Add files using upload-large-folder tool

README.md

@@ -0,0 +1,39 @@
+---
+language:
+- en
+- zh
+tags:
+- medical
+- mlx
+base_model: baichuan-inc/Baichuan-M1-14B-Instruct
+pipeline_tag: text-generation
+library_name: mlx
+---
+
+# mlx-community/Baichuan-M1-14B-Instruct
+
+This model [mlx-community/Baichuan-M1-14B-Instruct](https://huggingface.co/mlx-community/Baichuan-M1-14B-Instruct) was
+converted to MLX format from [baichuan-inc/Baichuan-M1-14B-Instruct](https://huggingface.co/baichuan-inc/Baichuan-M1-14B-Instruct)
+using mlx-lm version **0.23.2**.
+
+## Use with mlx
+
+```bash
+pip install mlx-lm
+```
+
+```python
+from mlx_lm import load, generate
+
+model, tokenizer = load("mlx-community/Baichuan-M1-14B-Instruct")
+
+prompt = "hello"
+
+if tokenizer.chat_template is not None:
+    messages = [{"role": "user", "content": prompt}]
+    prompt = tokenizer.apply_chat_template(
+        messages, add_generation_prompt=True
+    )
+
+response = generate(model, tokenizer, prompt=prompt, verbose=True)
+```

config.json

@@ -0,0 +1,56 @@
+{
+    "architectures": [
+        "BaichuanM1ForCausalLM"
+    ],
+    "attention_dropout": 0.0,
+    "auto_map": {
+        "AutoConfig": "configuration_baichuan.BaichuanM1Config",
+        "AutoModelForCausalLM": "modeling_baichuan.BaichuanM1ForCausalLM"
+    },
+    "bos_token_id": 1,
+    "conv_window": 2,
+    "eos_token_id": 2,
+    "hidden_act": "silu",
+    "hidden_size": 5120,
+    "initializer_range": 0.02,
+    "intermediate_size": 17408,
+    "max_position_embeddings": 32768,
+    "model_max_length": 32768,
+    "model_type": "baichuan_m1",
+    "num_attention_heads": 20,
+    "num_hidden_layers": 40,
+    "num_key_value_heads": 2,
+    "num_swa_attention_heads": 40,
+    "num_swa_key_value_heads": 8,
+    "pad_token_id": 0,
+    "rms_norm_eps": 1e-06,
+    "rope_theta": 1000000.0,
+    "sliding_window": 8192,
+    "sliding_window_layers": [
+        1,
+        3,
+        5,
+        7,
+        9,
+        11,
+        13,
+        15,
+        17,
+        19,
+        21,
+        23,
+        25,
+        27,
+        29,
+        31,
+        33,
+        35,
+        37,
+        39
+    ],
+    "tie_word_embeddings": false,
+    "torch_dtype": "bfloat16",
+    "transformers_version": "4.48.1",
+    "use_cache": true,
+    "vocab_size": 133120
+}

configuration_baichuan.py

@@ -0,0 +1,119 @@
+# coding=utf-8
+#
+# 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.
+
+from transformers import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+
+class BaichuanM1Config(PretrainedConfig):
+    r"""
+    Configuration objects inherit from [`PretrainedConfig`] and control the behavior of model outputs. For more details, 
+    refer to the documentation of [`PretrainedConfig`].
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 133120):
+            The size of the vocabulary used by the model.
+        hidden_size (`int`, *optional*, defaults to 4096):
+            The dimensionality of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 22016):
+            The dimensionality of the intermediate (MLP) representations.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            The number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            The number of attention heads for each attention layer in the Transformer encoder.
+        num_key_value_heads (`int`, *optional*, defaults to 32):
+            The number of key-value heads used to implement Grouped Query Attention (GQA). 
+            - If `num_key_value_heads == num_attention_heads`, the model uses Multi-Head Attention (MHA).
+            - If `num_key_value_heads == 1`, the model uses Multi-Query Attention (MQA).
+            - Otherwise, the model uses Grouped Query Attention (GQA). 
+            When converting a multi-head checkpoint to a GQA checkpoint, each group's key and value heads are constructed 
+            by mean-pooling the original heads within that group. For more details, refer to [this paper](https://arxiv.org/pdf/2305.13245.pdf). 
+            If not specified, this defaults to `32`.
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (either a string or a callable function) used in the decoder.
+        max_position_embeddings (`int`, *optional*, defaults to 32768):
+            The maximum sequence length the model can handle.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated normal initializer for initializing all weight matrices.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-06):
+            The epsilon value used by the RMS normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether the model should return the last key/value attentions. This is only relevant if `config.is_decoder=True`.
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether to tie the model's input and output word embeddings.
+        rope_theta (`float`, *optional*, defaults to 10000.0):
+            The base period of the Rotary Position Embeddings (RoPE).
+        use_sliding_window (`bool`, *optional*, defaults to `False`):
+            Whether to enable sliding window attention.
+        sliding_window (`int`, *optional*, defaults to 4096):
+            The size of the sliding window for sliding window attention (SWA). If not specified, it defaults to `2048`.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio applied to the attention probabilities.
+    """
+
+    model_type = "baichuan"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+            self,
+            vocab_size=133120,
+            hidden_size=5120,
+            intermediate_size=17408,
+            num_hidden_layers=40,
+            num_attention_heads=40,
+            num_key_value_heads=2,
+            num_swa_attention_heads: int = 20,
+            num_swa_key_value_heads=8,
+            sliding_window_layers: list = None,
+            hidden_act="silu",
+            max_position_embeddings=32768,
+            initializer_range=0.02,
+            rms_norm_eps=1e-6,
+            use_cache=True,
+            tie_word_embeddings=False,
+            rope_theta=100000.0,
+            sliding_window=2048,
+            attention_dropout=0.0,
+            conv_window = 2,
+            **kwargs,
+    ):
+        self.sliding_window_layers = sliding_window_layers
+        self.num_swa_key_value_heads = num_swa_key_value_heads
+        self.num_swa_attention_heads = num_swa_attention_heads
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.sliding_window = sliding_window
+
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.attention_dropout = attention_dropout
+        self.conv_window = conv_window
+        super().__init__(
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )

model-00001-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b105daa6506025d4089d5e8652858aa654f2553e0c5d488bd2b828f64450a65c
+size 5358352875

model-00002-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7369b201c0555e531198af8ebccb7e0374c0a06204056287e05bc7866f10c78e
+size 5295481563

model-00003-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b72d3ee28d023a41cc30ba604ba023eb6ef707b9fabb529bd3c899fb8627ab47
+size 5243052458

model-00004-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59e516bdfdf3ebf8178258e22872084944d47c190952062d317c3ec5344a1c0c
+size 5243052448

model-00005-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d996bd0881187f8fd10d715539d9e5c293e39b4ace2fbcc445accef68e77aadf
+size 5243052428

model-00006-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf9ef33fb9d8780e4c4a189bd3f2552821277df855bc6fdc20714afb37fa6975
+size 2558578518

