,0 +1,138 @@
+# -*- coding: utf-8 -*-
+# Copyright (c) 2023-2025, Songlin Yang, Yu Zhang
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Optional, Tuple, Union
+
+import torch
+import torch.nn as nn
+from einops import rearrange
+from transformers.utils import logging
+
+from fla.modules import RotaryEmbedding
+from fla.ops.nsa.parallel import parallel_nsa
+
+if TYPE_CHECKING:
+    from fla.models.utils import Cache
+
+logger = logging.get_logger(__name__)
+
+
+class NativeSparseAttention(nn.Module):
+
+    def __init__(
+        self,
+        hidden_size: int = 2048,
+        num_heads: int = 64,
+        num_kv_heads: Optional[int] = 4,
+        head_dim: int = 64,
+        qkv_bias: bool = False,
+        block_size: Optional[int] = 64,
+        block_counts: Optional[Union[torch.LongTensor, int]] = 16,
+        window_size: Optional[int] = 512,
+        rope_theta: Optional[float] = 10000.,
+        max_position_embeddings: Optional[int] = None,
+        layer_idx: int = None
+    ):
+        super().__init__()
+
+        self.hidden_size = hidden_size
+        self.num_heads = num_heads
+        if num_kv_heads is None:
+            self.num_kv_heads = self.num_heads
+        else:
+            self.num_kv_heads = num_kv_heads
+        self.num_kv_groups = num_heads // self.num_kv_heads
+        self.head_dim = head_dim
+        self.kv_dim = self.num_kv_heads * self.head_dim
+        self.qkv_bias = qkv_bias
+
+        self.block_size = block_size
+        self.block_counts = block_counts
+        self.window_size = window_size
+        self.rope_theta = rope_theta
+        self.max_position_embeddings = max_position_embeddings
+        self.layer_idx = layer_idx
+
+        self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=self.qkv_bias)
+        self.k_proj = nn.Linear(self.hidden_size, self.kv_dim, bias=self.qkv_bias)
+        self.v_proj = nn.Linear(self.hidden_size, self.kv_dim, bias=self.qkv_bias)
+        self.g_proj = nn.Linear(self.hidden_size, self.num_heads * 3, bias=False)
+        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
+
+        self.rotary = RotaryEmbedding(dim=self.head_dim, base=self.rope_theta)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        **kwargs,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        if attention_mask is not None:
+            assert len(attention_mask.shape) == 2, (
+                "Expected attention_mask as a 0-1 matrix with shape [batch_size, seq_len] "
+                "for padding purposes (0 indicating padding). "
+                "Arbitrary attention masks of shape [batch_size, seq_len, seq_len] are not allowed."
+            )
+
+        batch_size, seq_len, _ = hidden_states.size()
+
+        q = rearrange(self.q_proj(hidden_states), '... (h d) -> ... h d', d=self.head_dim)
+        k = rearrange(self.k_proj(hidden_states), '... (h d) -> ... h d', d=self.head_dim)
+        v = rearrange(self.v_proj(hidden_states), '... (h d) -> ... h d', d=self.head_dim)
+        g = rearrange(self.g_proj(hidden_states), '... (h d) -> ... h d', d=3)
+        g_cmp, g_slc, g_swa = g.sigmoid().unbind(-1)
+
+        cu_seqlens = kwargs.get('cu_seqlens', None)
+
+        seqlen_offset, max_seqlen = 0, seq_len
+        if past_key_values is not None:
+            seqlen_offset = past_key_values.get_seq_length(self.layer_idx)
+            max_seqlen = q.shape[1] + seqlen_offset
+
+            if attention_mask is not None:
+                # to deliminate the offsets of padding tokens
+                seqlen_offset = seqlen_offset + attention_mask.sum(-1) - attention_mask.shape[-1]
+                max_seqlen = q.shape[1] + max(seqlen_offset)
+
+        if self.max_position_embeddings is not None:
+            max_seqlen = max(max_seqlen, self.max_position_embeddings)
+        q, k = self.rotary(q, k, seqlen_offset=seqlen_offset, max_seqlen=max_seqlen, cu_seqlens=cu_seqlens)
+
+        if past_key_values is not None:
+            cache_has_content = past_key_values.get_seq_length(self.layer_idx) > 0
+            k_cached, v_cached = past_key_values.update(
+                attn_state=(k.flatten(-2, -1), v.flatten(-2, -1)),
+                layer_idx=self.layer_idx,
+                offset=seq_len,
+                cache_kwargs=dict(window_size=self.window_size)
+            )['attn_state']
+            if cache_has_content:
+                k, v = k_cached, v_cached
+                k = rearrange(k, '... (h d) -> ... h d', d=self.head_dim)
+                v = rearrange(v, '... (h d) -> ... h d', d=self.head_dim)
+
+        o = parallel_nsa(
+            q=q,
+            k=k,
+            v=v,
+            g_cmp=g_cmp,
+            g_slc=g_slc,
+            g_swa=g_swa,
+            block_size=self.block_size,
+            block_counts=self.block_counts,
+            window_size=self.window_size,
+            cu_seqlens=cu_seqlens,
+            head_first=False
+        )
+        o = o.reshape(batch_size, seq_len, -1)
+        o = self.o_proj(o)
+
+        if not output_attentions:
+            attentions = None
+
+        return o, attentions, past_key_values

fla/layers/rwkv7.py

@@ -0,0 +1,221 @@
+# -*- coding: utf-8 -*-
+# Copyright (c) 2023-2025, Songlin Yang, Yu Zhang
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Optional, Tuple
+
+import torch
+import torch.nn as nn
+from einops import rearrange
+from torch.nn import functional as F
+
+from fla.layers.rwkv6 import LoRA
+from fla.modules import GroupNorm
+from fla.modules.l2norm import l2_norm
+from fla.ops.rwkv7 import chunk_rwkv7, fused_recurrent_rwkv7
+
+if TYPE_CHECKING:
+    from fla.models.utils import Cache
+
+
+class RWKV7Attention(nn.Module):
+
+    def __init__(
+        self,
+        mode: str = 'chunk',
+        hidden_size: int = 1024,
+        head_dim: Optional[int] = 64,
+        num_heads: Optional[int] = None,
+        decay_low_rank_dim: int = 64,
+        gate_low_rank_dim: int = 128,
+        a_low_rank_dim: int = 64,
+        v_low_rank_dim: int = 16,
+        elementwise_affine: Optional[bool] = True,
+        norm_eps: float = 1e-5,
+        layer_idx: int = None,
+        fuse_norm: bool = False,
+        value_dim: int = None,
+        **kwargs
+    ) -> RWKV7Attention:
+        super().__init__()
+
+        self.mode = mode
+        assert mode in ['chunk', 'fused_recurrent'], f"Not supported mode `{mode}`."
+        self.hidden_size = hidden_size
+
+        self.key_dim = hidden_size
+        self.value_dim = value_dim if value_dim is not None else hidden_size
+        if head_dim is None and num_heads is None:
+            raise ValueError("Either `head_dim` or `num_heads` must be specified.")
+        elif head_dim is not None:
+            self.head_dim = head_dim
+            self.num_heads = int(hidden_size // head_dim)
+        elif num_heads is not None:
+            self.head_dim = int(hidden_size // num_heads)
+            self.num_heads = num_heads
+        self.head_v_dim = int(self.value_dim // self.num_heads)
+
+        self.decay_low_rank_dim = decay_low_rank_dim
+        self.gate_low_rank_dim = gate_low_rank_dim
+        self.a_low_rank_dim = a_low_rank_dim
+        self.v_low_rank_dim = v_low_rank_dim
+        self.layer_idx = layer_idx
+        self.fuse_norm = fuse_norm
+
+        self.time_shift = nn.ZeroPad2d((0, 0, 1, -1))
+
+        self.x_x = nn.Parameter(torch.zeros(6, hidden_size))
+
+        self.k_k = nn.Parameter(torch.zeros(self.key_dim))
+        self.k_a = nn.Parameter(torch.zeros(self.key_dim))
+        self.r_k = nn.Parameter(torch.zeros(self.num_heads, self.head_dim))
+
+        self.r_proj = nn.Linear(hidden_size, self.key_dim, bias=False)
+        self.k_proj = nn.Linear(hidden_size, self.key_dim, bias=False)
+        self.v_proj = nn.Linear(hidden_size, self.value_dim, bias=False)
+        self.o_proj = nn.Linear(self.value_dim, hidden_size, bias=False)
+
+        self.w_lora = LoRA(hidden_size, self.key_dim, low_rank_dim=decay_low_rank_dim, activation='tanh')
+        if self.layer_idx != 0:
+            self.v_lora = LoRA(hidden_size, self.value_dim, low_rank_dim=v_low_rank_dim, activation=None)
+        self.a_lora = LoRA(hidden_size, self.key_dim, low_rank_dim=a_low_rank_dim, activation=None)
+        self.g_lora = LoRA(hidden_size, self.value_dim, low_rank_dim=gate_low_rank_dim, activation='sigmoid', bias=False)
+
+        if self.fuse_norm:
+            self.g_norm = GroupNorm(
+                num_groups=self.num_heads,
+                hidden_size=self.value_dim,
+                elementwise_affine=elementwise_affine,
+                eps=self.head_dim*norm_eps,
+                bias=True,
+            )
+        else:
+            self.g_norm = nn.GroupNorm(
+                num_groups=self.num_heads,
+                num_channels=self.value_dim,
+                eps=self.head_dim*norm_eps,
+                affine=elementwise_affine
+            )
+
+        self.apply(self._initialize_weights)
+
+    def _initialize_weights(self, module: nn.Module):
+        if getattr(module, "_is_hf_initialized", False):
+            return
+        if isinstance(module, nn.Linear):
+            nn.init.xavier_uniform_(module.weight, gain=2 ** -2.5)
+            if module.bias is not None:
+                nn.init.zeros_(module.bias)
+        if isinstance(module, nn.Parameter):
+            nn.init.xavier_uniform_(module, gain=2 ** -2.5)
+        module._is_hf_initialized = True
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Cache] = None,
+        use_cache: Optional[bool] = False,
+        output_attentions: Optional[bool] = False,
+        v_first: torch.Tensor = None,
+        **kwargs
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Cache]]:
+        if attention_mask is not None:
+            assert len(attention_mask.shape) == 2, (
+                "Expected attention_mask as a 0-1 matrix with shape [batch_size, seq_len] "
+                "for padding purposes (0 indicating padding). "
+                "Arbitrary attention masks of shape [batch_size, seq_len, seq_len] are not allowed."
+            )
+
+        batch_size, seq_len, _ = hidden_states.shape
+
+        if self.training:
+            # if training, use chunk mode no matter how short the sequence is
+            mode = 'chunk'
+        else:
+            # launching the triton kernel for just one token will actually be slower
+            mode = 'fused_recurrent' if hidden_states.shape[1] <= 64 else self.mode
+
+        last_state = None
+        if past_key_values is not None and len(past_key_values) > self.layer_idx:
+            last_state = past_key_values[self.layer_idx]
+
+        if attention_mask is not None:
+            hidden_states = hidden_states.mul(attention_mask[:, -hidden_states.shape[-2]:, None])
+        if hidden_states.shape[1] == 1 and last_state is not None:
+            shifted = last_state['conv_state'].unsqueeze(1)
+        else:
+            shifted = self.time_shift(hidden_states)
+            if last_state is not None:
+                shifted[:, 0] = last_state['conv_state']
+
+        # [batch_size, seq_len, hidden_size]
+        delta = shifted - hidden_states
+        xr, xw, xk, xv, xa, xg = hidden_states.addcmul(delta, self.x_x.view(6, 1, 1, -1)).unbind(0)
+
+        r = self.r_proj(xr)
+        # -math.exp(-0.5) = -0.6065306597126334
+        # I think .to(torch.float) is unnecessary here, since we calculate lora in bloat16
+        # when we apply sigmoid, bf16 input will not have numerical issue
+        # FIXME: check if we can remove .to(torch.float)
+        w = -0.6065306597126334 * self.w_lora(xw).to(torch.float).sigmoid()
+
+        k = self.k_proj(xk)
+        v = self.v_proj(xv)
+
+        if self.layer_idx == 0:
+            v_first = v
+        else:
+            v = torch.lerp(v, v_first, self.v_lora(xv).sigmoid())
+        a = self.a_lora(xa).sigmoid()
+        g = self.g_lora(xg)
+
+        if self.fuse_norm:
+            kk = l2_norm(rearrange(k * self.k_k, 'b t (h d) -> b t h d', d=self.head_dim))
+        else:
+            kk = F.normalize(rearrange(k * self.k_k, 'b t (h d) -> b t h d', d=self.head_dim), dim=-1, p=2.0)
+
+        k = k.addcmul(k * (a - 1), self.k_a)
+
+        # dealing with left-padding
+        if attention_mask is not None:
+            v = v * attention_mask[:, -v.shape[-2]:, None]
+        r, w, k, a = map(lambda x: rearrange(x, 'b t (h d) -> b t h d', d=self.head_dim), (r, w, k, a))
+        v = rearrange(v, 'b t (h d) -> b t h d', d=self.head_v_dim)
+
+        recurrent_state = last_state['recurrent_state'] if last_state is not None else None
+
+        rwkv7_fn = chunk_rwkv7 if mode == 'chunk' else fused_recurrent_rwkv7
+        cu_seqlens = kwargs.get('cu_seqlens', None)
+        o, recurrent_state = rwkv7_fn(
+            r=r,
+            w=w,
+            k=k,
+            v=v,
+            a=-kk,
+            b=kk * a,
+            scale=1.,
+            initial_state=recurrent_state,
+            output_final_state=use_cache,
+            cu_seqlens=cu_seqlens,
+            head_first=False
+        )
+
+        if past_key_values is not None:
+            past_key_values.update(
+                recurrent_state=recurrent_state,
+                conv_state=hidden_states[:, -1],
+                layer_idx=self.layer_idx,
+                offset=r.shape[1]
+            )
+
+        if self.fuse_norm:
+            o = self.g_norm(rearrange(o, '... h d -> ... (h d)'))
+        else:
+            o = self.g_norm(rearrange(o, 'b t h d -> (b t) (h d)')).view(batch_size, seq_len, -1)
+
+        o = o + ((r * k * self.r_k).sum(-1, keepdim=True) * v).view(batch_size, seq_len, -1)
+        o = self.o_proj(o * g)
+
+        return o, None, past_key_values, v_first

