/LLaMA-Factory

Easy-to-use LLM fine-tuning framework (LLaMA, BLOOM, Mistral, Baichuan, Qwen, ChatGLM)

Primary LanguagePythonApache License 2.0Apache-2.0

# LLaMA Factory

GitHub Repo stars GitHub Code License GitHub last commit PyPI Downloads GitHub pull request Discord Spaces Studios

πŸ‘‹ Join our WeChat.

[ English | δΈ­ζ–‡ ]

LLaMA Board: A One-stop Web UI for Getting Started with LLaMA Factory

Preview LLaMA Board at πŸ€— Spaces or ModelScope.

Launch LLaMA Board via CUDA_VISIBLE_DEVICES=0 python src/train_web.py. (multiple GPUs are not supported yet in this mode)

Here is an example of altering the self-cognition of an instruction-tuned language model within 10 minutes on a single GPU.

tutorial.mp4

Table of Contents

Benchmark

Compared to ChatGLM's P-Tuning, LLaMA-Factory's LoRA tuning offers up to 3.7 times faster training speed with a better Rouge score on the advertising text generation task. By leveraging 4-bit quantization technique, LLaMA-Factory's QLoRA further improves the efficiency regarding the GPU memory.

benchmark

Definitions
  • Training Speed: the number of training samples processed per second during the training. (bs=4, cutoff_len=1024)
  • Rouge Score: Rouge-2 score on the development set of the advertising text generation task. (bs=4, cutoff_len=1024)
  • GPU Memory: Peak GPU memory usage in 4-bit quantized training. (bs=1, cutoff_len=1024)
  • We adopt pre_seq_len=128 for ChatGLM's P-Tuning and lora_rank=32 for LLaMA-Factory's LoRA tuning.

Changelog

[23/12/23] We supported unsloth's implementation to boost LoRA tuning for the LLaMA, Mistral and Yi models. Try --use_unsloth argument to activate unsloth patch. It achieves 1.7x speed in our benchmark, check this page for details.

[23/12/12] We supported fine-tuning the latest MoE model Mixtral 8x7B in our framework. See hardware requirement here.

[23/12/01] We supported downloading pre-trained models and datasets from the ModelScope Hub for Chinese mainland users. See this tutorial for usage.

Full Changelog

[23/10/21] We supported NEFTune trick for fine-tuning. Try --neftune_noise_alpha argument to activate NEFTune, e.g., --neftune_noise_alpha 5.

[23/09/27] We supported $S^2$-Attn proposed by LongLoRA for the LLaMA models. Try --shift_attn argument to enable shift short attention.

[23/09/23] We integrated MMLU, C-Eval and CMMLU benchmarks in this repo. See this example to evaluate your models.

[23/09/10] We supported FlashAttention-2. Try --flash_attn argument to enable FlashAttention-2 if you are using RTX4090, A100 or H100 GPUs.

[23/08/12] We supported RoPE scaling to extend the context length of the LLaMA models. Try --rope_scaling linear argument in training and --rope_scaling dynamic argument at inference to extrapolate the position embeddings.

[23/08/11] We supported DPO training for instruction-tuned models. See this example to train your models.

[23/07/31] We supported dataset streaming. Try --streaming and --max_steps 10000 arguments to load your dataset in streaming mode.

[23/07/29] We released two instruction-tuned 13B models at Hugging Face. See these Hugging Face Repos (LLaMA-2 / Baichuan) for details.

[23/07/18] We developed an all-in-one Web UI for training, evaluation and inference. Try train_web.py to fine-tune models in your Web browser. Thank @KanadeSiina and @codemayq for their efforts in the development.

[23/07/09] We released FastEdit ⚑🩹, an easy-to-use package for editing the factual knowledge of large language models efficiently. Please follow FastEdit if you are interested.

[23/06/29] We provided a reproducible example of training a chat model using instruction-following datasets, see Baichuan-7B-sft for details.

[23/06/22] We aligned the demo API with the OpenAI's format where you can insert the fine-tuned model in arbitrary ChatGPT-based applications.

[23/06/03] We supported quantized training and inference (aka QLoRA). Try --quantization_bit 4/8 argument to work with quantized models.