model.safetensors.index.json

@@ -0,0 +1,370 @@
+{
+    "metadata": {
+        "total_size": 28941528640
+    },
+    "weight_map": {
+        "lm_head.weight": "model-00006-of-00006.safetensors",
+        "model.embed_tokens.weight": "model-00001-of-00006.safetensors",
+        "model.layers.0.input_layernorm.weight": "model-00001-of-00006.safetensors",
+        "model.layers.0.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.0.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+        "model.layers.0.self_attn.W_pack.weight": "model-00001-of-00006.safetensors",
+        "model.layers.0.self_attn.conv_k": "model-00001-of-00006.safetensors",
+        "model.layers.0.self_attn.conv_v": "model-00001-of-00006.safetensors",
+        "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.1.input_layernorm.weight": "model-00001-of-00006.safetensors",
+        "model.layers.1.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.1.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+        "model.layers.1.self_attn.W_pack.weight": "model-00001-of-00006.safetensors",
+        "model.layers.1.self_attn.conv_k": "model-00001-of-00006.safetensors",
+        "model.layers.1.self_attn.conv_v": "model-00001-of-00006.safetensors",
+        "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.10.input_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.10.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.10.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.10.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.10.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.10.self_attn.W_pack.weight": "model-00002-of-00006.safetensors",
+        "model.layers.10.self_attn.conv_k": "model-00002-of-00006.safetensors",
+        "model.layers.10.self_attn.conv_v": "model-00002-of-00006.safetensors",
+        "model.layers.10.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.11.input_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.11.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.11.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.11.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.11.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.11.self_attn.W_pack.weight": "model-00002-of-00006.safetensors",
+        "model.layers.11.self_attn.conv_k": "model-00002-of-00006.safetensors",
+        "model.layers.11.self_attn.conv_v": "model-00002-of-00006.safetensors",
+        "model.layers.11.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.12.input_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.12.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.12.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.12.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.12.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.12.self_attn.W_pack.weight": "model-00002-of-00006.safetensors",
+        "model.layers.12.self_attn.conv_k": "model-00002-of-00006.safetensors",
+        "model.layers.12.self_attn.conv_v": "model-00002-of-00006.safetensors",
+        "model.layers.12.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.13.input_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.13.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.13.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.13.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.13.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.13.self_attn.W_pack.weight": "model-00002-of-00006.safetensors",
+        "model.layers.13.self_attn.conv_k": "model-00002-of-00006.safetensors",
+        "model.layers.13.self_attn.conv_v": "model-00002-of-00006.safetensors",
+        "model.layers.13.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.14.input_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.14.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.14.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.14.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.14.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.14.self_attn.W_pack.weight": "model-00002-of-00006.safetensors",
+        "model.layers.14.self_attn.conv_k": "model-00002-of-00006.safetensors",
+        "model.layers.14.self_attn.conv_v": "model-00002-of-00006.safetensors",
+        "model.layers.14.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.15.input_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.15.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.15.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.15.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.15.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.15.self_attn.W_pack.weight": "model-00003-of-00006.safetensors",
+        "model.layers.15.self_attn.conv_k": "model-00003-of-00006.safetensors",
+        "model.layers.15.self_attn.conv_v": "model-00003-of-00006.safetensors",
+        "model.layers.15.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.16.input_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.16.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.16.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.16.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.16.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.16.self_attn.W_pack.weight": "model-00003-of-00006.safetensors",
+        "model.layers.16.self_attn.conv_k": "model-00003-of-00006.safetensors",
+        "model.layers.16.self_attn.conv_v": "model-00003-of-00006.safetensors",
+        "model.layers.16.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.17.input_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.17.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.17.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.17.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.17.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.17.self_attn.W_pack.weight": "model-00003-of-00006.safetensors",
+        "model.layers.17.self_attn.conv_k": "model-00003-of-00006.safetensors",
+        "model.layers.17.self_attn.conv_v": "model-00003-of-00006.safetensors",
+        "model.layers.17.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.18.input_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.18.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.18.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.18.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.18.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.18.self_attn.W_pack.weight": "model-00003-of-00006.safetensors",
+        "model.layers.18.self_attn.conv_k": "model-00003-of-00006.safetensors",
+        "model.layers.18.self_attn.conv_v": "model-00003-of-00006.safetensors",
+        "model.layers.18.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.19.input_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.19.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.19.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.19.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.19.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.19.self_attn.W_pack.weight": "model-00003-of-00006.safetensors",
+        "model.layers.19.self_attn.conv_k": "model-00003-of-00006.safetensors",
+        "model.layers.19.self_attn.conv_v": "model-00003-of-00006.safetensors",
+        "model.layers.19.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.2.input_layernorm.weight": "model-00001-of-00006.safetensors",
+        "model.layers.2.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.2.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.2.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+        "model.layers.2.self_attn.W_pack.weight": "model-00001-of-00006.safetensors",
+        "model.layers.2.self_attn.conv_k": "model-00001-of-00006.safetensors",
+        "model.layers.2.self_attn.conv_v": "model-00001-of-00006.safetensors",
+        "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.20.input_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.20.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.20.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.20.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.20.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.20.self_attn.W_pack.weight": "model-00003-of-00006.safetensors",
+        "model.layers.20.self_attn.conv_k": "model-00003-of-00006.safetensors",
+        "model.layers.20.self_attn.conv_v": "model-00003-of-00006.safetensors",
+        "model.layers.20.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.21.input_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.21.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.21.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.21.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.21.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+        "model.layers.21.self_attn.W_pack.weight": "model-00003-of-00006.safetensors",
+        "model.layers.21.self_attn.conv_k": "model-00003-of-00006.safetensors",
+        "model.layers.21.self_attn.conv_v": "model-00003-of-00006.safetensors",
+        "model.layers.21.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.22.input_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.22.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.22.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.22.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.22.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.22.self_attn.W_pack.weight": "model-00003-of-00006.safetensors",
+        "model.layers.22.self_attn.conv_k": "model-00003-of-00006.safetensors",
+        "model.layers.22.self_attn.conv_v": "model-00003-of-00006.safetensors",
+        "model.layers.22.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+        "model.layers.23.input_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.23.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.23.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.23.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.23.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.23.self_attn.W_pack.weight": "model-00004-of-00006.safetensors",
+        "model.layers.23.self_attn.conv_k": "model-00004-of-00006.safetensors",
+        "model.layers.23.self_attn.conv_v": "model-00004-of-00006.safetensors",
+        "model.layers.23.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.24.input_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.24.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.24.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.24.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.24.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.24.self_attn.W_pack.weight": "model-00004-of-00006.safetensors",
+        "model.layers.24.self_attn.conv_k": "model-00004-of-00006.safetensors",
+        "model.layers.24.self_attn.conv_v": "model-00004-of-00006.safetensors",
+        "model.layers.24.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.25.input_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.25.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.25.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.25.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.25.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.25.self_attn.W_pack.weight": "model-00004-of-00006.safetensors",
+        "model.layers.25.self_attn.conv_k": "model-00004-of-00006.safetensors",
+        "model.layers.25.self_attn.conv_v": "model-00004-of-00006.safetensors",
+        "model.layers.25.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.26.input_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.26.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.26.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.26.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.26.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.26.self_attn.W_pack.weight": "model-00004-of-00006.safetensors",
+        "model.layers.26.self_attn.conv_k": "model-00004-of-00006.safetensors",
+        "model.layers.26.self_attn.conv_v": "model-00004-of-00006.safetensors",