fla/models/delta_net/configuration_delta_net.py

@@ -0,0 +1,91 @@
+# -*- coding: utf-8 -*-
+
+from typing import Dict, Optional
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class DeltaNetConfig(PretrainedConfig):
+
+    model_type = 'delta_net'
+    keys_to_ignore_at_inference = ['past_key_values']
+
+    def __init__(
+        self,
+        attn_mode: str = "chunk",
+        hidden_size: int = 2048,
+        expand_k: int = 1,
+        expand_v: int = 1,
+        use_gate: bool = False,
+        use_short_conv: bool = True,
+        conv_size: int = 4,
+        use_beta: bool = True,
+        use_output_norm: bool = True,
+        num_heads: int = 16,
+        qk_norm: str = 'l2',
+        qk_activation: str = 'silu',
+        max_position_embeddings: int = 2048,
+        hidden_ratio: Optional[int] = 4,
+        intermediate_size: Optional[int] = None,
+        hidden_act: str = "swish",
+        num_hidden_layers: int = 24,
+        norm_eps: float = 1e-6,
+        attn: Optional[Dict] = None,
+        use_cache: bool = True,
+        pad_token_id: int = None,
+        bos_token_id: int = 1,
+        eos_token_id: int = 2,
+        tie_word_embeddings: bool = False,
+        initializer_range: float = 0.006,
+        fuse_norm: bool = True,
+        fuse_swiglu: bool = True,
+        fuse_cross_entropy: bool = True,
+        vocab_size: int = 32000,
+        **kwargs
+    ):
+        self.attn_mode = attn_mode
+        self.hidden_size = hidden_size
+        self.expand_k = expand_k
+        self.expand_v = expand_v
+        self.use_gate = use_gate
+        self.use_short_conv = use_short_conv
+        self.conv_size = conv_size
+        self.use_beta = use_beta
+        self.use_output_norm = use_output_norm
+        self.num_heads = num_heads
+        self.qk_norm = qk_norm
+        self.qk_activation = qk_activation
+        self.max_position_embeddings = max_position_embeddings
+
+        self.hidden_ratio = hidden_ratio
+        self.intermediate_size = intermediate_size
+        self.hidden_act = hidden_act
+        self.num_hidden_layers = num_hidden_layers
+        self.norm_eps = norm_eps
+        self.attn = attn
+        self.use_cache = use_cache
+        self.initializer_range = initializer_range
+        self.fuse_norm = fuse_norm
+        self.fuse_swiglu = fuse_swiglu
+        self.fuse_cross_entropy = fuse_cross_entropy
+        self.vocab_size = vocab_size
+
+        if attn is not None:
+            if not isinstance(attn, Dict):
+                raise ValueError("attn must be a dictionary")
+            if 'layers' not in attn:
+                raise ValueError("Layer indices must be provided to initialize hybrid attention layers")
+            if 'num_heads' not in attn:
+                raise ValueError("Number of heads must be provided to initialize hybrid attention layers")
+            attn['num_kv_heads'] = attn.get('num_kv_heads', attn['num_heads'])
+            attn['qkv_bias'] = attn.get('qkv_bias', False)
+            attn['window_size'] = attn.get('window_size', None)
+            attn['rope_theta'] = attn.get('rope_theta', 10000.)
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )

fla/models/gated_deltanet/configuration_gated_deltanet.py

@@ -0,0 +1,83 @@
+# -*- coding: utf-8 -*-
+
+from typing import Dict, Optional
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class GatedDeltaNetConfig(PretrainedConfig):
+    model_type = 'gated_deltanet'
+    keys_to_ignore_at_inference = ['past_key_values']
+
+    def __init__(
+        self,
+        attn_mode: str = "chunk",
+        hidden_size: int = 2048,
+        expand_v: int = 2,
+        use_gate: bool = True,
+        use_short_conv: bool = True,
+        conv_size: int = 4,
+        head_dim: int = 256,
+        num_heads: int = 6,
+        max_position_embeddings: int = 2048,
+        hidden_ratio: Optional[int] = 4,
+        intermediate_size: Optional[int] = None,
+        hidden_act: str = "swish",
+        num_hidden_layers: int = 21,
+        norm_eps: float = 1e-6,
+        attn: Optional[Dict] = None,
+        use_cache: bool = True,
+        pad_token_id: int = None,
+        bos_token_id: int = 1,
+        eos_token_id: int = 2,
+        tie_word_embeddings: bool = False,
+        initializer_range: float = 0.006,
+        fuse_norm: bool = True,
+        fuse_swiglu: bool = True,
+        fuse_cross_entropy: bool = True,
+        vocab_size: int = 32000,
+        **kwargs
+    ):
+        self.attn_mode = attn_mode
+        self.hidden_size = hidden_size
+        self.expand_v = expand_v
+        self.use_gate = use_gate
+        self.use_short_conv = use_short_conv
+        self.conv_size = conv_size
+        self.head_dim = head_dim
+        self.num_heads = num_heads
+        self.max_position_embeddings = max_position_embeddings
+
+        self.hidden_ratio = hidden_ratio
+        self.intermediate_size = intermediate_size
+        self.hidden_act = hidden_act
+        self.num_hidden_layers = num_hidden_layers
+        self.norm_eps = norm_eps
+        self.attn = attn
+        self.use_cache = use_cache
+        self.initializer_range = initializer_range
+
+        self.fuse_norm = fuse_norm
+        self.fuse_swiglu = fuse_swiglu
+        self.fuse_cross_entropy = fuse_cross_entropy
+        self.vocab_size = vocab_size
+
+        if attn is not None:
+            if not isinstance(attn, Dict):
+                raise ValueError("attn must be a dictionary")
+            if 'layers' not in attn:
+                raise ValueError("Layer indices must be provided to initialize hybrid attention layers")
+            if 'num_heads' not in attn:
+                raise ValueError("Number of heads must be provided to initialize hybrid attention layers")
+            attn['num_kv_heads'] = attn.get('num_kv_heads', attn['num_heads'])
+            attn['qkv_bias'] = attn.get('qkv_bias', False)
+            attn['window_size'] = attn.get('window_size', None)
+            attn['rope_theta'] = attn.get('rope_theta', 10000.)
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )

fla/models/gated_deltanet/modeling_gated_deltanet.py

@@ -0,0 +1,412 @@
+# -*- coding: utf-8 -*-
+
+from __future__ import annotations
+
+import math
+import warnings
+from typing import TYPE_CHECKING, Dict, List, Optional, Tuple, Union
+
+import torch
+import torch.nn as nn
+import torch.utils.checkpoint
+from transformers.generation import GenerationMixin
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import logging
+from transformers.utils.deprecation import deprecate_kwarg
+
+from fla.layers.attn import Attention
+from fla.layers.gated_deltanet import GatedDeltaNet
+from fla.models.gated_deltanet.configuration_gated_deltanet import GatedDeltaNetConfig
+from fla.models.utils import Cache
+from fla.modules import FusedCrossEntropyLoss, FusedLinearCrossEntropyLoss
+from fla.modules import GatedMLP as GatedDeltaNetMLP
+from fla.modules import RMSNorm
+
+if TYPE_CHECKING:
+    from transformers.processing_utils import Unpack
+
+
+logger = logging.get_logger(__name__)
+
+
+class GatedDeltaNetBlock(nn.Module):
+    def __init__(self, config: GatedDeltaNetConfig, layer_idx: int):
+        super().__init__()
+
+        self.config = config
+        self.layer_idx = layer_idx
+
+        self.attn_norm = (RMSNorm if config.fuse_norm else nn.RMSNorm)(config.hidden_size, eps=config.norm_eps)
+        if config.attn is not None and layer_idx in config.attn['layers']:
+            self.attn = Attention(
+                hidden_size=config.hidden_size,
+                num_heads=config.attn['num_heads'],
+                num_kv_heads=config.attn['num_kv_heads'],
+                qkv_bias=config.attn['qkv_bias'],
+                window_size=config.attn['window_size'],
+                rope_theta=config.attn['rope_theta'],
+                max_position_embeddings=config.max_position_embeddings,
+                layer_idx=layer_idx
+            )
+        else:
+            self.attn = GatedDeltaNet(
+                mode=config.attn_mode,
+                hidden_size=config.hidden_size,
+                expand_v=config.expand_v,
+                head_dim=config.head_dim,
+                num_heads=config.num_heads,
+                use_gate=config.use_gate,
+                use_short_conv=config.use_short_conv,
+                conv_size=config.conv_size,
+                norm_eps=config.norm_eps,
+                layer_idx=layer_idx
+            )
+        self.mlp_norm = (RMSNorm if config.fuse_norm else nn.RMSNorm)(config.hidden_size, eps=config.norm_eps)
+        self.mlp = GatedDeltaNetMLP(
+            hidden_size=config.hidden_size,
+            hidden_ratio=config.hidden_ratio,
+            intermediate_size=config.intermediate_size,
+            hidden_act=config.hidden_act,
+            fuse_swiglu=config.fuse_swiglu
+        )
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        use_cache: Optional[bool] = False,
+        output_attentions: Optional[bool] = False,
+        **kwargs: Unpack[Dict]
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        residual = hidden_states
+        hidden_states = self.attn_norm(hidden_states)
+        hidden_states, attentions, past_key_values = self.attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            **kwargs
+        )
+        if self.config.fuse_norm:
+            hidden_states, residual = self.mlp_norm(hidden_states, residual, True)
+        else:
+            hidden_states = residual + hidden_states
+            residual = hidden_states
+            hidden_states = self.mlp_norm(hidden_states)
+        hidden_states = self.mlp(hidden_states, **kwargs)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states, attentions, past_key_values)
+
+        return outputs
+
+
+class GatedDeltaNetPreTrainedModel(PreTrainedModel):
+
+    config_class = GatedDeltaNetConfig
+    base_model_prefix = 'model'
+    supports_gradient_checkpointing = True
+    _no_split_modules = ['GatedDeltaNetBlock']
+    _supports_cache_class = True
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(
+        self,
+        module: nn.Module,
+        prenorm_residual_strategy: Optional[str] = 'rescale',
+        num_residuals_per_layer: int = 2,
+    ):
+        if isinstance(module, (nn.Linear, nn.Conv1d)):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            nn.init.normal_(module.weight, mean=0.0, std=self.config.initializer_range)
+            if module.bias is not None:
+                nn.init.zeros_(module.bias)
+        elif isinstance(module, nn.Embedding):
+            nn.init.normal_(module.weight, mean=0.0, std=self.config.initializer_range)
+        elif hasattr(module, 'reset_parameters'):
+            module.reset_parameters()
+
+        if prenorm_residual_strategy is not None:
+            # Reinitialize selected weights subject to the OpenAI GPT-2 Paper Scheme:
+            #   > A modified initialization which accounts for the accumulation on the residual path with model depth. Scale
+            #   > the weights of residual layers at initialization by a factor of 1/√N where N is the # of residual layers.
+            #   >   -- GPT-2 :: https://openai.com/blog/better-language-models/
+            #
+            # Reference (Megatron-LM): https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/model/gpt_model.py
+            p = None
+            if hasattr(module, 'o_proj'):
+                p = module.o_proj.weight
+            elif hasattr(module, 'down_proj'):
+                p = module.down_proj.weight
+            if p is not None:
+                # Special Scaled Initialization --> There are 2 Layer Norms per Transformer Block
+                # Following Pytorch init, except scale by 1/sqrt(2 * n_layer)
+                # We need to reinit p since this code could be called multiple times
+                # Having just p *= scale would repeatedly scale it down
+                if prenorm_residual_strategy == 'rescale':
+                    nn.init.kaiming_uniform_(p, a=math.sqrt(5))
+                    with torch.no_grad():
+                        p /= math.sqrt(num_residuals_per_layer * self.config.num_hidden_layers)
+                elif prenorm_residual_strategy == 'zero':
+                    nn.init.zeros_(p)
+                else:
+                    raise ValueError(f"Invalid prenorm_residual_strategy: {prenorm_residual_strategy}")
+
+
+class GatedDeltaNetModel(GatedDeltaNetPreTrainedModel):
+
+    def __init__(self, config: GatedDeltaNetConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embeddings = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        self.layers = nn.ModuleList([GatedDeltaNetBlock(config, layer_idx) for layer_idx in range(config.num_hidden_layers)])
+        self.norm = (RMSNorm if config.fuse_norm else nn.RMSNorm)(config.hidden_size, eps=config.norm_eps)
+
+        self.gradient_checkpointing = False
+
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embeddings
+
+    def set_input_embeddings(self, value):
+        self.embeddings = value
+
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,  # noqa
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        **kwargs: Unpack[Dict]
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        if output_attentions:
+            warnings.warn("`GatedDeltaNetModel` does not `output_attentions` now, setting it to `False`.")
+            output_attentions = False
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        use_cache = use_cache if use_cache is not None else (self.config.use_cache if not self.training else False)
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # retrieve input_ids and inputs_embeds
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        if input_ids is None and inputs_embeds is None:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embeddings(input_ids)
+        hidden_states = inputs_embeds
+
+        if use_cache and not isinstance(past_key_values, Cache):
+            past_key_values = Cache.from_legacy_cache(past_key_values)
+
+        if self.gradient_checkpointing and self.training and use_cache:
+            logger.warning_once("`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...")
+            use_cache = False
+
+        all_hidden_states = () if output_hidden_states else None
+        all_attns = () if output_attentions else None
+        for layer in self.layers:
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+                hidden_states, attentions, past_key_values = self._gradient_checkpointing_func(
+                    layer.__call__,
+                    hidden_states,
+                    attention_mask,
+                    past_key_values,
+                    use_cache,
+                    output_attentions,
+                    **kwargs
+                )
+            else:
+                hidden_states, attentions, past_key_values = layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    past_key_values=past_key_values,
+                    use_cache=use_cache,
+                    output_attentions=output_attentions,
+                    **kwargs
+                )
+
+            if output_attentions:
+                all_attns += (attentions,)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        if not return_dict:
+            return tuple(i for i in [hidden_states, past_key_values, all_hidden_states, all_attns] if i is not None)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+            hidden_states=all_hidden_states,
+            attentions=all_attns
+        )
+
+
+class GatedDeltaNetForCausalLM(GatedDeltaNetPreTrainedModel, GenerationMixin):
+
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = GatedDeltaNetModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+        self.criterion = None
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embeddings
+
+    def set_input_embeddings(self, value):
+        self.model.embeddings = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    def generate(self, *args, **kwargs):
+        try:
+            return super().generate(*args, **kwargs)
+        except AttributeError as exception:
+            if 'past_key_values' in str(exception):
+                raise AttributeError(
+                    f"You tried to call `generate` with a decoding strategy that manipulates `past_key_values`, "
+                    f"which is not supported for {self.__class__.__name__}. "
+                    f"Try another generation strategy instead. "
+                    f"For the available generation strategies, check this doc: "
+                    f"https://huggingface.co/docs/transformers/en/generation_strategies#decoding-strategies"
+                )
+            else:
+                raise exception
+
+    @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
+    def prepare_inputs_for_generation(
+        self,
+        input_ids: torch.LongTensor = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        use_cache: bool = True,
+        logits_to_keep: Optional[int] = None,
+        **kwargs
+    ):
+        # only last token for `inputs_ids` if the `past_key_values` is not empty.
+        if past_key_values is not None and len(past_key_values) > 0:
+            input_ids = input_ids[:, -1:]
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and len(past_key_values) == 0:
+            model_inputs = {'inputs_embeds': inputs_embeds}
+        else:
+            # The `contiguous()` here is necessary to have a static stride during decoding. torchdynamo otherwise
+            # recompiles graphs as the stride of the inputs is a guard.
+            # Ref: https://github.com/huggingface/transformers/pull/29114
+            # TODO: use `next_tokens` directly instead.
+            model_inputs = {'input_ids': input_ids.contiguous()}
+
+        if logits_to_keep is not None:
+            model_inputs['logits_to_keep'] = logits_to_keep
+
+        model_inputs.update({
+            'past_key_values': past_key_values,
+            'use_cache': use_cache,
+            'attention_mask': attention_mask,
+        })
+        return model_inputs
+
+    @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        logits_to_keep: Optional[int] = 0,
+        **kwargs: Unpack[Dict]
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            inputs_embeds=inputs_embeds,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            **kwargs
+        )
+
+        hidden_states = outputs[0]
+        fuse_linear_and_cross_entropy = self.config.fuse_cross_entropy and self.training
+
+        loss, logits = None, None
+        if not fuse_linear_and_cross_entropy or labels is None:
+            logits = self.lm_head(hidden_states if logits_to_keep is None else hidden_states[:, -logits_to_keep:])
+        if labels is not None:
+            if getattr(self, 'criterion', None) is None:
+                if fuse_linear_and_cross_entropy:
+                    criterion = FusedLinearCrossEntropyLoss()
+                elif self.config.fuse_cross_entropy:
+                    criterion = FusedCrossEntropyLoss(inplace_backward=True)
+                else:
+                    criterion = nn.CrossEntropyLoss()
+            else:
+                criterion = self.criterion
+            labels = labels.to(hidden_states.device)
+            labels = torch.cat((labels[..., 1:], torch.full_like(labels[:, :1], criterion.ignore_index)), 1)
+            if fuse_linear_and_cross_entropy:
+                loss = criterion(hidden_states, labels, self.lm_head.weight, self.lm_head.bias)
+            else:
+                loss = criterion(logits.view(labels.numel(), -1), labels.view(-1))
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )

