Awesome
Lion: Adversarial Distillation of Proprietary Large Language Models (EMNLP 2023)
<p align="center" width="100%"> <a ><img src="pics/Lion.jpg" alt="Lion" style="width: 20%; min-width: 200px; display: block; margin: auto;"></a> </p> <p align="center"> <a href="https://arxiv.org/abs/2305.12870">[📄 Paper]</a> | <a href="https://huggingface.co/YuxinJiang/lion-7b">[🤗 Lion Weights]</a> <!-- <a href="https://7fc72e99b01b79af.gradio.app/">[:desktop_computer: Demo]</a> --> </p> <hr> <!-- The dataset is CC BY NC 4.0 (allowing only non-commercial use) and models trained using the dataset should not be used outside of research purposes. --> <!-- The weight diff is also CC BY NC 4.0 (allowing only non-commercial use). --> <!-- #### Tuned on 70k instruction-following data, Lion (7B) can achieve 95% capability of ChatGPT! -->News
- [October 8, 2023] Our paper has been accepted to EMNLP 2023.
- [June 10, 2023] We released insturctions for addressing OOM during fine-tuning, check it in Training Process.
- [May 26, 2023] We released the model weights. Check out the 7B model!
- [May 25, 2023] We released an online demo, try our model here!
- [May 23, 2023] We released the code for training and inference.
Contents
<!-- 2. [Online Demo](#online-demo) -->Overview
<p align="center"> <img width="700" height="320" src="https://github.com/YJiangcm/Lion/blob/master/pics/overview.jpg"> </p>The high-level overview of our adversarial distillation framework, where we craft a compact Student LLM based on a superior closed-source LLM that serves three roles: the Teacher, the Referee, and the Generator. From left to right, there are three stages in an iteration:
- an imitation stage to align the student’s response with the teacher’s response;
- a discrimination stage to identify hard samples;
- a generation stage to produce new hard samples for escalating the challenges presented to the student model.
Recovering Lion weights
We release Lion weights as delta weights to comply with the LLaMA model license.
You can add our delta to the original LLaMA weights to obtain the Lion weights. Instructions:
- Get the original LLaMA weights in the huggingface format by following the instructions here
- Please download our delta model from Hugging Face
- Use the following scripts to get Lion weights by applying our delta:
python src/weight_diff.py recover --path_raw huggyllama/llama-7b --path_diff YuxinJiang/lion-7b --path_tuned <path_to_store_recovered_weights>
Inference
For inference and training of Lion, please first install the requirements:
pip install -r requirements.txt
We provide the decoding script for Lion, which reads a input file and generates corresponding responses for each sample, and finally consolidates them into an output file. It can be run on a single machine with 16GB GPU.
python src/lion_inference.py \
--model_dir <path_to_hf_converted_lion_ckpt_and_tokenizer> \
--data_dir <path_to_input_json_file> \
--output_dir <path_to_output_json_file> \
--num_gpus 1
Training Process
Below shows one iteration of our adversarial distillation framework.
1. Imitation Stage
1.1 Acquire the teacher's response on the Train Pool
python src/chatgpt_inference.py \
-q <path_to_json_file_for_the_Train_Pool> \
-o <path_to_chatgpt_inference_for_the_Train_Pool> \
--api_key <your_openai_api_key>
1.2 Instruction-tuning the student based on the teacher’s response on the Train Pool
Fine-tuning was conducted on a machine with 8 A100 80G GPUs.