+        "model.layers.26.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.27.input_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.27.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.27.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.27.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.27.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.27.self_attn.W_pack.weight": "model-00004-of-00006.safetensors",
+        "model.layers.27.self_attn.conv_k": "model-00004-of-00006.safetensors",
+        "model.layers.27.self_attn.conv_v": "model-00004-of-00006.safetensors",
+        "model.layers.27.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.28.input_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.28.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.28.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.28.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.28.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.28.self_attn.W_pack.weight": "model-00004-of-00006.safetensors",
+        "model.layers.28.self_attn.conv_k": "model-00004-of-00006.safetensors",
+        "model.layers.28.self_attn.conv_v": "model-00004-of-00006.safetensors",
+        "model.layers.28.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.29.input_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.29.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.29.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.29.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.29.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+        "model.layers.29.self_attn.W_pack.weight": "model-00004-of-00006.safetensors",
+        "model.layers.29.self_attn.conv_k": "model-00004-of-00006.safetensors",
+        "model.layers.29.self_attn.conv_v": "model-00004-of-00006.safetensors",
+        "model.layers.29.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.3.input_layernorm.weight": "model-00001-of-00006.safetensors",
+        "model.layers.3.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.3.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+        "model.layers.3.self_attn.W_pack.weight": "model-00001-of-00006.safetensors",
+        "model.layers.3.self_attn.conv_k": "model-00001-of-00006.safetensors",
+        "model.layers.3.self_attn.conv_v": "model-00001-of-00006.safetensors",
+        "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.30.input_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.30.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.30.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.30.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.30.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.30.self_attn.W_pack.weight": "model-00004-of-00006.safetensors",
+        "model.layers.30.self_attn.conv_k": "model-00004-of-00006.safetensors",
+        "model.layers.30.self_attn.conv_v": "model-00004-of-00006.safetensors",
+        "model.layers.30.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+        "model.layers.31.input_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.31.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.31.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.31.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.31.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.31.self_attn.W_pack.weight": "model-00005-of-00006.safetensors",
+        "model.layers.31.self_attn.conv_k": "model-00005-of-00006.safetensors",
+        "model.layers.31.self_attn.conv_v": "model-00005-of-00006.safetensors",
+        "model.layers.31.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.32.input_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.32.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.32.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.32.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.32.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.32.self_attn.W_pack.weight": "model-00005-of-00006.safetensors",
+        "model.layers.32.self_attn.conv_k": "model-00005-of-00006.safetensors",
+        "model.layers.32.self_attn.conv_v": "model-00005-of-00006.safetensors",
+        "model.layers.32.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.33.input_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.33.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.33.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.33.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.33.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.33.self_attn.W_pack.weight": "model-00005-of-00006.safetensors",
+        "model.layers.33.self_attn.conv_k": "model-00005-of-00006.safetensors",
+        "model.layers.33.self_attn.conv_v": "model-00005-of-00006.safetensors",
+        "model.layers.33.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.34.input_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.34.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.34.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.34.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.34.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.34.self_attn.W_pack.weight": "model-00005-of-00006.safetensors",
+        "model.layers.34.self_attn.conv_k": "model-00005-of-00006.safetensors",
+        "model.layers.34.self_attn.conv_v": "model-00005-of-00006.safetensors",
+        "model.layers.34.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.35.input_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.35.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.35.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.35.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.35.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.35.self_attn.W_pack.weight": "model-00005-of-00006.safetensors",
+        "model.layers.35.self_attn.conv_k": "model-00005-of-00006.safetensors",
+        "model.layers.35.self_attn.conv_v": "model-00005-of-00006.safetensors",
+        "model.layers.35.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.36.input_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.36.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.36.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.36.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.36.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.36.self_attn.W_pack.weight": "model-00005-of-00006.safetensors",
+        "model.layers.36.self_attn.conv_k": "model-00005-of-00006.safetensors",
+        "model.layers.36.self_attn.conv_v": "model-00005-of-00006.safetensors",
+        "model.layers.36.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.37.input_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.37.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.37.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.37.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.37.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+        "model.layers.37.self_attn.W_pack.weight": "model-00005-of-00006.safetensors",
+        "model.layers.37.self_attn.conv_k": "model-00005-of-00006.safetensors",
+        "model.layers.37.self_attn.conv_v": "model-00005-of-00006.safetensors",
+        "model.layers.37.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.38.input_layernorm.weight": "model-00006-of-00006.safetensors",
+        "model.layers.38.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+        "model.layers.38.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+        "model.layers.38.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+        "model.layers.38.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+        "model.layers.38.self_attn.W_pack.weight": "model-00005-of-00006.safetensors",
+        "model.layers.38.self_attn.conv_k": "model-00005-of-00006.safetensors",
+        "model.layers.38.self_attn.conv_v": "model-00005-of-00006.safetensors",
+        "model.layers.38.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+        "model.layers.39.input_layernorm.weight": "model-00006-of-00006.safetensors",
+        "model.layers.39.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+        "model.layers.39.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+        "model.layers.39.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+        "model.layers.39.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+        "model.layers.39.self_attn.W_pack.weight": "model-00006-of-00006.safetensors",
+        "model.layers.39.self_attn.conv_k": "model-00006-of-00006.safetensors",
+        "model.layers.39.self_attn.conv_v": "model-00006-of-00006.safetensors",
+        "model.layers.39.self_attn.o_proj.weight": "model-00006-of-00006.safetensors",
+        "model.layers.4.input_layernorm.weight": "model-00001-of-00006.safetensors",
+        "model.layers.4.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.4.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.4.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+        "model.layers.4.self_attn.W_pack.weight": "model-00001-of-00006.safetensors",
+        "model.layers.4.self_attn.conv_k": "model-00001-of-00006.safetensors",
+        "model.layers.4.self_attn.conv_v": "model-00001-of-00006.safetensors",
+        "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.5.input_layernorm.weight": "model-00001-of-00006.safetensors",
+        "model.layers.5.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.5.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.5.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+        "model.layers.5.self_attn.W_pack.weight": "model-00001-of-00006.safetensors",
+        "model.layers.5.self_attn.conv_k": "model-00001-of-00006.safetensors",
+        "model.layers.5.self_attn.conv_v": "model-00001-of-00006.safetensors",
+        "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+        "model.layers.6.input_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.6.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.6.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.6.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.6.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.6.self_attn.W_pack.weight": "model-00001-of-00006.safetensors",
+        "model.layers.6.self_attn.conv_k": "model-00002-of-00006.safetensors",
+        "model.layers.6.self_attn.conv_v": "model-00002-of-00006.safetensors",
+        "model.layers.6.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.7.input_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.7.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.7.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.7.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.7.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.7.self_attn.W_pack.weight": "model-00002-of-00006.safetensors",
+        "model.layers.7.self_attn.conv_k": "model-00002-of-00006.safetensors",
+        "model.layers.7.self_attn.conv_v": "model-00002-of-00006.safetensors",
+        "model.layers.7.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.8.input_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.8.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.8.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.8.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.8.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.8.self_attn.W_pack.weight": "model-00002-of-00006.safetensors",
+        "model.layers.8.self_attn.conv_k": "model-00002-of-00006.safetensors",
+        "model.layers.8.self_attn.conv_v": "model-00002-of-00006.safetensors",
+        "model.layers.8.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.9.input_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.9.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.9.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.9.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+        "model.layers.9.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+        "model.layers.9.self_attn.W_pack.weight": "model-00002-of-00006.safetensors",
+        "model.layers.9.self_attn.conv_k": "model-00002-of-00006.safetensors",
+        "model.layers.9.self_attn.conv_v": "model-00002-of-00006.safetensors",
+        "model.layers.9.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+        "model.norm.weight": "model-00006-of-00006.safetensors"
+    }
+}