fla/models/gla/configuration_gla.py

@@ -0,0 +1,95 @@
+# -*- coding: utf-8 -*-
+
+from typing import Dict, Optional
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class GLAConfig(PretrainedConfig):
+
+    model_type = 'gla'
+    keys_to_ignore_at_inference = ['past_key_values']
+
+    def __init__(
+        self,
+        hidden_size: int = 2048,
+        expand_k: int = 0.5,
+        expand_v: int = 1,
+        hidden_ratio: Optional[int] = 4,
+        intermediate_size: Optional[int] = None,
+        num_hidden_layers: int = 24,
+        num_heads: int = 4,
+        num_kv_heads: Optional[int] = None,
+        feature_map: Optional[str] = None,
+        attn_mode: str = "chunk",
+        use_short_conv: bool = False,
+        conv_size: int = 4,
+        use_output_gate: bool = True,
+        clamp_min: Optional[float] = None,
+        hidden_act: str = "swish",
+        max_position_embeddings: int = 2048,
+        elementwise_affine: Optional[bool] = True,
+        norm_eps: float = 1e-6,
+        use_gk: bool = True,
+        use_gv: bool = False,
+        attn: Optional[Dict] = None,
+        use_cache: bool = True,
+        pad_token_id: int = None,
+        bos_token_id: int = 1,
+        eos_token_id: int = 2,
+        tie_word_embeddings: bool = False,
+        initializer_range: float = 0.006,
+        fuse_norm: bool = True,
+        fuse_swiglu: bool = True,
+        fuse_cross_entropy: bool = True,
+        vocab_size: int = 32000,
+        **kwargs
+    ):
+        self.hidden_size = hidden_size
+        self.expand_k = expand_k
+        self.expand_v = expand_v
+        self.hidden_ratio = hidden_ratio
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_heads = num_heads
+        self.num_kv_heads = num_kv_heads
+        self.feature_map = feature_map
+        self.attn_mode = attn_mode
+        self.use_short_conv = use_short_conv
+        self.conv_size = conv_size
+        self.use_output_gate = use_output_gate
+        self.clamp_min = clamp_min
+        self.hidden_act = hidden_act
+        self.max_position_embeddings = max_position_embeddings
+        self.elementwise_affine = elementwise_affine
+        self.norm_eps = norm_eps
+        self.use_gk = use_gk
+        self.use_gv = use_gv
+        self.attn = attn
+        self.use_cache = use_cache
+        self.initializer_range = initializer_range
+
+        self.fuse_norm = fuse_norm
+        self.fuse_swiglu = fuse_swiglu
+        self.fuse_cross_entropy = fuse_cross_entropy
+        self.vocab_size = vocab_size
+
+        if attn is not None:
+            if not isinstance(attn, Dict):
+                raise ValueError("attn must be a dictionary")
+            if 'layers' not in attn:
+                raise ValueError("Layer indices must be provided to initialize hybrid attention layers")
+            if 'num_heads' not in attn:
+                raise ValueError("Number of heads must be provided to initialize hybrid attention layers")
+            attn['num_kv_heads'] = attn.get('num_kv_heads', attn['num_heads'])
+            attn['qkv_bias'] = attn.get('qkv_bias', False)
+            attn['window_size'] = attn.get('window_size', None)
+            attn['rope_theta'] = attn.get('rope_theta', 10000.)
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )

fla/models/gla/modeling_gla.py

@@ -0,0 +1,417 @@
+# -*- coding: utf-8 -*-
+
+from __future__ import annotations
+
+import math
+import warnings
+from typing import TYPE_CHECKING, Dict, List, Optional, Tuple, Union
+
+import torch
+import torch.nn as nn
+import torch.utils.checkpoint
+from transformers.generation import GenerationMixin
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import logging
+from transformers.utils.deprecation import deprecate_kwarg
+
+from fla.layers.attn import Attention
+from fla.layers.gla import GatedLinearAttention
+from fla.models.gla.configuration_gla import GLAConfig
+from fla.models.utils import Cache
+from fla.modules import FusedCrossEntropyLoss, FusedLinearCrossEntropyLoss
+from fla.modules import GatedMLP as GLAMLP
+from fla.modules import RMSNorm
+
+if TYPE_CHECKING:
+    from transformers.processing_utils import Unpack
+
+logger = logging.get_logger(__name__)
+
+
+class GLABlock(nn.Module):
+    def __init__(self, config: GLAConfig, layer_idx: int):
+        super().__init__()
+
+        self.config = config
+        self.layer_idx = layer_idx
+
+        self.attn_norm = (RMSNorm if config.fuse_norm else nn.RMSNorm)(config.hidden_size, eps=config.norm_eps)
+        if config.attn is not None and layer_idx in config.attn['layers']:
+            self.attn = Attention(
+                hidden_size=config.hidden_size,
+                num_heads=config.attn['num_heads'],
+                num_kv_heads=config.attn['num_kv_heads'],
+                qkv_bias=config.attn['qkv_bias'],
+                window_size=config.attn['window_size'],
+                rope_theta=config.attn['rope_theta'],
+                max_position_embeddings=config.max_position_embeddings,
+                layer_idx=layer_idx
+            )
+        else:
+            self.attn = GatedLinearAttention(
+                mode=config.attn_mode,
+                hidden_size=config.hidden_size,
+                expand_k=config.expand_k,
+                expand_v=config.expand_v,
+                num_heads=config.num_heads,
+                num_kv_heads=config.num_kv_heads,
+                feature_map=config.feature_map,
+                use_short_conv=config.use_short_conv,
+                conv_size=config.conv_size,
+                use_output_gate=config.use_output_gate,
+                gate_fn=config.hidden_act,
+                elementwise_affine=config.elementwise_affine,
+                norm_eps=config.norm_eps,
+                clamp_min=config.clamp_min,
+                fuse_norm=config.fuse_norm,
+                layer_idx=layer_idx
+            )
+        self.mlp_norm = (RMSNorm if config.fuse_norm else nn.RMSNorm)(config.hidden_size, eps=config.norm_eps)
+        self.mlp = GLAMLP(
+            hidden_size=config.hidden_size,
+            hidden_ratio=config.hidden_ratio,
+            intermediate_size=config.intermediate_size,
+            hidden_act=config.hidden_act,
+            fuse_swiglu=config.fuse_swiglu
+        )
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        use_cache: Optional[bool] = False,
+        output_attentions: Optional[bool] = False,
+        **kwargs: Unpack[Dict]
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        residual = hidden_states
+        hidden_states = self.attn_norm(hidden_states)
+        hidden_states, attentions, past_key_values = self.attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            **kwargs
+        )
+        if self.config.fuse_norm:
+            hidden_states, residual = self.mlp_norm(hidden_states, residual, True)
+        else:
+            hidden_states = residual + hidden_states
+            residual = hidden_states
+            hidden_states = self.mlp_norm(hidden_states)
+        hidden_states = self.mlp(hidden_states, **kwargs)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states, attentions, past_key_values)
+
+        return outputs
+
+
+class GLAPreTrainedModel(PreTrainedModel):
+
+    config_class = GLAConfig
+    base_model_prefix = 'model'
+    supports_gradient_checkpointing = True
+    _no_split_modules = ['GLABlock']
+    _supports_cache_class = True
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(
+        self,
+        module: nn.Module,
+        prenorm_residual_strategy: Optional[str] = 'rescale',
+        num_residuals_per_layer: int = 2,
+    ):
+        if isinstance(module, (nn.Linear, nn.Conv1d)):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            nn.init.normal_(module.weight, mean=0.0, std=self.config.initializer_range)
+            if module.bias is not None:
+                nn.init.zeros_(module.bias)
+        elif isinstance(module, nn.Embedding):
+            nn.init.normal_(module.weight, mean=0.0, std=self.config.initializer_range)
+        elif hasattr(module, 'reset_parameters'):
+            module.reset_parameters()
+
+        if prenorm_residual_strategy is not None:
+            # Reinitialize selected weights subject to the OpenAI GPT-2 Paper Scheme:
+            #   > A modified initialization which accounts for the accumulation on the residual path with model depth. Scale
+            #   > the weights of residual layers at initialization by a factor of 1/√N where N is the # of residual layers.
+            #   >   -- GPT-2 :: https://openai.com/blog/better-language-models/
+            #
+            # Reference (Megatron-LM): https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/model/gpt_model.py
+            p = None
+            if hasattr(module, 'o_proj'):
+                p = module.o_proj.weight
+            elif hasattr(module, 'down_proj'):
+                p = module.down_proj.weight
+            if p is not None:
+                # Special Scaled Initialization --> There are 2 Layer Norms per Transformer Block
+                # Following Pytorch init, except scale by 1/sqrt(2 * n_layer)
+                # We need to reinit p since this code could be called multiple times
+                # Having just p *= scale would repeatedly scale it down
+                if prenorm_residual_strategy == 'rescale':
+                    nn.init.kaiming_uniform_(p, a=math.sqrt(5))
+                    with torch.no_grad():
+                        p /= math.sqrt(num_residuals_per_layer * self.config.num_hidden_layers)
+                elif prenorm_residual_strategy == 'zero':
+                    nn.init.zeros_(p)
+                else:
+                    raise ValueError(f"Invalid prenorm_residual_strategy: {prenorm_residual_strategy}")
+
+
+class GLAModel(GLAPreTrainedModel):
+
+    def __init__(self, config: GLAConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embeddings = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        self.layers = nn.ModuleList([GLABlock(config, layer_idx) for layer_idx in range(config.num_hidden_layers)])
+        self.norm = (RMSNorm if config.fuse_norm else nn.RMSNorm)(config.hidden_size, eps=config.norm_eps)
+
+        self.gradient_checkpointing = False
+
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embeddings
+
+    def set_input_embeddings(self, value):
+        self.embeddings = value
+
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,  # noqa
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        **kwargs: Unpack[Dict]
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        if output_attentions:
+            warnings.warn("`GLAModel` does not `output_attentions` now, setting it to `False`.")
+            output_attentions = False
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        use_cache = use_cache if use_cache is not None else (self.config.use_cache if not self.training else False)
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # retrieve input_ids and inputs_embeds
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        if input_ids is None and inputs_embeds is None:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embeddings(input_ids)
+        hidden_states = inputs_embeds
+
+        if use_cache and not isinstance(past_key_values, Cache):
+            past_key_values = Cache.from_legacy_cache(past_key_values)
+
+        if self.gradient_checkpointing and self.training and use_cache:
+            logger.warning_once("`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...")
+            use_cache = False
+
+        all_hidden_states = () if output_hidden_states else None
+        all_attns = () if output_attentions else None
+        for layer in self.layers:
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+                hidden_states, attentions, past_key_values = self._gradient_checkpointing_func(
+                    layer.__call__,
+                    hidden_states,
+                    attention_mask,
+                    past_key_values,
+                    use_cache,
+                    output_attentions,
+                    **kwargs
+                )
+            else:
+                hidden_states, attentions, past_key_values = layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    past_key_values=past_key_values,
+                    use_cache=use_cache,
+                    output_attentions=output_attentions,
+                    **kwargs
+                )
+
+            if output_attentions:
+                all_attns += (attentions,)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        if not return_dict:
+            return tuple(i for i in [hidden_states, past_key_values, all_hidden_states, all_attns] if i is not None)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+            hidden_states=all_hidden_states,
+            attentions=all_attns
+        )
+
+
+class GLAForCausalLM(GLAPreTrainedModel, GenerationMixin):
+
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = GLAModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+        self.criterion = None
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embeddings
+
+    def set_input_embeddings(self, value):
+        self.model.embeddings = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    def generate(self, *args, **kwargs):
+        try:
+            return super().generate(*args, **kwargs)
+        except AttributeError as exception:
+            if 'past_key_values' in str(exception):
+                raise AttributeError(
+                    f"You tried to call `generate` with a decoding strategy that manipulates `past_key_values`, "
+                    f"which is not supported for {self.__class__.__name__}. "
+                    f"Try another generation strategy instead. "
+                    f"For the available generation strategies, check this doc: "
+                    f"https://huggingface.co/docs/transformers/en/generation_strategies#decoding-strategies"
+                )
+            else:
+                raise exception
+
+    @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
+    def prepare_inputs_for_generation(
+        self,
+        input_ids: torch.LongTensor = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        use_cache: bool = True,
+        logits_to_keep: Optional[int] = None,
+        **kwargs
+    ):
+        # only last token for `inputs_ids` if the `past_key_values` is not empty.
+        if past_key_values is not None and len(past_key_values) > 0:
+            input_ids = input_ids[:, -1:]
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and len(past_key_values) == 0:
+            model_inputs = {'inputs_embeds': inputs_embeds}
+        else:
+            # The `contiguous()` here is necessary to have a static stride during decoding. torchdynamo otherwise
+            # recompiles graphs as the stride of the inputs is a guard.
+            # Ref: https://github.com/huggingface/transformers/pull/29114
+            # TODO: use `next_tokens` directly instead.
+            model_inputs = {'input_ids': input_ids.contiguous()}
+
+        if logits_to_keep is not None:
+            model_inputs['logits_to_keep'] = logits_to_keep
+
+        model_inputs.update({
+            'past_key_values': past_key_values,
+            'use_cache': use_cache,
+            'attention_mask': attention_mask,
+        })
+        return model_inputs
+
+    @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        logits_to_keep: Optional[int] = 0,
+        **kwargs: Unpack[Dict]
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            inputs_embeds=inputs_embeds,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            **kwargs
+        )
+
+        hidden_states = outputs[0]
+        fuse_linear_and_cross_entropy = self.config.fuse_cross_entropy and self.training
+
+        loss, logits = None, None
+        if not fuse_linear_and_cross_entropy or labels is None:
+            logits = self.lm_head(hidden_states if logits_to_keep is None else hidden_states[:, -logits_to_keep:])
+        if labels is not None:
+            if getattr(self, 'criterion', None) is None:
+                if fuse_linear_and_cross_entropy:
+                    criterion = FusedLinearCrossEntropyLoss()
+                elif self.config.fuse_cross_entropy:
+                    criterion = FusedCrossEntropyLoss(inplace_backward=True)
+                else:
+                    criterion = nn.CrossEntropyLoss()
+            else:
+                criterion = self.criterion
+            labels = labels.to(hidden_states.device)
+            labels = torch.cat((labels[..., 1:], torch.full_like(labels[:, :1], criterion.ignore_index)), 1)
+            if fuse_linear_and_cross_entropy:
+                loss = criterion(hidden_states, labels, self.lm_head.weight, self.lm_head.bias)
+            else:
+                loss = criterion(logits.view(labels.numel(), -1), labels.view(-1))
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )

