Hugging Face's logo

Magpie-Align
/
MagpieLM-8B-SFT-v0.1

Text Generation
Transformers
PyTorch
Safetensors
llama
axolotl
Generated from Trainer
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
1
MagpieLM-8B-SFT-v0.1 / README.md
flydust's picture
flydust
End of training
de82fb5 verified
preview code
|
raw
history blame
3.81 kB
metadata
library_name: transformers
license: llama3.1
base_model: meta-llama/Meta-Llama-3.1-8B
tags:
  - axolotl
  - generated_from_trainer
model-index:
  - name: Llama-3.1-8B-Magpie-SFT-GMix-550K
    results: []

Built with Axolotl

See axolotl config

axolotl version: 0.4.1


base_model: meta-llama/Meta-Llama-3.1-8B
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer
chat_template: llama3

load_in_8bit: false
load_in_4bit: false
strict: false
main_process_port: 0

datasets:
  - path: flydust/Magpie-Llama-3-70B-300K-Gemma2-9B
    type: sharegpt
    conversation: llama3
  - path: flydust/Magpie-Reasoning-150K-Gemma2-9B
    type: sharegpt
    conversation: llama3
  - path: flydust/Magpie-100k-Gemma2-9B
    type: sharegpt
    conversation: llama3
dataset_prepared_path: /data/zhangchen_xu/last_run_prepared
val_set_size: 0.001
output_dir: /data/zhangchen_xu/axolotl_out/Llama-3.1-8B-SFT-GMix-550K

sequence_len: 8192
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

wandb_project: SynDa
wandb_entity:
wandb_watch:
wandb_name: Llama-3.1-8B-Mix-SFT-GMix-550K
wandb_log_model:
hub_model_id: Magpie-Align/Llama-3.1-8B-Magpie-SFT-GMix-550K

gradient_accumulation_steps: 32
micro_batch_size: 1
num_epochs: 2
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 2e-5

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
resume_from_checkpoint:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_ratio: 0.1
evals_per_epoch: 5
eval_table_size:
saves_per_epoch: 
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  pad_token: <|end_of_text|>

Llama-3.1-8B-Magpie-SFT-GMix-550K

This model is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.4544

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 2e-05
  • train_batch_size: 1
  • eval_batch_size: 1
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 4
  • gradient_accumulation_steps: 32
  • total_train_batch_size: 128
  • total_eval_batch_size: 4
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 51
  • num_epochs: 2

Training results

Training Loss Epoch Step Validation Loss
0.9311 0.0038 1 0.9847
0.561 0.2015 53 0.5765
0.4843 0.4030 106 0.5039
0.4608 0.6045 159 0.4814
0.4454 0.8060 212 0.4678
0.4403 1.0075 265 0.4596
0.3965 1.1938 318 0.4574
0.3952 1.3953 371 0.4554
0.3962 1.5968 424 0.4547
0.3948 1.7983 477 0.4544

Framework versions

  • Transformers 4.45.0.dev0
  • Pytorch 2.3.0+cu121
  • Datasets 2.19.1
  • Tokenizers 0.19.1