modeling_baichuan.py

@@ -0,0 +1,1197 @@
+import math
+import os
+from typing import List, Optional, Tuple, Union, Dict, Any
+
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from einops import rearrange
+from torch import nn
+from torch.nn import CrossEntropyLoss
+from transformers import add_start_docstrings, PreTrainedModel, DynamicCache, \
+    GenerationMixin, StaticCache, GenerationConfig
+from transformers.activations import ACT2FN
+from transformers.modeling_attn_mask_utils import AttentionMaskConverter
+from transformers.modeling_flash_attention_utils import _flash_supports_window_size, \
+    _upad_input
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.utils import is_flash_attn_2_available, is_flash_attn_greater_or_equal_2_10, \
+    add_start_docstrings_to_model_forward, is_torchdynamo_compiling, logging, \
+    is_flash_attn_greater_or_equal
+
+if is_flash_attn_2_available():
+    from flash_attn.bert_padding import  pad_input
+    from flash_attn import flash_attn_func, flash_attn_varlen_func
+    from flash_attn.layers.rotary import apply_rotary_emb_func
+from .configuration_baichuan import BaichuanM1Config
+
+logger = logging.get_logger(__name__)
+
+
+class CustomCache(DynamicCache):
+    def __init__(self):
+        super().__init__()
+        self.past_len = []
+
+    def get_past_len(self, layer_idx: Optional[int] = 0) -> int:
+        if len(self.past_len) <= layer_idx:
+            return 0
+        return self.past_len[layer_idx]
+
+    def get_seq_length(self, layer_idx: Optional[int] = 0) -> int:
+        """Returns the sequence length of the cached states. A layer index can be optionally passed."""
+        # TODO: deprecate this function in favor of `cache_position`
+        if len(self.key_cache) <= layer_idx:
+            return 0
+        return self.key_cache[layer_idx].shape[1]
+
+    def update(
+            self,
+            key_states: torch.Tensor,
+            value_states: torch.Tensor,
+            layer_idx: int,
+            cache_kwargs: Optional[Dict[str, Any]] = None,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """
+        Updates the cache with the new `key_states` and `value_states` for the layer `layer_idx`.
+
+        Parameters:
+            key_states (`torch.Tensor`):
+                The new key states to cache.
+            value_states (`torch.Tensor`):
+                The new value states to cache.
+            layer_idx (`int`):
+                The index of the layer to cache the states for.
+            cache_kwargs (`Dict[str, Any]`, `optional`):
+                Additional arguments for the cache subclass. No additional arguments are used in `DynamicCache`.
+
+        Return:
+            A tuple containing the updated key and value states.
+        """
+        # Update the number of seen tokens
+        if layer_idx == 0:
+            self._seen_tokens += key_states.shape[1]
+
+        # Update the cache
+        if len(self.key_cache) <= layer_idx:
+            self.key_cache.append(key_states)
+            self.value_cache.append(value_states)
+        else:
+            self.key_cache[layer_idx] = torch.cat([self.key_cache[layer_idx], key_states], dim=1)
+            self.value_cache[layer_idx] = torch.cat([self.value_cache[layer_idx], value_states], dim=1)
+
+        if len(self.past_len) <= layer_idx:
+            self.past_len.append(key_states.shape[1])
+        else:
+            self.past_len[layer_idx] += key_states.shape[1]
+
+        return self.key_cache[layer_idx], self.value_cache[layer_idx]
+
+
+def _prepare_4d_causal_attention_mask_with_cache_position(
+        attention_mask: torch.Tensor,
+        sequence_length: int,
+        target_length: int,
+        dtype: torch.dtype,
+        device: torch.device,
+        min_dtype: float,
+        cache_position: torch.Tensor,
+        batch_size: int,
+):
+    """
+    Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape
+    `(batch_size, key_value_length)`, or if the input `attention_mask` is already 4D, do nothing.
+
+    Args:
+        attention_mask (`torch.Tensor`):
+            A 2D attention mask of shape `(batch_size, key_value_length)` or a 4D attention mask of shape `(batch_size, 1, query_length, key_value_length)`.
+        sequence_length (`int`):
+            The sequence length being processed.
+        target_length (`int`):
+            The target length: when generating with static cache, the mask should be as long as the static cache, to account for the 0 padding, the part of the cache that is not filled yet.
+        dtype (`torch.dtype`):
+            The dtype to use for the 4D attention mask.
+        device (`torch.device`):
+            The device to plcae the 4D attention mask on.
+        min_dtype (`float`):
+            The minimum value representable with the dtype `dtype`.
+        cache_position (`torch.Tensor`):
+            Indices depicting the position of the input sequence tokens in the sequence.
+        batch_size (`torch.Tensor`):
+            Batch size.
+    """
+    if attention_mask is not None and attention_mask.dim() == 4:
+        # In this case we assume that the mask comes already in inverted form and requires no inversion or slicing.
+        causal_mask = attention_mask
+    else:
+        causal_mask = torch.full((sequence_length, target_length), fill_value=min_dtype, dtype=dtype, device=device)
+        if sequence_length != 1:
+            causal_mask = torch.triu(causal_mask, diagonal=1)
+        causal_mask *= torch.arange(target_length, device=device) > cache_position.reshape(-1, 1)
+        causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
+        if attention_mask is not None:
+            causal_mask = causal_mask.clone()  # copy to contiguous memory for in-place edit
+            mask_length = attention_mask.shape[-1]
+            padding_mask = causal_mask[:, :, :, :mask_length] + attention_mask[:, None, None, :]
+            padding_mask = padding_mask == 0
+            causal_mask[:, :, :, :mask_length] = causal_mask[:, :, :, :mask_length].masked_fill(
+                padding_mask, min_dtype
+            )
+
+    return causal_mask
+
+
+class BaichuanRMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        RMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+
+        # convert into half-precision if necessary
+        if self.weight.dtype in [torch.float16, torch.bfloat16]:
+            hidden_states = hidden_states.to(self.weight.dtype)
+
+        return self.weight * hidden_states
+
+
+class RotaryEmbedding(torch.nn.Module):
+    def __init__(self, dim, max_position_embeddings=2048, base=1e5, device=None, interleaved=False):
+        super().__init__()
+        self.inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float() / dim))
+        self.base = base
+        self.dim = dim
+        # Build here to make `torch.jit.trace` work.
+        self.max_seq_len_cached = 0
+        self.interleaved = interleaved
+
+    def forward(self, q, k, seqlen_offset=None, cu_seqlens=None, max_seqlen=None):
+        # x: [bs, num_attention_heads, seq_len, head_size]
+        # This `if` block is unlikely to be run after we build sin/cos in `__init__`. Keep the logic here just in case.
+        seq_len_dim = 1
+        seq_len = q.shape[seq_len_dim] + seqlen_offset
+        if seq_len > self.max_seq_len_cached:
+            self.max_seq_len_cached = seq_len
+            self.inv_freq = 1.0 / (
+                    self.base ** (torch.arange(0, self.dim, 2).float().to(self.inv_freq.device) / self.dim))
+            t = torch.arange(self.max_seq_len_cached, device=self.inv_freq.device, dtype=self.inv_freq.dtype)
+            # freqs = torch.einsum("i,j->ij", t, self.inv_freq) # dont use this, bug in fp16
+            freqs = torch.outer(t, self.inv_freq)
+            self.cos_cached = freqs.cos().to(q.device)
+            self.sin_cached = freqs.sin().to(k.device)
+        q_ori_size = q.size()
+        k_ori_size = k.size()
+        if cu_seqlens is not None:
+            q = flatten_one_dim(q)
+            k = flatten_one_dim(k)
+        q_new = apply_rotary_emb_func(
+            q.float(), self.cos_cached[seqlen_offset:], self.sin_cached[seqlen_offset:],
+            self.interleaved, True,  # inplace=True
+            cu_seqlens=cu_seqlens,
+            max_seqlen=max_seqlen
+        ).to(q.dtype)
+        k_new = apply_rotary_emb_func(
+            k.float(), self.cos_cached[seqlen_offset:], self.sin_cached[seqlen_offset:],
+            self.interleaved, True,
+            cu_seqlens=cu_seqlens,
+            max_seqlen=max_seqlen
+        ).to(k.dtype)
+        if cu_seqlens is not None:
+            q_new = q_new.reshape(*q_ori_size)
+            k_new = k_new.reshape(*k_ori_size)
+        return q_new, k_new
+
+
+class BaichuanMLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = config.intermediate_size
+        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, hidden_state):
+        return self.down_proj(self.act_fn(self.gate_proj(hidden_state)) * self.up_proj(hidden_state))
+
+
+class BaichuanAttention(nn.Module):
+    """
+    Multi-headed attention from 'Attention Is All You Need' paper. Modified to use sliding window attention: Longformer
+    and "Generating Long Sequences with Sparse Transformers".
+    """
+
+    def __init__(self, config: BaichuanM1Config, layer_idx: Optional[int] = None):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        if layer_idx is None:
+            raise ValueError(
+                f"Instantiating {self.__class__.__name__} without passing `layer_idx` is not recommended and will "
+                "to errors during the forward call, if caching is used. Please make sure to provide a `layer_idx` "
+                "when creating this class."
+            )
+
+        self.hidden_size = config.hidden_size
+        self.is_swa = layer_idx in self.config.sliding_window_layers
+        self.num_heads = config.num_swa_attention_heads if self.is_swa else config.num_attention_heads
+        self.head_dim = self.hidden_size // self.num_heads
+        self.num_key_value_heads = config.num_swa_key_value_heads if self.is_swa else config.num_key_value_heads
+        self.max_position_embeddings = config.max_position_embeddings
+        self.rope_theta = config.rope_theta
+        self.is_causal = True
+        self.attention_dropout = config.attention_dropout
+
+        if (self.head_dim * self.num_heads) != self.hidden_size:
+            raise ValueError(
+                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
+                f" and `num_heads`: {self.num_heads})."
+            )
+        self.W_pack = nn.Linear(config.hidden_size, self.hidden_size + 2 * self.num_key_value_heads * self.head_dim,
+                                bias=False)
+        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
+
+        self.rotary_emb = RotaryEmbedding(dim=self.head_dim, base=self.config.rope_theta,