fla/models/gsa/__init__.py

@@ -0,0 +1,13 @@
+# -*- coding: utf-8 -*-
+
+from transformers import AutoConfig, AutoModel, AutoModelForCausalLM
+
+from fla.models.gsa.configuration_gsa import GSAConfig
+from fla.models.gsa.modeling_gsa import GSAForCausalLM, GSAModel
+
+AutoConfig.register(GSAConfig.model_type, GSAConfig)
+AutoModel.register(GSAConfig, GSAModel)
+AutoModelForCausalLM.register(GSAConfig, GSAForCausalLM)
+
+
+__all__ = ['GSAConfig', 'GSAForCausalLM', 'GSAModel']

fla/models/hgrn/__init__.py

@@ -0,0 +1,13 @@
+# -*- coding: utf-8 -*-
+
+from transformers import AutoConfig, AutoModel, AutoModelForCausalLM
+
+from fla.models.hgrn.configuration_hgrn import HGRNConfig
+from fla.models.hgrn.modeling_hgrn import HGRNForCausalLM, HGRNModel
+
+AutoConfig.register(HGRNConfig.model_type, HGRNConfig)
+AutoModel.register(HGRNConfig, HGRNModel)
+AutoModelForCausalLM.register(HGRNConfig, HGRNForCausalLM)
+
+
+__all__ = ['HGRNConfig', 'HGRNForCausalLM', 'HGRNModel']

fla/models/lightnet/__init__.py

@@ -0,0 +1,13 @@
+# -*- coding: utf-8 -*-
+
+from transformers import AutoConfig, AutoModel, AutoModelForCausalLM
+
+from fla.models.lightnet.configuration_lightnet import LightNetConfig
+from fla.models.lightnet.modeling_lightnet import LightNetForCausalLM, LightNetModel
+
+AutoConfig.register(LightNetConfig.model_type, LightNetConfig)
+AutoModel.register(LightNetConfig, LightNetModel)
+AutoModelForCausalLM.register(LightNetConfig, LightNetForCausalLM)
+
+
+__all__ = ['LightNetConfig', 'LightNetForCausalLM', 'LightNetModel']

fla/models/lightnet/modeling_lightnet.py

@@ -0,0 +1,410 @@
+# -*- coding: utf-8 -*-
+
+from __future__ import annotations
+
+import math
+import warnings
+from typing import TYPE_CHECKING, Dict, List, Optional, Tuple, Union
+
+import torch
+import torch.nn as nn
+import torch.utils.checkpoint
+from transformers.generation import GenerationMixin
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import logging
+from transformers.utils.deprecation import deprecate_kwarg
+
+from fla.layers.attn import Attention
+from fla.layers.lightnet import LightNetAttention
+from fla.models.lightnet.configuration_lightnet import LightNetConfig
+from fla.models.utils import Cache
+from fla.modules import FusedCrossEntropyLoss, FusedLinearCrossEntropyLoss
+from fla.modules import GatedMLP as LightNetMLP
+from fla.modules import RMSNorm
+
+if TYPE_CHECKING:
+    from transformers.processing_utils import Unpack
+
+logger = logging.get_logger(__name__)
+
+
+class LightNetBlock(nn.Module):
+    def __init__(self, config: LightNetConfig, layer_idx: int):
+        super().__init__()
+
+        self.config = config
+        self.layer_idx = layer_idx
+
+        self.attn_norm = (RMSNorm if config.fuse_norm else nn.RMSNorm)(config.hidden_size, eps=config.norm_eps)
+        if config.attn is not None and layer_idx in config.attn['layers']:
+            self.attn = Attention(
+                hidden_size=config.hidden_size,
+                num_heads=config.attn['num_heads'],
+                num_kv_heads=config.attn['num_kv_heads'],
+                qkv_bias=config.attn['qkv_bias'],
+                window_size=config.attn['window_size'],
+                max_position_embeddings=config.max_position_embeddings,
+                layer_idx=layer_idx
+            )
+        else:
+            self.attn = LightNetAttention(
+                mode=config.attn_mode,
+                hidden_size=config.hidden_size,
+                num_heads=config.num_heads,
+                expand_ratio=config.expand_ratio,
+                use_short_conv=config.use_short_conv,
+                conv_size=config.conv_size,
+                gate_low_rank_dim=config.gate_low_rank_dim,
+                elementwise_affine=config.elementwise_affine,
+                norm_eps=config.norm_eps,
+                layer_idx=layer_idx
+            )
+        self.mlp_norm = (RMSNorm if config.fuse_norm else nn.RMSNorm)(config.hidden_size, eps=config.norm_eps)
+        self.mlp = LightNetMLP(
+            hidden_size=config.hidden_size,
+            hidden_ratio=config.hidden_ratio,
+            intermediate_size=config.intermediate_size,
+            hidden_act=config.hidden_act,
+            fuse_swiglu=config.fuse_swiglu
+        )
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        use_cache: Optional[bool] = False,
+        output_attentions: Optional[bool] = False,
+        **kwargs: Unpack[Dict]
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        residual = hidden_states
+        hidden_states = self.attn_norm(hidden_states)
+        hidden_states, attentions, past_key_values = self.attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            **kwargs
+        )
+        if self.config.fuse_norm:
+            hidden_states, residual = self.mlp_norm(hidden_states, residual, True)
+        else:
+            hidden_states = residual + hidden_states
+            residual = hidden_states
+            hidden_states = self.mlp_norm(hidden_states)
+        hidden_states = self.mlp(hidden_states, **kwargs)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states, attentions, past_key_values)
+
+        return outputs
+
+
+class LightNetPreTrainedModel(PreTrainedModel):
+
+    config_class = LightNetConfig
+    supports_gradient_checkpointing = True
+    _no_split_modules = ['LightNetBlock']
+    _supports_cache_class = True
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(
+        self,
+        module: nn.Module,
+        prenorm_residual_strategy: Optional[str] = 'rescale',
+        num_residuals_per_layer: int = 2,
+    ):
+        if isinstance(module, (nn.Linear, nn.Conv1d)):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            nn.init.normal_(module.weight, mean=0.0, std=self.config.initializer_range)
+            if module.bias is not None:
+                nn.init.zeros_(module.bias)
+        elif isinstance(module, nn.Embedding):
+            nn.init.normal_(module.weight, mean=0.0, std=self.config.initializer_range)
+        elif hasattr(module, 'reset_parameters'):
+            module.reset_parameters()
+
+        if prenorm_residual_strategy is not None:
+            # Reinitialize selected weights subject to the OpenAI GPT-2 Paper Scheme:
+            #   > A modified initialization which accounts for the accumulation on the residual path with model depth. Scale
+            #   > the weights of residual layers at initialization by a factor of 1/√N where N is the # of residual layers.
+            #   >   -- GPT-2 :: https://openai.com/blog/better-language-models/
+            #
+            # Reference (Megatron-LM): https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/model/gpt_model.py
+            p = None
+            if hasattr(module, 'o_proj'):
+                p = module.o_proj.weight
+            elif hasattr(module, 'down_proj'):
+                p = module.down_proj.weight
+            if p is not None:
+                # Special Scaled Initialization --> There are 2 Layer Norms per Transformer Block
+                # Following Pytorch init, except scale by 1/sqrt(2 * n_layer)
+                # We need to reinit p since this code could be called multiple times
+                # Having just p *= scale would repeatedly scale it down
+                if prenorm_residual_strategy == 'rescale':
+                    nn.init.kaiming_uniform_(p, a=math.sqrt(5))
+                    with torch.no_grad():
+                        p /= math.sqrt(num_residuals_per_layer * self.config.num_hidden_layers)
+                elif prenorm_residual_strategy == 'zero':
+                    nn.init.zeros_(p)
+                else:
+                    raise ValueError(f"Invalid prenorm_residual_strategy: {prenorm_residual_strategy}")
+
+
+class LightNetModel(LightNetPreTrainedModel):
+
+    def __init__(self, config: LightNetConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embeddings = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        self.layers = nn.ModuleList([LightNetBlock(config, layer_idx) for layer_idx in range(config.num_hidden_layers)])
+        self.norm = (RMSNorm if config.fuse_norm else nn.RMSNorm)(config.hidden_size, eps=config.norm_eps)
+
+        self.gradient_checkpointing = False
+
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embeddings
+
+    def set_input_embeddings(self, value):
+        self.embeddings = value
+
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,  # noqa
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        **kwargs: Unpack[Dict]
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        if output_attentions:
+            warnings.warn("`LightNetModel` does not `output_attentions` now, setting it to `False`.")
+            output_attentions = False
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        use_cache = use_cache if use_cache is not None else (self.config.use_cache if not self.training else False)
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # retrieve input_ids and inputs_embeds
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        if input_ids is None and inputs_embeds is None:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embeddings(input_ids)
+        hidden_states = inputs_embeds
+
+        if use_cache and not isinstance(past_key_values, Cache):
+            past_key_values = Cache.from_legacy_cache(past_key_values)
+
+        if self.gradient_checkpointing and self.training and use_cache:
+            logger.warning_once("`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...")
+            use_cache = False
+
+        all_hidden_states = () if output_hidden_states else None
+        all_attns = () if output_attentions else None
+
+        for i, layer in enumerate(self.layers):
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+                hidden_states, attentions, past_key_values = self._gradient_checkpointing_func(
+                    layer.__call__,
+                    hidden_states,
+                    attention_mask,
+                    past_key_values,
+                    use_cache,
+                    output_attentions,
+                    **kwargs
+                )
+            else:
+                hidden_states, attentions, past_key_values = layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    past_key_values=past_key_values,
+                    use_cache=use_cache,
+                    output_attentions=output_attentions,
+                    **kwargs
+                )
+
+            if output_attentions:
+                all_attns += (attentions,)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        if not return_dict:
+            return tuple(i for i in [hidden_states, past_key_values, all_hidden_states, all_attns] if i is not None)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+            hidden_states=all_hidden_states,
+            attentions=all_attns
+        )
+
+
+class LightNetForCausalLM(LightNetPreTrainedModel, GenerationMixin):
+
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = LightNetModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+        self.criterion = None
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embeddings
+
+    def set_input_embeddings(self, value):
+        self.model.embeddings = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    def generate(self, *args, **kwargs):
+        try:
+            return super().generate(*args, **kwargs)
+        except AttributeError as exception:
+            if 'past_key_values' in str(exception):
+                raise AttributeError(
+                    f"You tried to call `generate` with a decoding strategy that manipulates `past_key_values`, "
+                    f"which is not supported for {self.__class__.__name__}. "
+                    f"Try another generation strategy instead. "
+                    f"For the available generation strategies, check this doc: "
+                    f"https://huggingface.co/docs/transformers/en/generation_strategies#decoding-strategies"
+                )
+            else:
+                raise exception
+
+    @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
+    def prepare_inputs_for_generation(
+        self,
+        input_ids: torch.LongTensor = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        use_cache: bool = True,
+        logits_to_keep: Optional[int] = None,
+        **kwargs: Unpack[Dict]
+    ):
+        # only last token for `inputs_ids` if the `past_key_values` is not empty.
+        if past_key_values is not None and len(past_key_values) > 0:
+            input_ids = input_ids[:, -1:]
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and len(past_key_values) == 0:
+            model_inputs = {'inputs_embeds': inputs_embeds}
+        else:
+            # The `contiguous()` here is necessary to have a static stride during decoding. torchdynamo otherwise
+            # recompiles graphs as the stride of the inputs is a guard.
+            # Ref: https://github.com/huggingface/transformers/pull/29114
+            # TODO: use `next_tokens` directly instead.
+            model_inputs = {'input_ids': input_ids.contiguous()}
+
+        if logits_to_keep is not None:
+            model_inputs['logits_to_keep'] = logits_to_keep
+
+        model_inputs.update({
+            'past_key_values': past_key_values,
+            'use_cache': use_cache,
+            'attention_mask': attention_mask,
+        })
+        return model_inputs
+
+    @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        logits_to_keep: Optional[int] = 0,
+        **kwargs: Unpack[Dict]
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            inputs_embeds=inputs_embeds,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            **kwargs
+        )
+
+        hidden_states = outputs[0]
+        fuse_linear_and_cross_entropy = self.config.fuse_cross_entropy and self.training
+
+        loss, logits = None, None
+        if not fuse_linear_and_cross_entropy or labels is None:
+            logits = self.lm_head(hidden_states if logits_to_keep is None else hidden_states[:, -logits_to_keep:])
+        if labels is not None:
+            if getattr(self, 'criterion', None) is None:
+                if fuse_linear_and_cross_entropy:
+                    criterion = FusedLinearCrossEntropyLoss()
+                elif self.config.fuse_cross_entropy:
+                    criterion = FusedCrossEntropyLoss(inplace_backward=True)
+                else:
+                    criterion = nn.CrossEntropyLoss()
+            else:
+                criterion = self.criterion
+            labels = labels.to(hidden_states.device)
+            labels = torch.cat((labels[..., 1:], torch.full_like(labels[:, :1], criterion.ignore_index)), 1)
+            if fuse_linear_and_cross_entropy:
+                loss = criterion(hidden_states, labels, self.lm_head.weight, self.lm_head.bias)
+            else:
+                loss = criterion(logits.view(labels.numel(), -1), labels.view(-1))
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )

fla/models/linear_attn/__init__.py

@@ -0,0 +1,12 @@
+# -*- coding: utf-8 -*-
+
+from transformers import AutoConfig, AutoModel, AutoModelForCausalLM
+
+from fla.models.linear_attn.configuration_linear_attn import LinearAttentionConfig
+from fla.models.linear_attn.modeling_linear_attn import LinearAttentionForCausalLM, LinearAttentionModel
+
+AutoConfig.register(LinearAttentionConfig.model_type, LinearAttentionConfig)
+AutoModel.register(LinearAttentionConfig, LinearAttentionModel)
+AutoModelForCausalLM.register(LinearAttentionConfig, LinearAttentionForCausalLM)
+
+__all__ = ['LinearAttentionConfig', 'LinearAttentionForCausalLM', 'LinearAttentionModel']