Supported Models

Model Model size Default module Template
Baichuan 7B/13B W_pack baichuan
Baichuan2 7B/13B W_pack baichuan2
BLOOM 560M/1.1B/1.7B/3B/7.1B/176B query_key_value -
BLOOMZ 560M/1.1B/1.7B/3B/7.1B/176B query_key_value -
ChatGLM3 6B query_key_value chatglm3
Falcon 7B/40B/180B query_key_value falcon
InternLM 7B/20B q_proj,v_proj intern
LLaMA 7B/13B/33B/65B q_proj,v_proj -
LLaMA-2 7B/13B/70B q_proj,v_proj llama2
Mistral 7B q_proj,v_proj mistral
Mixtral 8x7B q_proj,v_proj mistral
Phi-1.5/2 1.3B/2.7B Wqkv -
Qwen 1.8B/7B/14B/72B c_attn qwen
XVERSE 7B/13B/65B q_proj,v_proj xverse
Yi 6B/34B q_proj,v_proj yi
Yuan 2B/51B/102B q_proj,v_proj yuan

Note

Default module is used for the --lora_target argument, you can use --lora_target all to specify all the available modules.

For the "base" models, the --template argument can be chosen from default, alpaca, vicuna etc. But make sure to use the corresponding template for the "chat" models.

Please refer to constants.py for a full list of models we supported.

Supported Training Approaches

Approach Full-parameter Partial-parameter LoRA QLoRA
Pre-Training βœ… βœ… βœ… βœ…
Supervised Fine-Tuning βœ… βœ… βœ… βœ…
Reward Modeling βœ… βœ… βœ… βœ…
PPO Training βœ… βœ… βœ… βœ…
DPO Training βœ… βœ… βœ… βœ…

Note

Use --quantization_bit 4 argument to enable QLoRA.

Provided Datasets

Pre-training datasets
Supervised fine-tuning datasets
Preference datasets

Please refer to data/README.md for details.

Some datasets require confirmation before using them, so we recommend logging in with your Hugging Face account using these commands.

pip install --upgrade huggingface_hub
huggingface-cli login

Requirement

  • Python 3.8+ and PyTorch 1.13.1+
  • πŸ€—Transformers, Datasets, Accelerate, PEFT and TRL
  • sentencepiece, protobuf and tiktoken
  • jieba, rouge-chinese and nltk (used at evaluation and predict)
  • gradio and matplotlib (used in web UI)
  • uvicorn, fastapi and sse-starlette (used in API)

Hardware Requirement

Method Bits 7B 13B 30B 65B 8x7B
Full 16 160GB 320GB 600GB 1200GB 900GB
Freeze 16 20GB 40GB 120GB 240GB 200GB
LoRA 16 16GB 32GB 80GB 160GB 120GB
QLoRA 8 10GB 16GB 40GB 80GB 80GB
QLoRA 4 6GB 12GB 24GB 48GB 32GB

Getting Started

Data Preparation (optional)

Please refer to data/README.md for checking the details about the format of dataset files. You can either use a single .json file or a dataset loading script with multiple files to create a custom dataset.

Note

Please update data/dataset_info.json to use your custom dataset. About the format of this file, please refer to data/README.md.

Dependence Installation (optional)

git clone https://github.com/hiyouga/LLaMA-Factory.git
conda create -n llama_factory python=3.10
conda activate llama_factory
cd LLaMA-Factory
pip install -r requirements.txt

If you want to enable the quantized LoRA (QLoRA) on the Windows platform, you will be required to install a pre-built version of bitsandbytes library, which supports CUDA 11.1 to 12.1.

pip install https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.39.1-py3-none-win_amd64.whl

Use ModelScope Hub (optional)

If you have trouble with downloading models and datasets from Hugging Face, you can use LLaMA-Factory together with ModelScope in the following manner.

export USE_MODELSCOPE_HUB=1 # `set USE_MODELSCOPE_HUB=1` for Windows

Then you can train the corresponding model by specifying a model ID of the ModelScope Hub. (find a full list of model IDs at ModelScope Hub)

CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --model_name_or_path modelscope/Llama-2-7b-ms \
    ... # arguments (same as above)

LLaMA Board also supports using the models and datasets on the ModelScope Hub.

CUDA_VISIBLE_DEVICES=0 USE_MODELSCOPE_HUB=1 python src/train_web.py

Train on a single GPU

Important

If you want to train models on multiple GPUs, please refer to Distributed Training.