+                                          max_position_embeddings=self.config.max_position_embeddings)
+        self.conv_window = config.conv_window
+        assert self.conv_window == 2 #%% Currently, only supported window=2 when inference
+        self.conv_k = nn.Parameter(torch.softmax(torch.randn((1, 1, self.num_key_value_heads, 1, self.conv_window)), dim=-1))
+        self.conv_v = nn.Parameter(torch.softmax(torch.randn((1, 1, self.num_key_value_heads, 1, self.conv_window)), dim=-1))
+        self.last_k, self.last_v = None, None
+
+
+def get_max_seqlen(cu_seqlens):
+    max_seqlen = (cu_seqlens[1:] - cu_seqlens[:-1]).max().item()
+    return max_seqlen
+
+
+def flatten_one_dim(tensor):
+    tensor = tensor.view(-1, tensor.size(-2), tensor.size(-1))
+    return tensor
+
+
+def prepare_for_flash_attention_varlen(query, key, value, cu_seqlens):
+    query = query.view(-1, query.size(-2), query.size(-1))
+    key = key.view(-1, key.size(-2), key.size(-1))
+    value = value.view(-1, value.size(-2), value.size(-1))
+    return query, key, value, get_max_seqlen(cu_seqlens)
+
+
+def flash_attention_forward(
+        query_states: torch.Tensor,
+        key_states: torch.Tensor,
+        value_states: torch.Tensor,
+        query_length: int,
+        is_causal: bool,
+        dropout: float = 0.0,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        seqlens: Optional[torch.LongTensor] = None,
+        softmax_scale: Optional[float] = None,
+        sliding_window: Optional[int] = None,
+        use_top_left_mask: bool = False,
+        softcap: Optional[float] = None,
+        deterministic: bool = None,
+):
+    """
+    Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
+    first unpad the input, then computes the attention scores and pad the final attention scores.
+
+    Args:
+        query_states (`torch.Tensor`):
+            Input query states to be passed to Flash Attention API
+        key_states (`torch.Tensor`):
+            Input key states to be passed to Flash Attention API
+        value_states (`torch.Tensor`):
+            Input value states to be passed to Flash Attention API
+        attention_mask (`torch.Tensor`):
+            The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
+            position of padding tokens and 1 for the position of non-padding tokens.
+        dropout (`float`):
+            Attention dropout
+        softmax_scale (`float`, *optional*):
+            The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
+        use_top_left_mask (`bool`, defaults to `False`):
+            flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference.
+        softcap (`float`, *optional*):
+            Softcap for the attention logits, used e.g. in gemma2.
+        deterministic (`bool`, *optional*):
+            Determines if the deterministic option introduced in flash_attn>=2.4.1 is enabled.
+    """
+    if not use_top_left_mask:
+        causal = is_causal
+    else:
+        # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. .
+        causal = is_causal and query_length != 1
+
+    # Assuming 4D tensors, key_states.shape[1] is the key/value sequence length (source length).
+    use_sliding_windows = (
+            _flash_supports_window_size and sliding_window is not None and key_states.shape[1] > sliding_window
+    )
+    flash_kwargs = {"window_size": (sliding_window - 1, 0)} if use_sliding_windows else {}
+
+    if is_flash_attn_greater_or_equal("2.4.1"):
+        if deterministic is None:
+            deterministic = os.environ.get("FLASH_ATTENTION_DETERMINISTIC", "0") == "1"
+        flash_kwargs["deterministic"] = deterministic
+
+    if softcap is not None:
+        flash_kwargs["softcap"] = softcap
+    # Contains at least one padding token in the sequence
+    if seqlens is not None:
+        batch_size = query_states.shape[0]
+        query_states, key_states, value_states, max_seqlen = prepare_for_flash_attention_varlen(query_states,
+                                                                                                key_states,
+                                                                                                value_states, seqlens)
+        attn_output = flash_attn_varlen_func(
+            query_states,
+            key_states,
+            value_states,
+            cu_seqlens_q=seqlens,
+            cu_seqlens_k=seqlens,
+            max_seqlen_q=max_seqlen,
+            max_seqlen_k=max_seqlen,
+            dropout_p=dropout,
+            softmax_scale=softmax_scale,
+            causal=causal,
+            **flash_kwargs,
+        )
+
+        attn_output = attn_output.reshape(batch_size, -1, attn_output.size(-2), attn_output.size(-1))
+
+    elif attention_mask is not None:
+        batch_size = query_states.shape[0]
+        query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = _upad_input(
+            query_states, key_states, value_states, attention_mask, query_length
+        )
+        cu_seqlens_q, cu_seqlens_k = cu_seq_lens
+        max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
+        attn_output_unpad = flash_attn_varlen_func(
+            query_states,
+            key_states,
+            value_states,
+            cu_seqlens_q=cu_seqlens_q,
+            cu_seqlens_k=cu_seqlens_k,
+            max_seqlen_q=max_seqlen_in_batch_q,
+            max_seqlen_k=max_seqlen_in_batch_k,
+            dropout_p=dropout,
+            softmax_scale=softmax_scale,
+            causal=causal,
+            **flash_kwargs,
+        )
+        attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
+
+    else:
+        attn_output = flash_attn_func(
+            query_states, key_states, value_states, dropout, softmax_scale=softmax_scale, causal=causal, **flash_kwargs
+        )
+
+    return attn_output
+
+
+def custom_convolution(U, K): 
+    """
+    U: Input matrix, shape (bs, seq, h, d)
+    K: Convolution kernel, shape (w, h)
+    Returns: Output matrix V, shape (bs, seq, h, d)
+    """
+    # h, w = K.shape
+    w = K.size(-1)
+    padding = (w - 1, 0)
+    U_padded = F.pad(U, (0, 0, 0, 0, *padding))  # Shape becomes (bs, seq+w-1, h, d)
+    U_unfolded = U_padded.unfold(1, w, 1)  # Shape becomes (bs, seq+w-1, h, d, w)
+    V_unfolded = U_unfolded * K  # Shape remains (bs, seq, h, d, w)
+    V = V_unfolded.sum(dim=-1)  # Shape becomes (bs, seq, h, d)
+    return V
+
+
+def custom_convolution_with_splits(U, K, cu_seqlens):
+    """
+    U: Input matrix, shape (bs, seq, h, d)
+    K: Convolution kernel, shape (w, h)
+    cu_seqlens: Cumulative sequence lengths, indicating how to split the input.
+    Returns: Output matrix, shape (bs, seq, h, d)
+    """
+    ori_shape = U.size()  # Save the original shape of U
+    # Flatten U to handle variable-length sequences
+    U_flatten = U.reshape(1, -1, ori_shape[-2], ori_shape[-1])  # Shape: (1, total_seq, h, d)
+
+    # Perform convolution on each subsequence separately
+    V_parts = []  # Store the results of each subsequence
+    start = 0  # Start index of the current subsequence
+    for end in cu_seqlens[1:]:
+        end = end.item()  # Convert scalar tensor to int
+        U_part = U_flatten[:, start:end, :, :]  # Slice the subsequence (1, seq_sub, h, d)
+        V_part = custom_convolution(U_part, K)  # Apply custom convolution
+        V_parts.append(V_part)  # Append the result
+        start = end  # Update the start index for the next subsequence
+
+    # Concatenate the results along the sequence dimension
+    V = torch.cat(V_parts, dim=1).to(U)  # Shape: (1, total_seq, h, d)
+
+    # Reshape the output to match the original input shape
+    return V.reshape(ori_shape)
+
+
+class BaichuanFlashAttention2(BaichuanAttention):
+    """
+    Baichuan flash attention module, following Baichuan attention module. This module inherits from `BaichuanAttention`
+    as the weights of the module stays untouched. The only required change would be on the forward pass
+    where it needs to correctly call the public API of flash attention and deal with padding tokens
+    in case the input contains any of them. 
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
+        # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
+        # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
+        self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            seqlens: Optional[torch.LongTensor] = None,
+            past_key_value: Optional[CustomCache] = None,
+            output_attentions: bool = False,
+            use_cache: bool = False,
+            cache_position: Optional[torch.LongTensor] = None,
+            position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,  # will become mandatory in v4.46
+    ):
+
+        bsz, q_len, _ = hidden_states.size()
+        proj = self.W_pack(hidden_states)
+        proj = rearrange(proj, 'bs seq_len (n_head head_dim) -> n_head bs seq_len head_dim', head_dim=self.head_dim)
+        query_states = rearrange(proj[:self.num_heads], 'n_head bs seq_len head_dim -> bs seq_len n_head head_dim')
+        key_states = rearrange(proj[self.num_heads:self.num_heads + self.num_key_value_heads],
+                               'n_head bs seq_len head_dim -> bs seq_len n_head head_dim')
+        value_states = rearrange(proj[self.num_heads + self.num_key_value_heads:],
+                                 'n_head bs seq_len head_dim -> bs seq_len n_head head_dim')
+
+
+        if past_key_value is None or past_key_value.get_seq_length(self.layer_idx) == 0:# prefill
+            if not self.training:
+                self.last_k = key_states[:, -1:]
+                self.last_v = value_states[:, -1:]
+            if seqlens is None:
+                key_states = custom_convolution(key_states, self.conv_k)
+                value_states = custom_convolution(value_states, self.conv_v)
+            else:
+                assert seqlens.ndim==1
+                key_states=custom_convolution_with_splits(key_states,self.conv_k,seqlens)
+                value_states=custom_convolution_with_splits(value_states,self.conv_v,seqlens)
+        else: # decode
+            self.last_k, key_states = key_states, self.conv_k[0, 0, :, 0, :1] * self.last_k + self.conv_k[0, 0, :, 0, 1:] * key_states
+            self.last_v, value_states = value_states, self.conv_v[0, 0, :, 0, :1] * self.last_v + self.conv_v[0, 0, :, 0, 1:] * value_states
+        if seqlens is not None:
+            max_seqlen = get_max_seqlen(seqlens)
+        else:
+            max_seqlen = None
+
+        past_len = past_key_value.get_past_len(self.layer_idx) if past_key_value is not None else 0
+        query_states, key_states = self.rotary_emb(
+            query_states,
+            key_states,
+            seqlen_offset=past_len,
+            cu_seqlens=seqlens,
+            max_seqlen=max_seqlen
+        )
+
+        if past_key_value is not None:
+            cache_has_contents = past_key_value.get_seq_length(self.layer_idx) > 0