fla/models/mamba/configuration_mamba.py

@@ -0,0 +1,166 @@
+# coding=utf-8
+# Copyright 2024 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""MAMBA configuration"""
+
+import math
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class MambaConfig(PretrainedConfig):
+    """
+    This is the configuration class to store the configuration of a [`MambaModel`]. It is used to instantiate a MAMBA
+    model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
+    defaults will yield a similar configuration to that of the MAMBA
+    [state-spaces/mamba-2.8b](https://huggingface.co/state-spaces/mamba-2.8b) architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*):
+            Vocabulary size of the Mamba model.
+        hidden_size (`int`, *optional*):
+            Dimensionality of the embeddings and hidden states. Default: 2048.
+        state_size (`int`, *optional*):
+            Shape of the state space latents. Default: 16.
+        num_hidden_layers (`int`, *optional*):
+            Number of hidden layers in the model. Default: 48.
+        layer_norm_epsilon (`float`, *optional*):
+            The epsilon to use in the layer normalization layers. Default: 1e-5.
+        pad_token_id (`int`, *optional*):
+            Padding token id. Default: 0.
+        bos_token_id (`int`, *optional*):
+            The id of the beginning of sentence token in the vocabulary. Default: 0.
+        eos_token_id (`int`, *optional*):
+            The id of the end of sentence token in the vocabulary. Default: 0.
+        expand (`int`, *optional*):
+            Expanding factor used to determine the intermediate size. Default: 2.
+        conv_kernel (`int`, *optional*):
+            Size of the convolution kernel. Default: 4.
+        use_bias (`bool`, *optional*):
+            Whether or not to use bias in ["in_proj", "out_proj"] of the mixer block. Default: `False`.
+        use_conv_bias (`bool`, *optional*):
+            Whether or not to use bias in the convolution layer of the mixer block. Default: `True`.
+        hidden_act (`str`, *optional*):
+            The non-linear activation function (function or string) in the decoder. Default: `"silu"`.
+        initializer_range (`float`, *optional*):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices. Default: 0.1.
+        residual_in_fp32 (`bool`, *optional*):
+            Whether or not residuals should be in `float32`.
+            If set to `False` residuals will keep the same `dtype` as the rest of the model. Default: `True`.
+        time_step_rank (`Union[int,str]`, *optional*):
+            Rank of the the discretization projection matrix.
+            `"auto"` means that it will default to `math.ceil(self.hidden_size / 16)`. Default: `"auto"`.
+        time_step_scale (`float`, *optional*):
+            Scale used used to scale `dt_proj.bias`. Default: 1.0.
+        time_step_min (`float`, *optional*):
+            Minimum `time_step` used to bound `dt_proj.bias`. Default: 0.001.
+        time_step_max (`float`, *optional*):
+            Maximum `time_step` used to bound `dt_proj.bias`. Default: 0.1.
+        time_step_init_scheme (`float`, *optional*):
+            Init scheme used for `dt_proj.weight`. Should be one of `["random","uniform"]`. Default: `"random"`.
+        time_step_floor (`float`, *optional*):
+            Minimum clamping value of the `dt_proj.bias` layer initialization. Default: 0.0001.
+        window_size (`int`, *optional*):
+            The window size used for sliding window attention. Default: 2048.
+        rescale_prenorm_residual (`bool`, *optional*):
+            Whether or not to rescale `out_proj` weights when initializing. Default: `False`.
+        use_cache (`bool`, *optional*):
+            Whether or not the cache should be used. Default: `True`.
+
+
+    Example:
+
+    ```python
+    >>> from transformers import MambaConfig, MambaModel
+
+    >>> # Initializing a Mamba configuration
+    >>> configuration = MambaConfig()
+
+    >>> # Initializing a model (with random weights) from the configuration
+    >>> model = MambaModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "mamba"
+
+    def __init__(
+        self,
+        vocab_size: int = 32000,
+        hidden_size: int = 2048,
+        state_size: int = 16,
+        num_hidden_layers: int = 48,
+        layer_norm_epsilon=1e-5,
+        pad_token_id: int = 0,
+        bos_token_id: int = 1,
+        eos_token_id: int = 2,
+        expand: int = 2,
+        conv_kernel: int = 4,
+        use_bias: bool = False,
+        use_conv_bias: bool = True,
+        hidden_act: str = "silu",
+        initializer_range: str = 0.1,
+        residual_in_fp32: bool = False,
+        time_step_rank: str = "auto",
+        time_step_scale: float = 1.0,
+        time_step_min: float = 0.001,
+        time_step_max: float = 0.1,
+        time_step_init_scheme: str = "random",
+        time_step_floor: float = 1e-4,
+        rescale_prenorm_residual: bool = False,
+        use_cache: bool = True,
+        fuse_norm: bool = True,
+        fuse_cross_entropy: bool = True,
+        tie_word_embeddings: bool = False,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.state_size = state_size
+        self.num_hidden_layers = num_hidden_layers
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.conv_kernel = conv_kernel
+        self.expand = expand
+        self.intermediate_size = int(expand * self.hidden_size)
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        self.pad_token_id = pad_token_id
+        self.use_bias = use_bias
+        self.use_conv_bias = use_conv_bias
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.time_step_rank = math.ceil(self.hidden_size / 16) if time_step_rank == "auto" else time_step_rank
+        self.time_step_scale = time_step_scale
+        self.time_step_min = time_step_min
+        self.time_step_max = time_step_max
+        self.time_step_init_scheme = time_step_init_scheme
+        self.time_step_floor = time_step_floor
+        self.rescale_prenorm_residual = rescale_prenorm_residual
+        self.residual_in_fp32 = residual_in_fp32
+        self.use_cache = use_cache
+        self.fuse_norm = fuse_norm
+        self.fuse_cross_entropy = fuse_cross_entropy
+
+        super().__init__(
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            pad_token_id=pad_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs
+        )

fla/models/mamba2/__init__.py

@@ -0,0 +1,13 @@
+# -*- coding: utf-8 -*-
+
+from transformers import AutoConfig, AutoModel, AutoModelForCausalLM
+
+from fla.models.mamba2.configuration_mamba2 import Mamba2Config
+from fla.models.mamba2.modeling_mamba2 import Mamba2ForCausalLM, Mamba2Model
+
+AutoConfig.register(Mamba2Config.model_type, Mamba2Config, True)
+AutoModel.register(Mamba2Config, Mamba2Model, True)
+AutoModelForCausalLM.register(Mamba2Config, Mamba2ForCausalLM, True)
+
+
+__all__ = ['Mamba2Config', 'Mamba2ForCausalLM', 'Mamba2Model']

flame/config_manager.py

@@ -0,0 +1,940 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+import argparse
+import sys
+from collections import defaultdict
+from typing import Tuple
+
+import torch
+
+try:
+    import tomllib
+except ModuleNotFoundError:
+    import tomli as tomllib
+
+from torchtitan.tools.logging import logger
+
+TORCH_DTYPE_MAP = {
+    "float16": torch.float16,
+    "float32": torch.float32,
+    "bfloat16": torch.bfloat16,
+}
+
+
+def string_list(raw_arg):
+    """Comma-separated string list argument."""
+    return [s.strip() for s in raw_arg.split(",") if s.strip()]
+
+
+def check_string_list_argument(args_dict: dict[str, any], fullargname: str):
+    section, name = fullargname.split(".")
+    # Split string list which are still raw strings.
+    if (
+        section in args_dict
+        and name in args_dict[section]
+        and isinstance(args_dict[section][name], str)
+    ):
+        sec = args_dict[section]
+        sec[name] = string_list(sec[name])
+
+
+class JobConfig:
+    """
+    A helper class to manage the train configuration.
+    Semantics:
+    - Default config is loaded from a toml file. If no toml file is provided,
+    then the default config is loaded from argparse defaults.
+    - if toml file has missing keys, they are filled with argparse defaults.
+    - if additional explicit cmd args are provided in addition to the toml
+    file, they will override the toml config and the argparse defaults
+
+    precedence order: cmdline > toml > argparse default
+
+    Arg parsing semantics:
+
+    Each argument starts with <prefix>_ which is the section name in the toml file
+    followed by name of the option in the toml file. For ex,
+    model.name translates to:
+        [model]
+        name
+    in the toml file
+    """
+
+    def __init__(self):
+        self.args_dict = None
+        # main parser
+        self.parser = argparse.ArgumentParser(description="torchtitan arg parser.")
+
+        self.parser.add_argument(
+            "--job.config_file",
+            type=str,
+            default=None,
+            help="Job config file",
+        )
+
+        # job level configs
+        self.parser.add_argument(
+            "--job.dump_folder",
+            type=str,
+            default="./torchtitan/outputs",
+            help="Folder to dump job outputs",
+        )
+        self.parser.add_argument(
+            "--job.description",
+            type=str,
+            default="default job",
+            help="Description of the job",
+        )
+        self.parser.add_argument(
+            "--job.use_for_integration_test",
+            action="store_true",
+            help="Add this config to the integration test suite",
+        )
+        self.parser.add_argument(
+            "--job.print_args",
+            action="store_true",
+            help="Print the args to terminal",
+        )
+
+        # model configs
+        self.parser.add_argument(
+            "--model.name",
+            type=str,
+            default="fla",
+            help="Which model to train",
+        )
+        self.parser.add_argument(
+            "--model.config",
+            type=str,
+            default="fla-hub/transformer-1.3B-100B",
+            help="Path to the model config",
+        )
+        self.parser.add_argument(
+            "--model.tokenizer_path",
+            type=str,
+            default="fla-hub/transformer-1.3B-100B",
+            help="Tokenizer path",
+        )
+        self.parser.add_argument(
+            "--model.converters",
+            type=string_list,
+            nargs="+",
+            default=[],
+            help="""
+                Comma separated list of converters to apply to the model.
+                For instance, the `float8` converter swaps `torch.nn.Linear`
+                with `Float8Linear`. This feature requires you to install 'torchao'
+                which can be found here: https://github.com/pytorch/ao
+            """,
+        )
+        self.parser.add_argument(
+            "--model.print_after_conversion",
+            action="store_true",
+            help="""
+            If true, model definition will be printed to stdout after all model
+            converters have been applied.
+            """,
+        )
+
+        # profiling configs
+        self.parser.add_argument(
+            "--profiling.enable_profiling",
+            action="store_true",
+            help="Whether to enable pytorch profiler",
+        )
+        self.parser.add_argument(
+            "--profiling.save_traces_folder",
+            type=str,
+            default="profile_traces",
+            help="Trace files location",
+        )
+        self.parser.add_argument(
+            "--profiling.profile_freq",
+            type=int,
+            default=10,
+            help="How often to collect profiler traces, in iterations",
+        )
+        self.parser.add_argument(
+            "--profiling.enable_memory_snapshot",
+            action="store_true",
+            help="Whether to dump memory snapshot",
+        )
+        self.parser.add_argument(
+            "--profiling.save_memory_snapshot_folder",
+            type=str,
+            default="memory_snapshot",
+            help="Memeory snapshot files location",
+        )
+
+        # optimizer configs
+        self.parser.add_argument(
+            "--optimizer.name", type=str, default="AdamW", help="Optimizer to use"
+        )
+        self.parser.add_argument(
+            "--optimizer.eps",
+            type=float,
+            default=1e-8,
+            help="Epsilon value for the optimizer.",
+        )
+        self.parser.add_argument(
+            "--optimizer.lr", type=float, default=8e-4, help="Learning rate to use"
+        )
+        self.parser.add_argument(
+            "--optimizer.implementation",
+            type=str,
+            default="fused",
+            choices=["for-loop", "foreach", "fused"],
+            help="""
+            Specify which optimizer implementation to use:
+            - 'fused': Use fused implementation (CUDA only) for best performance.
+            - 'foreach': Use some horizontal fusion of tensors for better performance.
+            - 'for-loop': Use the default implementation for the optimizer (slowest).
+            - more info: https://pytorch.org/docs/stable/optim.html
+            """,
+        )
+        self.parser.add_argument(
+            "--optimizer.early_step_in_backward",
+            action="store_true",
+            help="""
+            Whether to apply optimizer in the backward. Caution, optimizer_in_backward
+            is not compatible with gradients clipping, users should not call
+            register_post_accumulate_grad_hook after the optimizer is built.""",
+        )
+
+        # lr scheduler configs
+        self.parser.add_argument(
+            "--lr_scheduler.warmup_steps",
+            type=int,
+            default=200,
+            help="Steps for lr scheduler warmup, normally 1/5 of --training.steps",
+        )
+        self.parser.add_argument(
+            "--lr_scheduler.decay_ratio",
+            type=float,
+            default=None,
+            help="""
+            Controls the proportion of the training steps allocated to the learning rate decay phase.
+
+            If `None`, the learning rate will begin decaying immediately after the warmup period.
+            Otherwise, the learning rate will remain stable after the warmup period and
+            only start decaying during the last `decay_ratio` portion of the total training steps.
+
+            This is known as the Warmup-Stable-Decay (WSD) schedule, as described in https://arxiv.org/abs/2404.06395.
+            """,
+        )
+        self.parser.add_argument(
+            "--lr_scheduler.decay_type",
+            type=str,
+            default="linear",
+            choices=["linear", "sqrt", "cosine"],
+            help="""
+            Learning rate decay type to use during training:
+            - 'linear': linearly decays learning rate from initial to final value
+            - 'sqrt': decays learning rate following a 1 minus square root curve
+            - 'cosine': smoothly decays learning rate following a cosine curve
+            """,
+        )
+        self.parser.add_argument(
+            "--lr_scheduler.lr_min",
+            type=float,
+            default=0.0,
+            help="""
+            Min lr ratio for lr scheduler.
+
+            If provided, the range of decay factor is scaled from 1 to `lr_min`
+            to ensure the learning rate does not drop below `optimizer.lr * lr_scheduler.lr_min`.
+            """,
+        )
+
+        # training configs
+        self.parser.add_argument(
+            "--training.batch_size", type=int, default=8, help="Batch size"
+        )
+        self.parser.add_argument(
+            "--training.seq_len", type=int, default=2048, help="Sequence length"
+        )
+        self.parser.add_argument(
+            "--training.context_len",
+            type=int,
+            default=2048,
+            help="Max length allowed for each sequence",
+        )
+        self.parser.add_argument(
+            "--training.varlen",
+            action="store_true",
+            help="Whether to take sequences of variable length as input",
+        )
+        self.parser.add_argument(
+            "--training.gradient_accumulation_steps",
+            type=int,
+            default=1,
+            help="Number of steps to accumulate gradients before updating parameters",
+        )
+        self.parser.add_argument(
+            "--training.steps",
+            type=int,
+            default=10000,
+            help="How many train steps to run",
+        )
+        self.parser.add_argument(
+            "--training.max_norm",
+            type=float,
+            default=1.0,
+            help="Max norm for gradient clipping",
+        )
+        self.parser.add_argument(
+            "--training.skip_nan_inf",
+            action="store_true",
+            help="Skip batch updates when NaN or INF gradients are encountered during training",
+        )
+        self.parser.add_argument(
+            "--training.dataset",
+            default="HuggingFaceFW/fineweb-edu",
+            help="Dataset to use, with comma separated values",
+        )
+        self.parser.add_argument(
+            "--training.dataset_name",
+            default=None,
+            help="The name of the dataset config, with comma separated values if provided",
+        )
+        self.parser.add_argument(
+            "--training.dataset_split",
+            default=None,
+            help="Dataset split to use, with comma separated values if provided",
+        )
+        self.parser.add_argument(
+            "--training.data_dir",
+            default=None,
+            help="Data dirs to use, with comma separated values if provided",
+        )
+        self.parser.add_argument(
+            "--training.data_files",
+            default=None,
+            help="Data files to use, with comma separated values if provided",
+        )
+        self.parser.add_argument(
+            "--training.data_probs",
+            default=None,
+            help="Data sampling probabilities, with comma separated values if provided",
+        )