Pre-Training

CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --stage pt \
    --do_train \
    --model_name_or_path path_to_llama_model \
    --dataset wiki_demo \
    --finetuning_type lora \
    --lora_target q_proj,v_proj \
    --output_dir path_to_pt_checkpoint \
    --overwrite_cache \
    --per_device_train_batch_size 4 \
    --gradient_accumulation_steps 4 \
    --lr_scheduler_type cosine \
    --logging_steps 10 \
    --save_steps 1000 \
    --learning_rate 5e-5 \
    --num_train_epochs 3.0 \
    --plot_loss \
    --fp16

Supervised Fine-Tuning

CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --stage sft \
    --do_train \
    --model_name_or_path path_to_llama_model \
    --dataset alpaca_gpt4_en \
    --template default \
    --finetuning_type lora \
    --lora_target q_proj,v_proj \
    --output_dir path_to_sft_checkpoint \
    --overwrite_cache \
    --per_device_train_batch_size 4 \
    --gradient_accumulation_steps 4 \
    --lr_scheduler_type cosine \
    --logging_steps 10 \
    --save_steps 1000 \
    --learning_rate 5e-5 \
    --num_train_epochs 3.0 \
    --plot_loss \
    --fp16

Reward Modeling

CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --stage rm \
    --do_train \
    --model_name_or_path path_to_llama_model \
    --adapter_name_or_path path_to_sft_checkpoint \
    --create_new_adapter \
    --dataset comparison_gpt4_en \
    --template default \
    --finetuning_type lora \
    --lora_target q_proj,v_proj \
    --output_dir path_to_rm_checkpoint \
    --per_device_train_batch_size 2 \
    --gradient_accumulation_steps 4 \
    --lr_scheduler_type cosine \
    --logging_steps 10 \
    --save_steps 1000 \
    --learning_rate 1e-6 \
    --num_train_epochs 1.0 \
    --plot_loss \
    --fp16

PPO Training

CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --stage ppo \
    --do_train \
    --model_name_or_path path_to_llama_model \
    --adapter_name_or_path path_to_sft_checkpoint \
    --create_new_adapter \
    --dataset alpaca_gpt4_en \
    --template default \
    --finetuning_type lora \
    --lora_target q_proj,v_proj \
    --reward_model path_to_rm_checkpoint \
    --output_dir path_to_ppo_checkpoint \
    --per_device_train_batch_size 2 \
    --gradient_accumulation_steps 4 \
    --lr_scheduler_type cosine \
    --top_k 0 \
    --top_p 0.9 \
    --logging_steps 10 \
    --save_steps 1000 \
    --learning_rate 1e-5 \
    --num_train_epochs 1.0 \
    --plot_loss \
    --fp16

Warning

Use --per_device_train_batch_size=1 for LLaMA-2 models in fp16 PPO training.

DPO Training

CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --stage dpo \
    --do_train \
    --model_name_or_path path_to_llama_model \
    --adapter_name_or_path path_to_sft_checkpoint \
    --create_new_adapter \
    --dataset comparison_gpt4_en \
    --template default \
    --finetuning_type lora \
    --lora_target q_proj,v_proj \
    --output_dir path_to_dpo_checkpoint \
    --per_device_train_batch_size 2 \
    --gradient_accumulation_steps 4 \
    --lr_scheduler_type cosine \
    --logging_steps 10 \
    --save_steps 1000 \
    --learning_rate 1e-5 \
    --num_train_epochs 1.0 \
    --plot_loss \
    --fp16

Distributed Training

Use Huggingface Accelerate

accelerate config # configure the environment
accelerate launch src/train_bash.py # arguments (same as above)
Example config for LoRA training
compute_environment: LOCAL_MACHINE
distributed_type: MULTI_GPU
downcast_bf16: 'no'
gpu_ids: all
machine_rank: 0
main_training_function: main
mixed_precision: fp16
num_machines: 1
num_processes: 4
rdzv_backend: static
same_network: true
tpu_env: []
tpu_use_cluster: false
tpu_use_sudo: false
use_cpu: false