+            kv_seq_len = key_states.shape[1] + past_key_value.get_seq_length(self.layer_idx)
+            if (
+                    self.is_swa
+                    and kv_seq_len > self.config.sliding_window
+                    and cache_has_contents
+            ):
+                slicing_tokens = 1 - self.config.sliding_window
+                past_key = past_key_value[self.layer_idx][0]
+                past_value = past_key_value[self.layer_idx][1]
+
+                past_key_value.key_cache[self.layer_idx] = past_key[:, slicing_tokens:, :, :].contiguous()
+                past_key_value.value_cache[self.layer_idx] = past_value[:, slicing_tokens:, :, :].contiguous()
+
+                if past_key_value[self.layer_idx][0].shape[1] != self.config.sliding_window - 1:
+                    raise ValueError(
+                        f"past key must have a shape of (`batch_size, num_heads, self.config.sliding_window-1, head_dim`), got"
+                        f" {past_key.shape}"
+                    )
+
+                # if attention_mask is not None:
+                #     # TODO: not check!!
+                #     attention_mask = attention_mask[:, slicing_tokens:]
+                #     attention_mask = torch.cat([attention_mask, torch.ones_like(attention_mask[:, -1:])], dim=-1)
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx)
+
+        input_dtype = query_states.dtype
+        if input_dtype == torch.float32:
+            if torch.is_autocast_enabled():
+                target_dtype = torch.get_autocast_gpu_dtype()
+            # Handle the case where the model is quantized
+            elif hasattr(self.config, "_pre_quantization_dtype"):
+                target_dtype = self.config._pre_quantization_dtype
+            else:
+                target_dtype = self.q_proj.weight.dtype
+
+            logger.warning_once(
+                f"The input hidden states seems to be silently casted in float32, this might be related to"
+                f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in"
+                f" {target_dtype}."
+            )
+
+            query_states = query_states.to(target_dtype)
+            key_states = key_states.to(target_dtype)
+            value_states = value_states.to(target_dtype)
+
+        if self.is_swa:
+            sliding_window = self.config.sliding_window
+        else:
+            sliding_window = None
+        attn_output = flash_attention_forward(
+            query_states,
+            key_states,
+            value_states,
+            query_length=q_len,
+            position_ids=position_ids,
+            seqlens=seqlens,
+            sliding_window=sliding_window,
+            is_causal=self.is_causal,
+            use_top_left_mask=self._flash_attn_uses_top_left_mask,
+        )
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size).contiguous()
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+
+Baichuan_ATTENTION_CLASSES = {
+    "eager": BaichuanAttention,
+    "flash_attention_2": BaichuanFlashAttention2,
+}
+
+
+class BaichuanDecoderLayer(nn.Module):
+    def __init__(self, config: BaichuanM1Config, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.layer_idx = layer_idx
+        self.self_attn = Baichuan_ATTENTION_CLASSES['flash_attention_2'](config, layer_idx)
+
+        self.mlp = BaichuanMLP(config)
+        self.input_layernorm = BaichuanRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = BaichuanRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            seqlens: Optional[torch.LongTensor] = None,
+            past_key_value: Optional[Tuple[torch.Tensor]] = None,
+            output_attentions: Optional[bool] = False,
+            use_cache: Optional[bool] = False,
+            cache_position: Optional[torch.LongTensor] = None,
+            position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,  
+            **kwargs,
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
+                `(batch, sequence_length)` where padding elements are indicated by 0.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+            cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*):
+                Indices depicting the position of the input sequence tokens in the sequence.
+            position_embeddings (`Tuple[torch.FloatTensor, torch.FloatTensor]`, *optional*):
+                Tuple containing the cosine and sine positional embeddings of shape `(batch_size, seq_len, head_dim)`,
+                with `head_dim` being the embedding dimension of each attention head.
+            kwargs (`dict`, *optional*):
+                Arbitrary kwargs to be ignored, used for FSDP and other methods that injects code
+                into the model
+        """
+
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            seqlens=seqlens,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            position_embeddings=position_embeddings,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+        return outputs
+
+
+Baichuan_START_DOCSTRING = r"""
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`BaichuanM1Config`]):
+            Model configuration class with all the parameters of the model. Initializing with a config file does not
+            load the weights associated with the model, only the configuration. Check out the
+            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+
+@add_start_docstrings(
+    "The bare Bai chuan Model outputting raw hidden-states without any specific head on top.",
+    Baichuan_START_DOCSTRING,
+)
+class BaichuanPreTrainedModel(PreTrainedModel):
+    config_class = BaichuanM1Config
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["BaichuanDecoderLayer"]
+    _skip_keys_device_placement = "past_key_values"
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
+    _supports_cache_class = True
+    _supports_quantized_cache = True
+    _supports_static_cache = True
+
+    def _init_weights(self, module):
+        std = self.config.initializer_range
+        if isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+
+Baichuan_INPUTS_DOCSTRING = r"""
+
+"""
+
+
+@add_start_docstrings(
+    "The bare Baichuan Model outputting raw hidden-states without any specific head on top.",
+    Baichuan_START_DOCSTRING,
+)
+class BaichuanModel(BaichuanPreTrainedModel):
+    """
+    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`BaichuanDecoderLayer`]
+
+    Args:
+        config: BaichuanM1Config
+    """
+
+    def __init__(self, config: BaichuanM1Config):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        self.layers = nn.ModuleList(
+            [BaichuanDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self._attn_implementation = config._attn_implementation
+        self.norm = BaichuanRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+        self.gradient_checkpointing = True
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    @add_start_docstrings_to_model_forward(Baichuan_INPUTS_DOCSTRING)
+    def forward(
+            self,
+            input_ids: torch.LongTensor = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            seqlens: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[CustomCache] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+            cache_position: Optional[torch.LongTensor] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        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
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError(
+                "You cannot specify both input_ids and inputs_embeds at the same time, and must specify either one"
+            )
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        if seqlens is not None:
+            assert seqlens.ndim == 2
+                # batch multi-pack 样本拉平
+            cu_seqlens = []
+            offset, seqlen = 0, seqlens.size(1)
+            for lens in seqlens:
+                cu_seqlens.append(offset)
+                cu_seqlens.extend((lens[(lens > 0) & (lens < seqlen)] + offset).tolist())
+                offset += seqlen
+            cu_seqlens.append(offset)
+            seqlens = torch.tensor(cu_seqlens, dtype=torch.int32, device=input_ids.device)
+            # unset attention_mask to save memory
+            attention_mask = None
+        # kept for BC (non `Cache` `past_key_values` inputs)
+        return_legacy_cache = False
+        if use_cache and not isinstance(past_key_values, CustomCache):
+            return_legacy_cache = False
+            if past_key_values is None:
+                past_key_values = CustomCache()
+            else:
+                past_key_values = CustomCache.from_legacy_cache(past_key_values)
+                logger.warning_once(
+                    "We detected that you are passing `past_key_values` as a tuple of tuples. This is deprecated and "
+                    "will be removed in v4.47. Please convert your cache or use an appropriate `Cache` class "
+                    "(https://huggingface.co/docs/transformers/kv_cache#legacy-cache-format)"
+                )
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if cache_position is None:
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position = torch.arange(
+                past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device
+            )
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        causal_mask = self._update_causal_mask(
+            attention_mask, inputs_embeds, cache_position, past_key_values, output_attentions
+        )
+
+        hidden_states = inputs_embeds
+
+        # create position embeddings to be shared across the decoder layers
+        # position_embeddings = self.rotary_emb(hidden_states, position_ids)
+        position_embeddings = None
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        next_decoder_cache = None
+
+        for decoder_layer in self.layers:
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    decoder_layer,
+                    hidden_states,
+                    causal_mask,
+                    position_ids,
+                    seqlens,
+                    past_key_values,
+                    output_attentions,
+                    use_cache,
+                    cache_position,
+                    position_embeddings,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=causal_mask,
+                    position_ids=position_ids,
+                    seqlens=seqlens,
+                    past_key_value=past_key_values,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                    cache_position=cache_position,
+                    position_embeddings=position_embeddings,
+                )
+
+            hidden_states = layer_outputs[0]
+            if use_cache:
+                next_decoder_cache = layer_outputs[2 if output_attentions else 1]
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        next_cache = next_decoder_cache if use_cache else None
+        if return_legacy_cache:
+            next_cache = next_cache.to_legacy_cache()