+        self.parser.add_argument(
+            "--training.streaming",
+            action="store_true",
+            help="Whether to load dataset in streaming mode, used for huge dataset",
+        )
+        self.parser.add_argument(
+            "--training.num_workers",
+            type=int,
+            default=32,
+            help="Number of subprocesses to use for data loading. 0 means that the data will be loaded in the main process.",
+        )
+        self.parser.add_argument(
+            "--training.prefetch_factor",
+            type=int,
+            default=2,
+            help="Number of batches loaded in advance by each worker."
+            "2 means there will be a total of 2 * num_workers batches prefetched across all workers.",
+        )
+        self.parser.add_argument(
+            "--training.data_parallel_replicate_degree",
+            type=int,
+            default=1,
+            help="""
+            The `data_parallel_replicate_degree` argument specifies the degree of
+            data parallelism for weight replication. When this value is greater
+            than 1, weights will be replicated across `data_parallel_replicate_degree`
+            ranks. If `data_parallel_shard_degree` is also greater than 1, the parallelism
+            method used is HSDP (Hybrid Sharded Data Parallelism). Otherwise, the
+            parallelism method used is DDP (Distributed Data Parallelism).
+            1 means disabled.""",
+        )
+        self.parser.add_argument(
+            "--training.data_parallel_shard_degree",
+            type=int,
+            default=-1,
+            help="""
+            The `data_parallel_shard_degree` argument specifies the degree of data
+            parallelism for weight sharding. When this value is greater than 1, weights
+            will be sharded across `data_parallel_shard_degree` ranks. If
+            `data_parallel_replicate_degree` is also greater than 1, the parallelism
+            method used is HSDP (Hybrid Sharded Data Parallelism).  Otherwise, the
+            parallelism method used is FSDP (Fully Sharded Data Parallelism).
+
+            -1 means leftover ranks will be used (After DP_REPLICATE/SP/PP). Note that
+            only `data_parallel_shard_degree` can be negative. 1 means disabled.""",
+        )
+        self.parser.add_argument(
+            "--training.enable_cpu_offload",
+            action="store_true",
+            help="""
+            Whether to apply CPU offloading of parameters, gradients, and optimizer states in FSDP""",
+        )
+        self.parser.add_argument(
+            "--training.tensor_parallel_degree",
+            type=int,
+            default=1,
+            help="Tensor Parallelism degree. 1 means disabled.",
+        )
+        self.parser.add_argument(
+            "--training.disable_loss_parallel",
+            action="store_true",
+            help="Whether to apply loss parallel when sequence parallel is enabled",
+        )
+        self.parser.add_argument(
+            "--training.fsdp_reshard_after_forward",
+            type=str,
+            default="default",
+            choices=["default", "always", "never"],
+            help="""
+            `reshard_after_forward` specifies the policy for applying `reshard_after_forward`
+            within an FSDP setup. `reshard_after_forward` controls parameter behavior after forward,
+            trading off memory and communication. See torch's `fully_shard` API for more documentation
+            on `reshard_after_forward`.
+            The supported policies include "default", "always" and "never":
+            - "default" applies default resharding behavior, implementing "smart defaults" for known optimal
+              scenarios.
+            - "always" will enable `reshard_after_forward` for all forward passes.
+            - "never" will disable `reshard_after_forward` for all forward passes.
+            """,
+        )
+        self.parser.add_argument(
+            "--training.mixed_precision_param",
+            type=str,
+            default="bfloat16",
+            choices=["bfloat16", "float32"],
+            help="""
+                torch dtype to use for parameters when applying mixed precision via FSDP.
+                This feature only takes effect when data_parallel_shard_degree > 1
+            """,
+        )
+        self.parser.add_argument(
+            "--training.mixed_precision_reduce",
+            type=str,
+            default="float32",
+            choices=["float32"],
+            help="""
+                torch dtype to use for reductions when applying mixed precision via FSDP.
+                This feature only takes effect when data_parallel_shard_degree > 1
+            """,
+        )
+        self.parser.add_argument(
+            "--training.compile",
+            action="store_true",
+            help="Whether to compile the model",
+        )
+        self.parser.add_argument(
+            "--training.gc_freq",
+            type=int,
+            default=50,
+            help="Python garbage control scheduling interval, in steps",
+        )
+        self.parser.add_argument(
+            "--training.seed",
+            type=int,
+            default=42,
+            help="Choose the base RNG seed used for training",
+        )
+        self.parser.add_argument(
+            "--training.deterministic",
+            action="store_true",
+            help="Use deterministic algorithms wherever possible, may be slower",
+        )
+        # metrics configs
+        self.parser.add_argument(
+            "--metrics.log_freq",
+            type=int,
+            default=10,
+            help="How often to log metrics to TensorBoard, in iterations",
+        )
+        self.parser.add_argument(
+            "--metrics.enable_tensorboard",
+            action="store_true",
+            help="Whether to log metrics to TensorBoard",
+        )
+        self.parser.add_argument(
+            "--metrics.disable_color_printing",
+            action="store_true",
+            help="Whether to disable color printing in logs",
+        )
+        self.parser.add_argument(
+            "--metrics.save_tb_folder",
+            type=str,
+            default="tb",
+            help="Folder to dump TensorBoard states",
+        )
+        self.parser.add_argument(
+            "--metrics.save_for_all_ranks",
+            action="store_true",
+            default=False,
+            help="""
+                Whether to save TensorBoard/Wandb metrics only for rank 0 or for all ranks.
+                When this option is False and pipeline_parallel_degree is > 1, the metrics
+                component uses the 0th rank of the last stage pipeline group, which is the
+                only stage that computes loss metrics.
+            """,
+        )
+        self.parser.add_argument(
+            "--metrics.enable_wandb",
+            action="store_true",
+            help="Whether to log metrics to Weights & Biases",
+        )
+
+        self.parser.add_argument(
+            "--experimental.enable_async_tensor_parallel",
+            action="store_true",
+            help="Whether to apply async tensor parallel (currently only effective when compile is enabled)",
+        )
+        self.parser.add_argument(
+            "--experimental.pipeline_parallel_degree",
+            type=int,
+            default=1,
+            help="""
+                Pipeline Parallelism degree, or number of ranks. 1 means disabled.
+                If using looped schedules, this still specifies the number of physical ranks, not the number
+                of stages.  Stages per rank are inferred from split points degree, and schedule.""",
+        )
+        self.parser.add_argument(
+            "--experimental.pipeline_parallel_split_points",
+            type=string_list,
+            nargs="+",
+            default=[],
+            help="""
+                Specify comma-separated names of modules to use as the beginning of a split point.
+
+                e.g. "layers.0,layers.2" will cause the model to be split into 3 stages,
+                the first containing all the layers up to layers.0,
+                the second containing layers.0 and up to layers.2,
+                the third containing layers.2 and all the remaining layers.
+
+                Note: fully-automated splitting may be enabled in the future,
+                but currently the split points must be specified manually.""",
+        )
+        self.parser.add_argument(
+            "--experimental.pipeline_parallel_schedule",
+            type=str,
+            default="1F1B",
+            help="""
+                Specify the Pipeline Parallel schedule to use. The supported schedules are:
+                https://github.com/pytorch/pytorch/blob/de4c2a3b4e89d96334dc678d1c3f2ae51a6630a0/torch/distributed/pipelining/schedules.py#L2161.
+                The schedule must be compatible with the split points and stages_per_rank.
+
+                Looped schedules (e.g. Interleaved1F1B) require specifying pipeline_parallel_degree = number of ranks,
+                and split_points = number of stages - 1
+                """,
+        )
+        self.parser.add_argument(
+            "--experimental.pipeline_parallel_schedule_csv",
+            type=str,
+            default="",
+            help="""
+                Specify the path to the pipeline parallel schedule csv file to use.
+                The pipeline_parallel_schedule argument must be either
+                PipelineScheduleSingle, PipelineScheduleMulti, or _PipelineScheduleRuntime.
+            """,
+        )
+
+        self.parser.add_argument(
+            "--experimental.pipeline_parallel_microbatches",
+            type=int,
+            default=None,
+            help="""
+                How many microbatches to split the global training batch into when using pipeline parallelism.
+
+                The global training batch size must be evenly divisible by the number of microbatches.
+
+                The default value will be the number of pipeline stages, if unspecified.
+            """,
+        )
+        self.parser.add_argument(
+            "--experimental.enable_compiled_autograd",
+            action="store_true",
+            help="Enable CompiledAutograd to compile the backward.",
+        )
+        self.parser.add_argument(
+            "--experimental.context_parallel_degree",
+            type=int,
+            default=1,
+            help="Context parallelism degree. 1 means disabled.",
+        )
+        self.parser.add_argument(
+            "--experimental.context_parallel_rotate_method",
+            type=str,
+            default="allgather",
+            help="""
+                The collective to use in context parallel SDPA for kv shards exchange.
+
+                'allgather' means to all-gather all kv shards on ranks after the first sub-SDPA computation,
+
+                'alltoall' means to all-to-all shuffle the kv shards.
+
+                The default value is 'allgather'.
+            """,
+        )
+        # I'm not particularly fond of this. Users can choose to write their own wrapper
+        # module and import TorchTitan training loop and execute it, which look cleaner.
+        # One reason to provide this option is to allow users to use the existing run script.
+        # While the script is pretty trivial now, we may add more logic when integrating
+        # with TorchFT.
+        # This option is subject to change and may be deleted in the future.
+        self.parser.add_argument(
+            "--experimental.custom_model_path",
+            type=str,
+            default="",
+            help="""
+                The --custom_model_path option allows to specify a custom path to a model module
+                that is not natively implemented within TorchTitan.
+                Acceptable values are the file system path to the module (e.g., my_models/model_x)
+                dotted import module  (e.g., some_package.model_x).
+            """,
+        )
+        # checkpointing configs
+        self.parser.add_argument(
+            "--checkpoint.enable_checkpoint",
+            action="store_true",
+            help="Whether to enable checkpoint",
+        )
+        self.parser.add_argument(
+            "--checkpoint.folder",
+            type=str,
+            default="checkpoint",
+            help="""
+                The folder to store the checkpoints.
+                When enable_checkpoint is set to true, checkpoints will be in {--job.dump_folder}/{--checkpoint.folder}.
+            """,
+        )
+        self.parser.add_argument(
+            "--checkpoint.interval",
+            type=int,
+            default=500,
+            help="Checkpointing interval in steps.",
+        )
+        self.parser.add_argument(
+            "--checkpoint.model_weights_only",
+            action="store_true",
+            help="""
+                When model_weights_only=True, only model weights will be saved at the end of training.
+                With this, checkpoints can be loaded using `torch.load(..., weights_only=True)` after conversion.
+                When model_weights_only=False, the full checkpoint will be saved.
+                A full checkpoint includes model, optimizer and train_state, which can be used to resume training.
+                The default value is false.
+            """,
+        )
+        self.parser.add_argument(
+            "--checkpoint.export_dtype",
+            type=str,
+            default="float32",
+            choices=["float16", "bfloat16", "float32"],
+            help="""
+                Converts to the specified precision when training completes and model_weights_only=true.
+                Currently supports float32, float16, and bfloat16.
+                The default value is float32.
+            """,
+        )
+        self.parser.add_argument(
+            "--checkpoint.create_seed_checkpoint",
+            action="store_true",
+            help="""
+                Initializes the full model without applying parallelisms, and then saves it as a seed checkpoint.
+                Note: requires user to call train.py without specifying any parallelisms, e.g. NGPU=1.
+                Could be implemented as a separate script, but this way shares more code.
+            """,
+        )
+        self.parser.add_argument(
+            "--checkpoint.async_mode",
+            type=str,
+            default="disabled",
+            help="""
+                Which async checkpoint mode to use. Currently there are 3 different modes.
+                1. "disabled": synchronized checkpointing will be used.
+                2. "async": torch.distributed.checkpoint.async_save will be used.
+                3. "async_with_pinned_mem": this option utilizes a dedicated pinned memory
+                   space and creates a separate process for faster GPU->CPU transfer
+                   performance and eliminating GIL contention. The cost is increased CPU
+                   memory usage. If insufficient CPU memory is available, performance may
+                   degrade due to memory paging. For most users, "async" should suffice as
+                   the performance overhead is typically small (on the order of tens of
+                   seconds) compared to checkpointing frequency. This mode can be employed
+                   to pursue near-zero checkpointing times (e.g., < 1 second) given
+                   appropriate hardware support such as ample CPU memory and fast PCIe.
+
+                "disabled" is the default mode.
+            """,
+        )
+        self.parser.add_argument(
+            "--checkpoint.keep_latest_k",
+            type=int,
+            default=0,
+            help="""
+                Keeps only the latest k checkpoints, and purging older ones. If 0, keep all checkpoints.
+                0 is the default value. k cannot be 1 as the last one may be in the process of being
+                saved. As a result, the metadata of the last one may not be ready yet.
+            """,
+        )
+        self.parser.add_argument(
+            "--checkpoint.load_step",
+            type=int,
+            default=-1,
+            help="Load the checkpoint at the specified step. If -1, load the latest checkpoint.",
+        )
+        self.parser.add_argument(
+            "--checkpoint.exclude_from_loading",
+            type=string_list,
+            nargs="*",
+            default=[],
+            help="""
+                Exclude specific keys from being loaded from the checkpoint.
+                Provide a comma-separated list of keys to exclude, e.g. 'optimizer,lr_scheduler,dataloader'.
+                This will load the model only, excluding the specified keys.
+            """,
+        )
+        self.parser.add_argument(
+            "--checkpoint.convert_to_hf_on_save",
+            action="store_true",
+            help="""
+                If true, automatically convert the saved DCP checkpoint to Hugging Face format
+                in a parallel directory (e.g., step-1000-hf) after each save.
+            """,
+        )
+        self.parser.add_argument(
+            "--checkpoint.hf_upload_enabled",
+            action="store_true",
+            help="Enable uploading converted Hugging Face checkpoints to the Hub.",
+        )
+        self.parser.add_argument(
+            "--checkpoint.hf_repo_base_name",
+            type=str,
+            default=None,
+            help="Hugging Face Hub repository ID to upload checkpoints to (e.g., 'username/repo').",
+        )
+        self.parser.add_argument(
+            "--checkpoint.hf_upload_format",
+            type=str,
+            default="dcp",
+            choices=["dcp", "hf"],
+            help="""
+                Format to upload to Hugging Face Hub. 'dcp' for DCP format, 'hf' for Hugging Face format.
+                Note: 'hf' is only supported for models with a single pipeline stage.
+            """,
+        )
+        # activation checkpointing configs
+        self.parser.add_argument(
+            "--activation_checkpoint.mode",
+            type=str,
+            default="selective",