Use DeepSpeed

deepspeed --num_gpus 8 --master_port=9901 src/train_bash.py \
    --deepspeed ds_config.json \
    ... # arguments (same as above)
Example config for full-parameter training with DeepSpeed ZeRO-2
{
  "train_batch_size": "auto",
  "train_micro_batch_size_per_gpu": "auto",
  "gradient_accumulation_steps": "auto",
  "gradient_clipping": "auto",
  "zero_allow_untested_optimizer": true,
  "fp16": {
    "enabled": "auto",
    "loss_scale": 0,
    "initial_scale_power": 16,
    "loss_scale_window": 1000,
    "hysteresis": 2,
    "min_loss_scale": 1
  },
  "zero_optimization": {
    "stage": 2,
    "allgather_partitions": true,
    "allgather_bucket_size": 5e8,
    "reduce_scatter": true,
    "reduce_bucket_size": 5e8,
    "overlap_comm": false,
    "contiguous_gradients": true
  }
}

Merge LoRA weights and export model

python src/export_model.py \
    --model_name_or_path path_to_llama_model \
    --adapter_name_or_path path_to_checkpoint \
    --template default \
    --finetuning_type lora \
    --export_dir path_to_export \
    --export_size 2 \
    --export_legacy_format False

Warning

Merging LoRA weights into a quantized model is not supported.

Tip

Use --export_quantization_bit 4 and --export_quantization_dataset data/c4_demo.json to quantize the model after merging the LoRA weights.

API Demo

python src/api_demo.py \
    --model_name_or_path path_to_llama_model \
    --adapter_name_or_path path_to_checkpoint \
    --template default \
    --finetuning_type lora

Tip

Visit http://localhost:8000/docs for API documentation.

CLI Demo

python src/cli_demo.py \
    --model_name_or_path path_to_llama_model \
    --adapter_name_or_path path_to_checkpoint \
    --template default \
    --finetuning_type lora

Web Demo

python src/web_demo.py \
    --model_name_or_path path_to_llama_model \
    --adapter_name_or_path path_to_checkpoint \
    --template default \
    --finetuning_type lora

Evaluation

CUDA_VISIBLE_DEVICES=0 python src/evaluate.py \
    --model_name_or_path path_to_llama_model \
    --adapter_name_or_path path_to_checkpoint \
    --template vanilla \
    --finetuning_type lora
    --task mmlu \
    --split test \
    --lang en \
    --n_shot 5 \
    --batch_size 4

Predict

CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
    --stage sft \
    --do_predict \
    --model_name_or_path path_to_llama_model \
    --adapter_name_or_path path_to_checkpoint \
    --dataset alpaca_gpt4_en \
    --template default \
    --finetuning_type lora \
    --output_dir path_to_predict_result \
    --per_device_eval_batch_size 8 \
    --max_samples 100 \
    --predict_with_generate \
    --fp16

Warning

Use --per_device_train_batch_size=1 for LLaMA-2 models in fp16 predict.

Tip

We recommend using --per_device_eval_batch_size=1 and --max_target_length 128 at 4/8-bit predict.

Projects using LLaMA Factory

  • StarWhisper: A large language model for Astronomy, based on ChatGLM2-6B and Qwen-14B.
  • DISC-LawLLM: A large language model specialized in Chinese legal domain, based on Baichuan-13B, is capable of retrieving and reasoning on legal knowledge.
  • Sunsimiao: A large language model specialized in Chinese medical domain, based on Baichuan-7B and ChatGLM-6B.
  • CareGPT: A series of large language models for Chinese medical domain, based on LLaMA2-7B and Baichuan-13B.

Tip

If you have a project that should be incorporated, please contact via email or create a pull request.

License

This repository is licensed under the Apache-2.0 License.

Please follow the model licenses to use the corresponding model weights: Baichuan / Baichuan2 / BLOOM / ChatGLM3 / Falcon / InternLM / LLaMA / LLaMA-2 / Mistral / Phi-1.5 / Qwen / XVERSE / Yi / Yuan

Citation

If this work is helpful, please kindly cite as:

@Misc{llama-factory,
  title = {LLaMA Factory},
  author = {hiyouga},
  howpublished = {\url{https://github.com/hiyouga/LLaMA-Factory}},
  year = {2023}
}

Acknowledgement

This repo benefits from PEFT, QLoRA and FastChat. Thanks for their wonderful works.

Star History

Star History Chart