+        if not return_dict:
+            return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+
+
+    def _update_causal_mask(
+            self,
+            attention_mask: torch.Tensor,
+            input_tensor: torch.Tensor,
+            cache_position: torch.Tensor,
+            past_key_values: CustomCache,
+            output_attentions: bool,
+    ):
+        if self.config._attn_implementation == "flash_attention_2":
+            if attention_mask is not None and 0.0 in attention_mask:
+                return attention_mask
+            return None
+
+        # For SDPA, when possible, we will rely on its `is_causal` argument instead of its `attn_mask` argument, in
+        # order to dispatch on Flash Attention 2. This feature is not compatible with static cache, as SDPA will fail
+        # to infer the attention mask.
+        past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+        using_static_cache = isinstance(past_key_values, StaticCache)
+
+        # When output attentions is True, sdpa implementation's forward method calls the eager implementation's forward
+        if self.config._attn_implementation == "sdpa" and not using_static_cache and not output_attentions:
+            if AttentionMaskConverter._ignore_causal_mask_sdpa(
+                    attention_mask,
+                    inputs_embeds=input_tensor,
+                    past_key_values_length=past_seen_tokens,
+                    is_training=self.training,
+            ):
+                return None
+
+        dtype, device = input_tensor.dtype, input_tensor.device
+        min_dtype = torch.finfo(dtype).min
+        sequence_length = input_tensor.shape[1]
+        if using_static_cache:
+            target_length = past_key_values.get_max_length()
+        else:
+            target_length = (
+                attention_mask.shape[-1]
+                if isinstance(attention_mask, torch.Tensor)
+                else past_seen_tokens + sequence_length + 1
+            )
+
+        # In case the provided `attention` mask is 2D, we generate a causal mask here (4D).
+        causal_mask = _prepare_4d_causal_attention_mask_with_cache_position(
+            attention_mask,
+            sequence_length=sequence_length,
+            target_length=target_length,
+            dtype=dtype,
+            device=device,
+            min_dtype=min_dtype,
+            cache_position=cache_position,
+            batch_size=input_tensor.shape[0],
+        )
+
+        if (
+                self.config._attn_implementation == "sdpa"
+                and attention_mask is not None
+                and attention_mask.device.type == "cuda"
+                and not output_attentions
+        ):
+            # Attend to all tokens in fully masked rows in the causal_mask, for example the relevant first rows when
+            # using left padding. This is required by F.scaled_dot_product_attention memory-efficient attention path.
+            # Details: https://github.com/pytorch/pytorch/issues/110213
+            causal_mask = AttentionMaskConverter._unmask_unattended(causal_mask, min_dtype)
+
+        return causal_mask
+
+
+class NormHead(nn.Module):
+    def __init__(self, hidden_size, vocab_size, bias=False):
+        super().__init__()
+        self.weight = nn.Parameter(torch.empty((vocab_size, hidden_size)))
+        nn.init.kaiming_uniform_(self.weight, a=math.sqrt(5))
+
+    def forward(self, hidden_states):
+        norm_weight = nn.functional.normalize(self.weight)
+        return nn.functional.linear(hidden_states, norm_weight)
+
+
+class BaichuanM1ForCausalLM(BaichuanPreTrainedModel, GenerationMixin):
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = BaichuanModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = NormHead(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = 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
+
+    @add_start_docstrings_to_model_forward(Baichuan_INPUTS_DOCSTRING)
+    def forward(
+            self,
+            input_ids: torch.LongTensor = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            seqlens: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[List[torch.FloatTensor]] = None,
+            inputs_embeds: Optional[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,
+            cache_position: Optional[torch.LongTensor] = None,
+            num_logits_to_keep: int = 0,
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        r"""
+        Args:
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+            num_logits_to_keep (`int`, *optional*):
+                Calculate logits for the last `num_logits_to_keep` tokens. If `0`, calculate logits for all
+                `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+                token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, BaichuanForCausalLM
+
+        >>> model = BaichuanForCausalLM.from_pretrained(PATH_TO_CONVERTED_WEIGHTS)
+        >>> tokenizer = AutoTokenizer.from_pretrained(PATH_TO_CONVERTED_TOKENIZER)
+
+        >>> prompt = "Hey, are you conscious? Can you talk to me?"
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+        ```"""
+        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
+        if input_ids is not None:
+            input_ids[input_ids == self.config.vocab_size] = 0
+        if labels is not None:
+            labels[labels == self.config.vocab_size] = 0
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            seqlens=seqlens,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            cache_position=cache_position,
+        )
+
+        hidden_states = outputs[0]
+        if labels is None and not is_torchdynamo_compiling():
+            logger.warning_once(
+                "Starting from v4.46, the `logits` model output will have the same type as the model (except at train time, where it will always be FP32)"
+            )
+        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+        # TODO: remove the float() operation in v4.46
+        logits = self.lm_head(hidden_states[:, -num_logits_to_keep:, :])
+
+        loss = None
+        if labels is not None:
+            # Upcast to float if we need to compute the loss to avoid potential precision issues
+            # logits = logits.float()
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            #shift_logits = logits
+            #shift_labels = labels
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+        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,
+        )
+
+    def prepare_inputs_for_generation(
+            self,
+            input_ids,
+            past_key_values=None,
+            attention_mask=None,
+            inputs_embeds=None,
+            cache_position=None,
+            position_ids=None,
+            use_cache=True,
+            num_logits_to_keep=None,
+            **kwargs,
+    ):
+        # If we have cache: let's slice `input_ids` through `cache_position`, to keep only the unprocessed tokens
+        # Exception 1: when passing input_embeds, input_ids may be missing entries
+        # Exception 2: some generation methods do special slicing of input_ids, so we don't need to do it here
+        if past_key_values is not None:
+            if inputs_embeds is not None:  # Exception 1
+                input_ids = input_ids[:, -cache_position.shape[0]:]
+            elif input_ids.shape[1] != cache_position.shape[0]:  # Default case (the "else", a no op, is Exception 2)
+                input_ids = input_ids[:, cache_position]
+
+        if attention_mask is not None and position_ids is None:
+            # create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if past_key_values:
+                position_ids = position_ids[:, -input_ids.shape[1]:]
+
+                # This `clone` call is needed to avoid recapturing cuda graphs with `torch.compile`'s  `mode="reduce-overhead`, as otherwise the input `position_ids` would have various stride during the decoding. Here, simply using `.contiguous()` is not sufficient as in the batch size = 1 case, `position_ids` is already contiguous but with varying stride which retriggers a capture.
+                position_ids = position_ids.clone(memory_format=torch.contiguous_format)
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and cache_position[0] == 0:
+            model_inputs = {"inputs_embeds": inputs_embeds, "input_ids": None}
+        else:
+            # The clone here is for the same reason as for `position_ids`.
+            model_inputs = {"input_ids": input_ids.clone(memory_format=torch.contiguous_format), "inputs_embeds": None}
+
+        if isinstance(past_key_values, StaticCache) and attention_mask.ndim == 2:
+            if model_inputs["inputs_embeds"] is not None:
+                batch_size, sequence_length, _ = model_inputs["inputs_embeds"].shape
+                device = model_inputs["inputs_embeds"].device
+            else:
+                batch_size, sequence_length = model_inputs["input_ids"].shape
+                device = model_inputs["input_ids"].device
+
+            dtype = self.lm_head.weight.dtype
+            min_dtype = torch.finfo(dtype).min
+
+            attention_mask = _prepare_4d_causal_attention_mask_with_cache_position(
+                attention_mask,
+                sequence_length=sequence_length,
+                target_length=past_key_values.get_max_length(),
+                dtype=dtype,
+                device=device,
+                min_dtype=min_dtype,
+                cache_position=cache_position,
+                batch_size=batch_size,
+            )
+
+        if num_logits_to_keep is not None:
+            model_inputs["num_logits_to_keep"] = num_logits_to_keep
+
+        model_inputs.update(
+            {
+                "position_ids": position_ids,
+                "cache_position": cache_position,
+                "past_key_values": past_key_values,
+                "use_cache": use_cache,
+                "attention_mask": attention_mask,
+            }
+        )
+        return model_inputs
+
+    @torch.no_grad()
+    def chat(self, tokenizer, messages: List[dict], stream=False,
+             generation_config: Optional[GenerationConfig] = None):
+        generation_config = generation_config or self.generation_config
+        input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True)
+        input_ids = torch.LongTensor([input_ids]).to(self.device)
+        if stream:
+            streamer = TextIterStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
+            Thread(target=self.generate, kwargs=dict(
+                inputs=input_ids, streamer=streamer,
+                generation_config=generation_config,
+            )).start()
+            return streamer
+        else:
+            outputs = self.generate(input_ids, generation_config=generation_config)
+            response = tokenizer.decode(outputs[0][len(input_ids[0]):], skip_special_tokens=True)
+            return response