+            help="Type of activation checkpointing to use ['none', 'full', 'selective']",
+        )
+        self.parser.add_argument(
+            "--activation_checkpoint.selective_ac_option",
+            type=str,
+            default="2",  # 2 = checkpoint every other layer
+            help="""
+                Selective activation checkpointing options ['int', 'op'].
+                'int' (e.g., 2) for every nth layer, or 'op' for op level ac.
+            """,
+        )
+
+        self.parser.add_argument(
+            "--activation_offload.mode",
+            type=str,
+            default="none",
+            help="""
+                if we are using activation offload or not. Options are ['none', 'full'].
+            """,
+        )
+
+        # float8 configs
+        self.parser.add_argument(
+            "--float8.enable_fsdp_float8_all_gather",
+            action="store_true",
+            help="Whether enable float8 all-gather in FSDP, recommended for tensorwise scaling",
+        )
+        self.parser.add_argument(
+            "--float8.precompute_float8_dynamic_scale_for_fsdp",
+            action="store_true",
+            help="Whether precompute float8 scales dynamically for FSDP, recommended for tensorwise scaling",
+        )
+        self.parser.add_argument(
+            "--float8.force_recompute_fp8_weight_in_bwd",
+            action="store_true",
+            help="""
+            Whether to force the recomputation of FP8 weights during backward pass.
+            When using FSDP with tensorwise scaling, it is recommended to enable
+            `force_recompute_fp8_weight_in_bwd` to prevent saving unsharded FP8 weights
+            for backward computation.
+            """,
+        )
+        self.parser.add_argument(
+            "--float8.recipe_name",
+            type=str,
+            default=None,
+            choices=["tensorwise", "rowwise", "rowwise_with_gw_hp"],
+            help="""
+            If specified, creates float8 config from recipe name, valid choices are
+            `tensorwise`, `rowwise` and `rowwise_with_gw_hp`.
+            """,
+        )
+
+        # communications library settings
+        self.parser.add_argument(
+            "--comm.init_timeout_seconds",
+            type=int,
+            default=300,
+            help="Timeout for communication operations, during initialization and first train step.",
+        )
+        self.parser.add_argument(
+            "--comm.train_timeout_seconds",
+            type=int,
+            default=100,
+            help=(
+                "Timeout for communication operations after the first train step -- "
+                "usually a tighter bound than during initialization."
+            ),
+        )
+        self.parser.add_argument(
+            "--comm.trace_buf_size",
+            type=int,
+            default=20000,
+            help="Flight recorder ring buffer size, >0 means recording by default, 0 means disabled",
+        )
+
+        # memory estimation settings
+        self.parser.add_argument(
+            "--memory_estimation.enabled",
+            help="Whether to estimate memory usage for FSDP",
+            action="store_true",
+        )
+
+        self.parser.add_argument(
+            "--memory_estimation.disable_fake_mode",
+            help="Whether to estimate memory under FakeTensorMode",
+            action="store_true",
+        )
+
+        self.parser.add_argument(
+            "--fault_tolerance.enable",
+            action="store_true",
+            help="""
+                Enable TorchFT integration. When TorchFT is enabled, HSDP will be used.
+                And --fault_tolerance.data_parallel_replicate_degree should be 1 and
+                --fault_tolerance.group_size will be used to control the maximum
+                replicate group size as the replicate group size is dynamic.
+
+                Note that this is still an experimental feature.
+            """,
+        )
+
+        self.parser.add_argument(
+            "--fault_tolerance.replica_id",
+            type=int,
+            default=0,
+            help="The TorchFT replica ID of this run.",
+        )
+
+        self.parser.add_argument(
+            "--fault_tolerance.group_size",
+            type=int,
+            default=0,
+            help="""
+                The number of TorchFT replicate groups. This number will be used for
+                dataloader to split the dataset across the replicate groups and FSDP
+                dimension
+            """,
+        )
+
+        self.parser.add_argument(
+            "--fault_tolerance.min_replica_size",
+            type=int,
+            default=1,
+            help="The minimum number of FT replica for each step.",
+        )
+
+    def to_dict(self):
+        return self.args_dict
+
+    def parse_args(self, args_list: list = sys.argv[1:]):
+        args, cmd_args = self.parse_args_from_command_line(args_list)
+        config_file = getattr(args, "job.config_file", None)
+        # build up a two level dict
+        args_dict = self._args_to_two_level_dict(args)
+        if config_file is not None:
+            try:
+                with open(config_file, "rb") as f:
+                    for k, v in tomllib.load(f).items():
+                        # to prevent overwrite of non-specified keys
+                        args_dict[k] |= v
+            except (FileNotFoundError, tomllib.TOMLDecodeError) as e:
+                logger.exception(
+                    f"Error while loading the configuration file: {config_file}"
+                )
+                logger.exception(f"Error details: {str(e)}")
+                raise e
+
+        # Checking string-list arguments are properly split into a list
+        # if split-points came from 'args' (from cmd line) it would have already been parsed into a list by that parser
+        string_list_argnames = self._get_string_list_argument_names()
+        for n in string_list_argnames:
+            check_string_list_argument(args_dict, n)
+
+        # override args dict with cmd_args
+        cmd_args_dict = self._args_to_two_level_dict(cmd_args)
+        for section, section_args in cmd_args_dict.items():
+            for k, v in section_args.items():
+                args_dict[section][k] = v
+
+        self.args_dict = args_dict
+
+        for k, v in args_dict.items():
+            class_type = type(k.title(), (), v)
+            setattr(self, k, class_type())
+        self._validate_config()
+
+    def _args_to_two_level_dict(self, args: argparse.Namespace) -> defaultdict:
+        args_dict = defaultdict(defaultdict)
+        for k, v in vars(args).items():
+            first_level_key, second_level_key = k.split(".", 1)
+            args_dict[first_level_key][second_level_key] = v
+        return args_dict
+
+    def _validate_config(self) -> None:
+        # TODO: Add more mandatory validations
+        assert self.model.config
+        assert self.model.tokenizer_path
+
+    def _get_string_list_argument_names(self) -> list[str]:
+        """Get the parser argument names of type `string_list`."""
+        string_list_args = [
+            v.dest for v in self.parser._actions if v.type is string_list
+        ]
+        return string_list_args
+
+    def parse_args_from_command_line(
+        self, args_list
+    ) -> Tuple[argparse.Namespace, argparse.Namespace]:
+        """
+        Parse command line arguments and return the parsed args and the command line only args
+        """
+        args = self.parser.parse_args(args_list)
+        string_list_argnames = set(self._get_string_list_argument_names())
+
+        # aux parser to parse the command line only args, with no defaults from main parser
+        aux_parser = argparse.ArgumentParser(argument_default=argparse.SUPPRESS)
+        for arg, val in vars(args).items():
+            if isinstance(val, bool):
+                aux_parser.add_argument(
+                    "--" + arg, action="store_true" if val else "store_false"
+                )
+            elif arg in string_list_argnames:
+                # without this special case, type inference breaks here,
+                # since the inferred type is just 'list' and it ends up flattening
+                # e.g. from ["layers.0", "layers.1"] into ["l", "a", "y", "e", "r", "s", ".0", ...]
+                aux_parser.add_argument("--" + arg, type=string_list)
+            else:
+                aux_parser.add_argument("--" + arg, type=type(val))
+
+        cmd_args, _ = aux_parser.parse_known_args(args_list)
+
+        return args, cmd_args

flame/data.py

@@ -0,0 +1,570 @@
+# -*- coding: utf-8 -*-
+
+from __future__ import annotations
+
+import copy
+import pickle
+from copy import deepcopy
+from dataclasses import dataclass
+from typing import Any, Callable, Dict, Iterable, List, Optional, Union
+
+import datasets
+import numpy as np
+import torch
+from datasets import Dataset, IterableDataset
+from datasets.iterable_dataset import ShufflingConfig
+from torch.distributed.checkpoint.stateful import Stateful
+from torchdata.stateful_dataloader import StatefulDataLoader
+from transformers import PreTrainedTokenizer
+
+from torchtitan.tools.logging import logger
+
+
+class BufferShuffledIterableDataset(IterableDataset):
+    def __init__(
+        self,
+        dataset: Dataset,
+        tokenizer: PreTrainedTokenizer,
+        seq_len: int = 2048,
+        rank: int = 0,
+        world_size: int = 1,
+        buffer_size: int = 1024,
+    ) -> BufferShuffledIterableDataset:
+        self.dataset = dataset
+        self.tokenizer = tokenizer
+
+        self.data = dataset.shard(world_size, rank)
+        self.seq_len = seq_len
+
+        self.rank = rank
+        self.world_size = world_size
+        self.buffer_size = buffer_size
+
+        if tokenizer.vocab_size < torch.iinfo(torch.int16).max:
+            self.dtype = torch.int16
+        elif tokenizer.vocab_size < torch.iinfo(torch.int32).max:
+            self.dtype = torch.int32
+        else:
+            self.dtype = torch.int64
+        self.states = None
+        self.buffer = torch.tensor([], dtype=self.dtype)
+        self.tokens = []
+        self.rand_id = 0
+        self.token_id = 0
+        self.rng_state = None
+        self._epoch = 0
+
+    def __iter__(self):
+        g = torch.Generator()
+        g.manual_seed(self._epoch + self.rank)
+        if self.rng_state is not None:
+            g.set_state(self.rng_state)
+
+        rand_it = self.randint(0, self.buffer_size, g=g)
+        if self.states is not None:
+            self.data.load_state_dict(self.states)
+
+        # max number of tokens allowed in the chunk buffer
+        n_tokens = self.buffer_size * self.seq_len
+
+        while True:
+            for sample in self.tokenize(self.data):
+                # keep appending the samples to the token buffer
+                self.tokens += sample
+                # if the token buffer is full, start sampling
+                # NOTE: we first convert the token ids to a tensor of shape [n_chunks, seq_len] for efficiency
+                if len(self.buffer) == 0 and len(self.tokens) >= n_tokens:
+                    self.buffer = torch.tensor(self.tokens[:n_tokens], dtype=self.dtype).view(self.buffer_size, -1)
+                    self.tokens = self.tokens[n_tokens:]
+                if len(self.buffer) == self.buffer_size:
+                    yield from self.sample(rand_it)
+
+            n_chunks = len(self.tokens) // self.seq_len
+            # handle the left tokens in the buffer
+            if n_chunks > 0:
+                n_tokens = n_chunks * self.seq_len
+                indices = torch.randperm(n_chunks, generator=g).tolist()
+                self.buffer = torch.tensor(self.tokens[:n_tokens], dtype=torch.long).view(n_chunks, -1)
+                self.tokens = self.tokens[n_tokens:]
+                for i in indices:
+                    yield {'input_ids': self.buffer[i]}
+
+    def tokenize(self, data, batch_size: int = 64):
+        texts, states = [], []
+        for sample in data:
+            texts.append(sample['text'])
+            states.append(self.data.state_dict())
+            if len(texts) == batch_size:
+                for s, tokenized in zip(states, self.tokenizer(texts, return_attention_mask=False)['input_ids']):
+                    self.states = s
+                    yield tokenized
+                texts, states = [], []
+        if len(texts) > 0:
+            for s, tokenized in zip(states, self.tokenizer(texts, return_attention_mask=False)['input_ids']):
+                self.states = s
+                yield tokenized
+
+    def sample(self, indices):
+        n_tokens = (len(self.tokens) // self.seq_len) * self.seq_len
+        while self.token_id < n_tokens:
+            i = next(indices)
+            start, end = self.token_id, self.token_id + self.seq_len
+            self.token_id += self.seq_len
+            yield {'input_ids': self.buffer[i].to(torch.long)}
+            self.buffer[i] = torch.tensor(self.tokens[start:end], dtype=self.dtype)
+        self.token_id = 0
+        self.tokens = self.tokens[n_tokens:]
+
+    def randint(self, low: int, high: int, buffer_size: int = 1024, g: torch.Generator = torch.Generator()) -> Iterable[int]:
+        indices = torch.empty(buffer_size, dtype=torch.long)
+        while True:
+            # record the generator states before sampling
+            self.rng_state = g.get_state()
+            indices = torch.randint(low, high, (buffer_size,), out=indices, generator=g)
+            for i in indices[self.rand_id:].tolist():
+                self.rand_id += 1
+                yield i
+            self.rand_id = 0
+
+    def set_epoch(self, epoch):
+        self._epoch = epoch
+        if hasattr(self.dataset, 'set_epoch'):
+            self.dataset.set_epoch(epoch)
+
+    def state_dict(self):
+        return {
+            'states': self.states,
+            'buffer': self.buffer.clone(),
+            'tokens': deepcopy(self.tokens),
+            'rand_id': self.rand_id,
+            'token_id': self.token_id,
+            'rng_state': self.rng_state,
+            'epoch': self._epoch,
+        }
+
+    def load_state_dict(self, state_dict):
+        self.states = state_dict['states']
+        self.buffer = state_dict['buffer'].clone()
+        self.tokens = deepcopy(state_dict['tokens'])
+        self.rand_id = state_dict['rand_id']
+        self.token_id = state_dict['token_id']
+        self.rng_state = state_dict['rng_state'].clone() if state_dict['rng_state'] is not None else None
+        self._epoch = state_dict['epoch']
+
+
+class OnlineTokenizedIterableDataset(IterableDataset):
+    def __init__(
+        self, dataset: Dataset, tokenizer: PreTrainedTokenizer, seq_len: int = 2048, rank: int = 0, world_size: int = 1
+    ) -> OnlineTokenizedIterableDataset:
+        self.dataset = dataset
+        self.tokenizer = tokenizer
+
+        self.data = dataset.shard(world_size, rank)
+        self.seq_len = seq_len
+        self.rank = rank
+        self.world_size = world_size
+
+        self.states = None
+        self.tokens = []
+
+    def __iter__(self):
+        if self.states is not None:
+            self.data.load_state_dict(self.states)
+
+        while True:
+            for sample in self.tokenize(self.data):
+                # keep appending the samples to the token buffer
+                self.tokens += sample
+
+                while len(self.tokens) >= self.seq_len:
+                    input_ids = torch.tensor(self.tokens[:self.seq_len], dtype=torch.long)
+                    self.tokens = self.tokens[self.seq_len:]
+                    yield {'input_ids': input_ids}
+
+    def tokenize(self, data, buffer_size: int = 64):
+        buffer, states = [], []
+        for sample in data:
+            if sample.get('text', None) is not None:
+                buffer.append(sample['text'])
+            elif sample.get('content', None) is not None:
+                buffer.append(sample['content'])
+            else:
+                raise ValueError(f"No 'text' or 'content' field found in sample:\n{sample}")
+            states.append(self.data.state_dict())
+            if len(buffer) == buffer_size:
+                for s, tokenized in zip(states, self.tokenizer(buffer, return_attention_mask=False)['input_ids']):
+                    self.states = s
+                    yield tokenized
+                buffer, states = [], []
+        if len(buffer) > 0:
+            for s, tokenized in zip(states, self.tokenizer(buffer, return_attention_mask=False)['input_ids']):
+                self.states = s
+                yield tokenized
+
+    def state_dict(self):
+        return {'states': self.states, 'tokens': deepcopy(self.tokens)}
+
+    def load_state_dict(self, state_dict):
+        self.states = state_dict['states']
+        self.tokens = deepcopy(state_dict['tokens'])
+
+
+class BufferShuffledExamplesIterable(datasets.iterable_dataset.BufferShuffledExamplesIterable):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+    def _init_state_dict(self) -> dict:
+        self._state_dict = self.ex_iterable._init_state_dict()
+        self._state_dict['mem_buffer'] = ([],)
+        self._state_dict['bit_generator_state'] = self.generator.bit_generator.state
+        self._state_dict['bit_generator_index_offset'] = 0
+        self._state_dict['bit_generator_index_offset_shuffle'] = 0
+        return self._state_dict
+
+    def __iter__(self):
+        buffer_size = self.buffer_size
+        rng = deepcopy(self.generator)
+        # this is the shuffle buffer that we keep in memory
+        mem_buffer = self._state_dict['mem_buffer'][0]
+        # this is an infinite iterator that randomly samples the index of the source to pick examples from
+        index_offset = self._state_dict['bit_generator_index_offset'] if self._state_dict else 0
+        if self._state_dict:
+            rng.bit_generator.state = self._state_dict['bit_generator_state']
+        indices_iterator = self._iter_random_indices(rng, buffer_size, random_batch_size=buffer_size)
+        # skip already consumed ones
+        for _ in range(index_offset):
+            i = next(indices_iterator)
+
+        for x in self.ex_iterable:
+            if len(mem_buffer) < buffer_size:  # if the buffer is not full, keep filling the buffer
+                mem_buffer.append(x)
+            else:  # otherwise, pick an example from it
+                i = next(indices_iterator)
+                index_offset = (index_offset + 1) % buffer_size
+                if self._state_dict:
+                    self._state_dict['bit_generator_index_offset'] = index_offset
+                    if index_offset == 0:
+                        self._state_dict['bit_generator_state'] = rng.bit_generator.state
+                selected = mem_buffer[i]
+                mem_buffer[i] = x  # replace the picked example by a new one
+                yield selected
+
+        index_offset = self._state_dict['bit_generator_index_offset_shuffle'] if self._state_dict else 0
+        if self._state_dict:
+            rng.bit_generator.state = self._state_dict['bit_generator_state']
+
+        # when we run out of examples, we shuffle the remaining examples in the buffer and yield them
+        for i in rng.permutation(len(mem_buffer))[index_offset:].tolist():
+            index_offset = index_offset + 1
+            if self._state_dict:
+                self._state_dict['bit_generator_index_offset_shuffle'] = index_offset
+            yield mem_buffer[i]
+
+    def shuffle_data_sources(self, generator: np.random.Generator) -> BufferShuffledExamplesIterable:
+        """Shuffle the wrapped examples iterable as well as the shuffling buffer."""
+        return BufferShuffledExamplesIterable(
+            self.ex_iterable.shuffle_data_sources(generator), buffer_size=self.buffer_size, generator=generator
+        )
+
+    def shard_data_sources(self, num_shards: int, index: int, contiguous=True) -> BufferShuffledExamplesIterable:
+        """Keep only the requested shard."""
+        return BufferShuffledExamplesIterable(
+            self.ex_iterable.shard_data_sources(num_shards, index, contiguous=contiguous),
+            buffer_size=self.buffer_size,
+            generator=self.generator,
+        )
+
+    def load_state_dict(self, state_dict: dict) -> dict:
+        def _inner_load_state_dict(state, new_state):
+            if new_state is not None and isinstance(state, dict):
+                for key in new_state:
+                    state[key] = _inner_load_state_dict(state[key], new_state[key])
+                return state
+            elif new_state is not None and isinstance(state, list):
+                for i in range(len(state)):
+                    state[i] = _inner_load_state_dict(state[i], new_state[i])
+                return state
+            return new_state
+
+        return _inner_load_state_dict(self._state_dict, state_dict)
+
+
+def shuffle(
+    dataset: IterableDataset,
+    seed: int = 42,
+    generator: np.random.Generator = None,
+    buffer_size: int = 1024,
+):
+    generator = np.random.default_rng(seed) if generator is None else deepcopy(generator)
+    return IterableDataset(
+        ex_iterable=BufferShuffledExamplesIterable(dataset._ex_iterable, buffer_size=buffer_size, generator=generator),
+        info=dataset._info.copy(),
+        split=dataset._split,
+        formatting=dataset._formatting,
+        shuffling=ShufflingConfig(generator=generator, _original_seed=seed),
+        distributed=copy.deepcopy(dataset._distributed),
+        token_per_repo_id=dataset._token_per_repo_id,
+    )
+
+
+@dataclass
+class DataCollatorForLanguageModeling:
+    """
+    Data collator used for language modeling. Inputs are dynamically padded if `varlen=False`.
+    If `varlen=True`, sequences are expected to be concatenated, and labels match inputs.
+
+    Args:
+        tokenizer ([`PreTrainedTokenizer`] or [`PreTrainedTokenizerFast`]):
+            The tokenizer used for encoding the data.
+        context_len (`int`, optional):
+            When `varlen=True`, sequences longer than this length within a document
+            (as determined by `cu_seqlens`) will be further chunked.
+        varlen (`bool`):
+            Whether to handle variable length concatenated sequences (`True`) or padded batches (`False`).
+
+    Returns:
+        A dictionary with the following keys:
+        - `input_ids`: Tensor of input IDs. Shape `[batch_size, seq_len]` if `varlen=False`, `[1, total_len]` if `varlen=True`.
+        - `labels`: Tensor of labels. Shape matches `input_ids`. Padding positions are masked with -100 if `varlen=False`.
+        - `attention_mask`: Tensor indicating non-padding tokens (only if `varlen=False`). Shape matches `input_ids`.
+        - `cu_seqlens`: Tensor of cumulative sequence lengths (only if `varlen=True`). Shape `[1, num_sequences + 1]`.
+
+    NOTE: When `varlen=True`, the `batch_size` must be 1.
+    """
+
+    tokenizer: PreTrainedTokenizer
+    context_len: Optional[int] = None
+    varlen: bool = False
+
+    def __call__(self, examples: List[Union[List[int], Dict[str, Any]]]) -> Dict[str, Any]:
+        if not isinstance(examples[0], Dict):
+            examples = [{'input_ids': example} for example in examples]
+
+        def tensorize(example: Dict[str, Any]) -> Dict[str, Any]:
+            tensorized = {}
+            for key in ['input_ids', 'cu_seqlens']:
+                if key not in example:
+                    continue
+                if isinstance(example[key], List):
+                    tensorized[key] = torch.tensor(example[key], dtype=torch.long)
+                elif isinstance(example[key], np.ndarray):
+                    tensorized[key] = torch.from_numpy(example[key])
+                else:
+                    tensorized[key] = example[key]
+            return tensorized
+
+        examples = list(map(tensorize, examples))
+
+        if not self.varlen:
+            # --- Handling for varlen=False (Batch Padding) ---
+            length_of_first = examples[0]['input_ids'].size(0)
+            needs_padding = not all(example['input_ids'].size(0) == length_of_first for example in examples)
+
+            if needs_padding:
+                # Check for pad token if padding is actually required
+                if self.tokenizer.pad_token_id is None:
+                    raise ValueError(
+                        f'You are attempting to pad samples but the tokenizer you are using '
+                        f'({self.tokenizer.__class__.__name__}) does not have a pad token.'
+                    )
+                # Pad using the tokenizer, ensuring attention_mask is returned
+                batch = self.tokenizer.pad(examples, return_tensors='pt', return_attention_mask=True)
+            else:
+                # No padding needed, stack directly and create a full attention mask
+                input_ids = torch.stack([example['input_ids'] for example in examples], dim=0)
+                batch = {
+                    'input_ids': input_ids,
+                    # Create attention mask of all ones
+                    'attention_mask': torch.ones_like(input_ids),
+                }
+
+            # Create labels by cloning input_ids
+            labels = batch['input_ids'].clone()
+            # Mask labels only where attention_mask is 0 (padding positions)
+            if 'attention_mask' in batch:
+                labels[batch['attention_mask'] == 0] = -100
+            batch['labels'] = labels
+
+        else:
+            # --- Handling for varlen=True (Concatenated Sequences) ---
+            if len(examples) > 1:
+                raise ValueError('The batch size must be 1 for inputs with variable lengths (varlen=True).')
+
+            batch = {'input_ids': torch.cat([example['input_ids'] for example in examples], dim=0).unsqueeze(0)}
+
+            # --- cu_seqlens calculation logic remains the same ---
+            if 'cu_seqlens' in examples[0]:
+                batch['cu_seqlens'] = (
+                    torch.cat([example['cu_seqlens'] for example in examples], dim=0).unsqueeze(0).to(dtype=torch.int32)
+                )  # Ensure int32
+            else:
+                # determine boundaries by bos/eos positions
+                # Check for bos_token_id first
+                if self.tokenizer.bos_token_id is not None:
+                    cu_seqlens = []
+                    # Handle case where the sequence doesn't start with BOS
+                    if batch['input_ids'][0, 0] != self.tokenizer.bos_token_id:
+                        cu_seqlens.append(torch.tensor([0], device=batch['input_ids'].device))  # Match device
+                    # Find all BOS token positions
+                    bos_positions = torch.where(batch['input_ids'].eq(self.tokenizer.bos_token_id))[1]
+                    # Ensure bos_positions is on the correct device if empty
+                    if bos_positions.numel() == 0 and len(cu_seqlens) > 0:
+                        cu_seqlens.append(bos_positions.to(cu_seqlens[0].device))
+                    elif bos_positions.numel() > 0:
+                        cu_seqlens.append(bos_positions)
+                    # Add the end of the entire batch
+                    cu_seqlens.append(
+                        torch.tensor([batch['input_ids'].size(1)], device=batch['input_ids'].device)
+                    )  # Match device and use size(1)
+                    # Filter out empty tensors before cat
+                    cu_seqlens = [t for t in cu_seqlens if t.numel() > 0]
+                    if not cu_seqlens:  # Handle case where input is empty or has no BOS
+                        batch['cu_seqlens'] = torch.tensor(
+                            [0, batch['input_ids'].size(1)], dtype=torch.int32, device=batch['input_ids'].device
+                        )
+                    else:
+                        batch['cu_seqlens'] = torch.cat(cu_seqlens, dim=0).to(dtype=torch.int32)
+
+                # Else, check for eos_token_id
+                elif self.tokenizer.eos_token_id is not None:
+                    cu_seqlens = [torch.tensor([0], device=batch['input_ids'].device)]  # Match device
+                    # Find positions *after* EOS tokens
+                    eos_positions = torch.where(batch['input_ids'].eq(self.tokenizer.eos_token_id))[1] + 1
+                    # Ensure eos_positions is on the correct device if empty
+                    if eos_positions.numel() > 0:
+                        cu_seqlens.append(eos_positions)
+                    # Handle case where the sequence doesn't end with EOS
+                    if batch['input_ids'][0, -1] != self.tokenizer.eos_token_id:
+                        # Only add the final length if the last found EOS wasn't already the end
+                        if eos_positions.numel() == 0 or eos_positions[-1] != batch['input_ids'].size(1):
+                            cu_seqlens.append(
+                                torch.tensor([batch['input_ids'].size(1)], device=batch['input_ids'].device)
+                            )  # Match device and use size(1)
+                    # Filter out empty tensors before cat
+                    cu_seqlens = [t for t in cu_seqlens if t.numel() > 0]
+                    if not cu_seqlens:  # Handle case where input is empty or has no EOS
+                        batch['cu_seqlens'] = torch.tensor(
+                            [0, batch['input_ids'].size(1)], dtype=torch.int32, device=batch['input_ids'].device
+                        )
+                    else:
+                        batch['cu_seqlens'] = torch.cat(cu_seqlens, dim=0).to(dtype=torch.int32)
+                # Else, neither BOS nor EOS is usable
+                else:
+                    raise ValueError(
+                        'For varlen=True without precomputed cu_seqlens, the tokenizer must have either a bos_token_id '
+                        'or an eos_token_id defined to act as sequence separators.'
+                    )
+
+                # --- cu_seqlens validation checks remain the same ---
+                if batch['cu_seqlens'].numel() < 2:
+                    raise ValueError(f'Calculated cu_seqlens must have at least start and end: {batch["cu_seqlens"]}')
+                if not torch.all(batch['cu_seqlens'][1:] >= batch['cu_seqlens'][:-1]):
+                    raise ValueError(f'Calculated cu_seqlens are not monotonically increasing: {batch["cu_seqlens"]}')
+                if batch['cu_seqlens'][0] != 0:
+                    raise ValueError(f'Calculated cu_seqlens do not start at 0: {batch["cu_seqlens"]}')
+                if batch['cu_seqlens'][-1] != batch['input_ids'].size(1):
+                    # Allow empty sequence case where cu_seqlens=[0, 0] and input_ids.size(1)=0
+                    if not (batch['cu_seqlens'].tolist() == [0, 0] and batch['input_ids'].size(1) == 0):
+                        raise ValueError(
+                            f'Calculated cu_seqlens do not end at total length {batch["input_ids"].size(1)}: '
+                            f'{batch["cu_seqlens"]}'
+                        )
+
+                # --- context_len splitting logic remains the same ---
+                if self.context_len is not None:
+                    # This logic splits sequences based on context_len *after* initial boundaries are found
+                    bos = batch['cu_seqlens'][:-1].tolist()
+                    eos = batch['cu_seqlens'][1:].tolist()
+                    # Handle empty sequences between boundaries
+                    split_boundaries = []
+                    for i, j in zip(bos, eos):
+                        if i < j:  # Only process non-empty sequences
+                            split_boundaries.append(torch.arange(i, j, self.context_len, device=batch['input_ids'].device))
+                    # Add the final end point if it wasn't included by arange
+                    final_end_point = torch.tensor([batch['input_ids'].size(1)], device=batch['input_ids'].device)
+                    # Concatenate all boundaries
+                    if not split_boundaries:  # Handle case of completely empty input
+                        batch['cu_seqlens'] = torch.tensor([0, 0], dtype=torch.int32, device=batch['input_ids'].device)
+                    else:
+                        batch['cu_seqlens'] = torch.cat(split_boundaries + [final_end_point]).to(dtype=torch.int32)
+                        # Ensure uniqueness and sort, as arange might duplicate the endpoint
+                        batch['cu_seqlens'] = torch.unique(batch['cu_seqlens'])
+
+            # Create labels directly from input_ids, NO padding mask needed for varlen
+            labels = batch['input_ids'].clone()
+            batch['labels'] = labels
+
+        return batch
+
+
+class ParallelAwareDataLoader(StatefulDataLoader, Stateful):
+    """
+    A wrapper around the StatefulDataLoader that ensures that the state is stored only once per DP rank.
+    """
+
+    def __init__(
+        self,
+        rank: int,
+        dataset: IterableDataset,
+        batch_size: int,
+        collate_fn: Callable,
+        num_workers: int = 0,
+        pin_memory: bool = False,
+        prefetch_factor: int = 2,
+        persistent_workers: bool = False,
+        snapshot_every_n_steps: Optional[int] = 1,
+    ):
+        super().__init__(
+            dataset=dataset,
+            batch_size=batch_size,
+            collate_fn=collate_fn,
+            num_workers=num_workers,
+            pin_memory=pin_memory,
+            prefetch_factor=prefetch_factor,
+            persistent_workers=persistent_workers,
+            snapshot_every_n_steps=snapshot_every_n_steps,
+        )
+        self.rank = rank
+
+    def state_dict(self) -> Dict[str, Any]:
+        # Store state only for dp rank to avoid replicating the same state across other dimensions
+        return {f'rank_{self.rank}': pickle.dumps(super().state_dict())}
+
+    def load_state_dict(self, state_dict: Dict[str, Any]) -> None:
+        # State being empty is valid
+        if not state_dict:
+            return
+
+        if f'rank_{self.rank}' not in state_dict:
+            logger.warning(f'DataLoader state is empty for dp rank {self.rank}, expected key rank_{self.rank}')
+            return
+        super().load_state_dict(pickle.loads(state_dict[f'rank_{self.rank}']))
+
+
+def build_dataloader(
+    dataset: IterableDataset,
+    tokenizer: PreTrainedTokenizer,
+    rank: int,
+    world_size: int,
+    batch_size: int,
+    seq_len: int,
+    context_len: Optional[int] = None,
+    varlen: bool = False,
+    num_workers: int = 0,
+    pin_memory: bool = False,
+    persistent_workers: bool = False,
+    snapshot_every_n_steps: Optional[int] = 1,
+):
+    dataset = OnlineTokenizedIterableDataset(
+        dataset=dataset, tokenizer=tokenizer, seq_len=seq_len, rank=rank, world_size=world_size
+    )
+    return ParallelAwareDataLoader(
+        rank=rank,
+        dataset=dataset,
+        batch_size=batch_size,
+        collate_fn=DataCollatorForLanguageModeling(tokenizer=tokenizer, context_len=context_len, varlen=varlen),
+        num_workers=num_workers,
+        pin_memory=pin_memory,
+        persistent_workers=persistent_workers,
+        snapshot_every_n_steps=snapshot_every_n_steps,
+    )

logs/none_75lcom2m/attempt_0/3/stderr.log

@@ -0,0 +1,17 @@
+Traceback (most recent call last):
+  File "<frozen runpy>", line 198, in _run_module_as_main
+  File "<frozen runpy>", line 88, in _run_code
+  File "/workspace/flame/flame/train.py", line 19, in <module>
+    import fla  # noqa
+    ^^^^^^^^^^
+  File "/workspace/flame/fla/__init__.py", line 23, in <module>
+    from fla.models import (
+  File "/workspace/flame/fla/models/__init__.py", line 4, in <module>
+    from fla.models.bitnet import BitNetConfig, BitNetForCausalLM, BitNetModel
+  File "/workspace/flame/fla/models/bitnet/__init__.py", line 8, in <module>
+    AutoConfig.register(BitNetConfig.model_type, BitNetConfig)
+  File "/workspace/flame/.venv/lib/python3.11/site-packages/transformers/models/auto/configuration_auto.py", line 1211, in register
+    CONFIG_MAPPING.register(model_type, config, exist_ok=exist_ok)
+  File "/workspace/flame/.venv/lib/python3.11/site-packages/transformers/models/auto/configuration_auto.py", line 905, in register
+    raise ValueError(f"'{key}' is already used by a Transformers config, pick another name.")
+ValueError: 'bitnet' is already used by a Transformers config, pick another name.