special_tokens_map.json

@@ -0,0 +1,46 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>",
+    "<B_SYS>",
+    "<B_USYS>",
+    "<C_Q>",
+    "<C_A>",
+    "<|im_sep|>",
+    "<|tool_call|>",
+    "<|arguments|>"
+  ],
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<pad>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenization_baichuan.py

@@ -0,0 +1,231 @@
+import os
+from shutil import copyfile
+from typing import Any, Dict, List, Optional, Tuple
+
+import sentencepiece as spm
+
+from transformers.tokenization_utils import AddedToken, PreTrainedTokenizer
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+VOCAB_FILES_NAMES = {"vocab_file": "tokenizer.model"}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    "vocab_file": {},
+    "tokenizer_file": {},
+}
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {}
+
+
+class BaichuanTokenizer(PreTrainedTokenizer):
+    """
+    Construct a Baichuan tokenizer. Based on byte-level Byte-Pair-Encoding.
+
+    Args:
+        vocab_file (`str`):
+            Path to the vocabulary file.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    model_input_names = ["input_ids", "attention_mask"]
+
+    def __init__(
+        self,
+        vocab_file,
+        unk_token="<unk>",
+        bos_token="<s>",
+        eos_token="</s>",
+        pad_token=None,
+        sp_model_kwargs: Optional[Dict[str, Any]] = None,
+        add_bos_token=True,
+        add_eos_token=False,
+        clean_up_tokenization_spaces=False,
+        **kwargs,
+    ):
+        self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs
+        bos_token = AddedToken(bos_token, lstrip=False, rstrip=False) if isinstance(bos_token, str) else bos_token
+        eos_token = AddedToken(eos_token, lstrip=False, rstrip=False) if isinstance(eos_token, str) else eos_token
+        unk_token = AddedToken(unk_token, lstrip=False, rstrip=False) if isinstance(unk_token, str) else unk_token
+        pad_token = AddedToken(pad_token, lstrip=False, rstrip=False) if isinstance(pad_token, str) else pad_token
+        self.vocab_file = vocab_file
+        self.add_bos_token = add_bos_token
+        self.add_eos_token = add_eos_token
+        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
+        self.sp_model.Load(vocab_file)
+        super().__init__(
+            bos_token=bos_token,
+            eos_token=eos_token,
+            unk_token=unk_token,
+            pad_token=pad_token,
+            add_bos_token=add_bos_token,
+            add_eos_token=add_eos_token,
+            sp_model_kwargs=self.sp_model_kwargs,
+            clean_up_tokenization_spaces=clean_up_tokenization_spaces,
+            **kwargs,
+        )
+        self.pad_token_id = self._convert_token_to_id(self.pad_token)
+
+    def __getstate__(self):
+        state = self.__dict__.copy()
+        state["sp_model"] = None
+        return state
+
+    def __setstate__(self, d):
+        self.__dict__ = d
+        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
+        self.sp_model.Load(self.vocab_file)
+
+    @property
+    def vocab_size(self):
+        """Returns vocab size"""
+        return self.sp_model.get_piece_size()
+
+    def get_vocab(self):
+        """Returns vocab as a dict"""
+        vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}
+        vocab.update(self.added_tokens_encoder)
+        return vocab
+
+    def _tokenize(self, text):
+        """Returns a tokenized string."""
+        return self.sp_model.encode(text, out_type=str)
+
+    def _convert_token_to_id(self, token):
+        """Converts a token (str) in an id using the vocab."""
+        return self.sp_model.piece_to_id(token)
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (str) using the vocab."""
+        token = self.sp_model.IdToPiece(index)
+        return token
+
+    def convert_tokens_to_string(self, tokens):
+        """Converts a sequence of tokens (string) in a single string."""
+        current_sub_tokens = []
+        out_string = ""
+        prev_is_special = False
+        for i, token in enumerate(tokens):
+            # make sure that special tokens are not decoded using sentencepiece model
+            if token in self.all_special_tokens:
+                if not prev_is_special and i != 0:
+                    out_string += " "
+                out_string += self.sp_model.decode(current_sub_tokens) + token
+                prev_is_special = True
+                current_sub_tokens = []
+            else:
+                current_sub_tokens.append(token)
+                prev_is_special = False
+        out_string += self.sp_model.decode(current_sub_tokens)
+        return out_string
+
+    def save_vocabulary(self, save_directory, filename_prefix: Optional[str] = None) -> Tuple[str]:
+        """
+        Save the vocabulary and special tokens file to a directory.
+
+        Args:
+            save_directory (`str`):
+                The directory in which to save the vocabulary.
+
+        Returns:
+            `Tuple(str)`: Paths to the files saved.
+        """
+        if not os.path.isdir(save_directory):
+            logger.error(f"Vocabulary path ({save_directory}) should be a directory")
+            return
+        out_vocab_file = os.path.join(
+            save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
+        )
+
+        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
+            copyfile(self.vocab_file, out_vocab_file)
+        elif not os.path.isfile(self.vocab_file):
+            with open(out_vocab_file, "wb") as fi:
+                content_spiece_model = self.sp_model.serialized_model_proto()
+                fi.write(content_spiece_model)
+
+        return (out_vocab_file,)
+
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
+        bos_token_id = [self.bos_token_id] if self.add_bos_token else []
+        eos_token_id = [self.eos_token_id] if self.add_eos_token else []
+
+        output = bos_token_id + token_ids_0 + eos_token_id
+
+        if token_ids_1 is not None:
+            output = output + bos_token_id + token_ids_1 + eos_token_id
+
+        return output
+
+    def get_special_tokens_mask(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False
+    ) -> List[int]:
+        """
+        Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
+        special tokens using the tokenizer `prepare_for_model` method.
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of IDs.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+            already_has_special_tokens (`bool`, *optional*, defaults to `False`):
+                Whether or not the token list is already formatted with special tokens for the model.
+
+        Returns:
+            `List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
+        """
+        if already_has_special_tokens:
+            return super().get_special_tokens_mask(
+                token_ids_0=token_ids_0, token_ids_1=token_ids_1, already_has_special_tokens=True
+            )
+
+        bos_token_id = [1] if self.add_bos_token else []
+        eos_token_id = [1] if self.add_eos_token else []
+
+        if token_ids_1 is None:
+            return bos_token_id + ([0] * len(token_ids_0)) + eos_token_id
+        return (
+            bos_token_id
+            + ([0] * len(token_ids_0))
+            + eos_token_id
+            + bos_token_id
+            + ([0] * len(token_ids_1))
+            + eos_token_id
+        )
+
+    def create_token_type_ids_from_sequences(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Creates a mask from the two sequences passed to be used in a sequence-pair classification task. An ALBERT
+        sequence pair mask has the following format:
+
+        ```
+        0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1
+        | first sequence    | second sequence |
+        ```
+
+        if token_ids_1 is None, only returns the first portion of the mask (0s).
+
+        Args:
+            token_ids_0 (`List[int]`):
+                List of ids.
+            token_ids_1 (`List[int]`, *optional*):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            `List[int]`: List of [token type IDs](../glossary#token-type-ids) according to the given sequence(s).
+        """
+        bos_token_id = [self.bos_token_id] if self.add_bos_token else []
+        eos_token_id = [self.eos_token_id] if self.add_eos_token else []
+
+        output = [0] * len(bos_token_id + token_ids_0 + eos_token_id)
+
+        if token_ids_1 is not None:
+            output += [1] * len(bos_token_id + token_ids_1 + eos_token_id)
+
+        return output

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f5af87706706ff930034b468c7f315c7da31de5f35d5b71a6458329ef5d9034
+size 2224601

tokenizer_config.json

@@ -0,0 +1,389 @@
+{
+  "add_bos_token": false,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<pad>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "50": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "51": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "52": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "53": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "54": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "55": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "56": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "57": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "58": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "59": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "60": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "61": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "62": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "63": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "64": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "65": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "66": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "67": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "68": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "69": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "70": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "71": {
+      "content": "<B_SYS>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "72": {
+      "content": "<B_USYS>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "73": {
+      "content": "<C_Q>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "74": {
+      "content": "<C_A>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "75": {
+      "content": "<B_FUNC>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "76": {
+      "content": "<B_CODE>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "77": {
+      "content": "<B_APE>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "78": {
+      "content": "<function_calling>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "79": {
+      "content": "<calc_start>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "80": {
+      "content": "<calc_end>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "81": {
+      "content": "<inner_think>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "82": {
+      "content": "<|im_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "83": {
+      "content": "<|tool_call|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "84": {
+      "content": "<|arguments|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "85": {
+      "content": "<|o1_step|>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "86": {
+      "content": "<|o1_answer|>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "87": {
+      "content": "<tree_node>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "88": {
+      "content": "</tree_node>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>",
+    "<B_SYS>",
+    "<B_USYS>",
+    "<C_Q>",
+    "<C_A>",
+    "<|im_sep|>",
+    "<|tool_call|>",
+    "<|arguments|>"
+  ],
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_baichuan.BaichuanTokenizer",
+      null
+    ]
+  },
+  "bos_token": "<s>",
+  "chat_template": "{% for message in messages %}{% if message['role'] == 'system' %}{{'<B_SYS>' + message['content']}}{% elif message['role'] == 'user_system' %}{{'<B_USYS>' + message['content']}}{% elif message['role'] == 'user' %}{{'<C_Q>' + message['content']}}{% elif message['role'] == 'assistant' %}{{'<C_A>' + message['content']}}{% elif message['role'] == 'function' %}{{'<B_FUNC>' + message['content']}}{% elif message['role'] == 'code' %}{{'<B_CODE>' + message['content']}}{% else %}{{ raise_exception('Invalid message role: ' + message['role']) }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{'<C_A>'}}{% endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "extra_special_tokens": {},
+  "model_max_length": 32768,
+  "pad_token": "<pad>",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "BaichuanTokenizer",
+  "unk_token": "<unk>",
+  "use_fast": false
+}