torchrun --nproc_per_node=8 --master_port=<your_random_port> src/train.py \
--model_name_or_path <path_to_hf_converted_ckpt_and_tokenizer> \
--data_path <path_to_chatgpt_inference_for_the_Train_Pool> \
--bf16 True \
--output_dir result \
--num_train_epochs 3 \
--model_max_length 1024 \
--per_device_train_batch_size 2 \
--per_device_eval_batch_size 2 \
--gradient_accumulation_steps 8 \
--evaluation_strategy "no" \
--save_strategy "steps" \
--save_steps 600 \
--save_total_limit 1 \
--learning_rate 2e-5 \
--weight_decay 0. \
--warmup_ratio 0.03 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--fsdp "full_shard auto_wrap" \
--fsdp_transformer_layer_cls_to_wrap 'LlamaDecoderLayer' \
--tf32 True
Addressing OOM
Naively, fine-tuning a 7B model requires about 7 x 8 x 2 = 112 GB of VRAM. Commands given above enable parameter sharding, so no redundant model copy is stored on any GPU. If you'd like to further reduce the memory footprint, here are some options:
-
Turn on CPU offload for FSDP with
--fsdp "full_shard auto_wrap offload"
. This saves VRAM at the cost of longer runtime. -
In our experience, DeepSpeed stage-3 (with offload) can at times be more memory efficient than FSDP with offload. Here's an example to use DeepSpeed stage-3 with 8 GPUs with both parameter and optimizer offload:
deepspeed src/train_deepspeed.py \ --model_name_or_path <path_to_hf_converted_ckpt_and_tokenizer> \ --data_path <path_to_chatgpt_inference_for_the_Train_Pool> \ --output_dir result \ --num_train_epochs 3 \ --model_max_length 1024 \ --per_device_train_batch_size 16 \ --per_device_eval_batch_size 1 \ --gradient_accumulation_steps 1 \ --evaluation_strategy "no" \ --save_strategy "steps" \ --save_steps 600 \ --save_total_limit 1 \ --learning_rate 2e-5 \ --warmup_ratio 0.03 \ --logging_steps 1 \ --lr_scheduler_type "cosine" \ --report_to "tensorboard" \ --gradient_checkpointing True \ --deepspeed srcs/configs/deepspeed_config.json \ --fp16 True
- The DeepSpeed library also provides some helpful functions to estimate memory usage.
-
LoRA fine-tunes low-rank slices of the query, key, and value embedding heads. This can reduce the total memory footprint from 112GB to about 7x4=28GB. We may release our re-implemention of this in the future, but for now the peft codebase can be a useful resource.
2. Discrimination Stage
2.1 Acquire the teacher's response on the Cache Pool
python src/chatgpt_inference.py \
-q <path_to_json_file_for_the_Cache_Pool> \
-o <path_to_chatgpt_inference_for_the_Cache_Pool> \
--api_key <your_openai_api_key>
2.2 Acquire the student's response on the Cache Pool
python src/lion_inference.py \
--model_dir <path_to_hf_converted_lion_ckpt_and_tokenizer> \
--data_dir <path_to_json_file_for_the_Cache_Pool> \
--output_dir <path_to_lion_inference_for_the_Cache_Pool> \
--num_gpus 8
2.3 Ask the referee to output two scores according to the respose quality of the teacher and the student
To mitigate the position bias of the LLM referee, we conduct two runs by exchanging the positions of the teacher's response and the student's response.
python src/chatgpt_referee.py \
-a <path_to_chatgpt_inference_for_the_Cache_Pool> <path_to_lion_inference_for_the_Cache_Pool> \
-o <path_to_output_review_chatgpt_lion_file> \
--api_key <your_openai_api_key>
python src/chatgpt_referee.py \
-a <path_to_lion_inference_for_the_Cache_Pool> <path_to_chatgpt_inference_for_the_Cache_Pool> \
-o <path_to_output_review_lion_chatgpt_file> \
--api_key <your_openai_api_key>
2.4 Discriminate hard instructions and easy instructions
python src/discrimination.py \
--review12_path <path_to_output_review_chatgpt_lion_file> \
--review21_path <path_to_output_review_lion_chatgpt_file> \
--chatgpt_inference_path <path_to_chatgpt_inference_for_the_Cache_Pool> \
--lion_inference_path <path_to_lion_inference_for_the_Cache_Pool> \
--hard_save_path <path_to_identified_hard_instructions> \
--easy_save_path <path_to_identified_easy_instructions>
3. Generation Stage
3.1 Generate new hard instructions
python -m src/generate_hard_instruction generate_instruction_following_data \
--seed_tasks_path <path_to_identified_hard_instructions> \
--all_tasks_path <path_to_json_file_for_the_Cache_Pool> \
--output_dir <path_to_generated_hard_instructions> \
--num_instructions_to_generate 3000 \
--api_key <your_openai_api_key>
3.2 Generate new easy instructions
python -m src/generate_easy_instruction generate_instruction_following_data \
--seed_tasks_path <path_to_identified_easy_instructions> \
--all_tasks_path <path_to_json_file_for_the_Cache_Pool> \
--output_dir <path_to_generated_easy_instructions> \
--num_instructions_to_generate 3000 \
--api_key <your_openai_api_key>
Evaluation
Results for Open-ended Generation Dataset
We leverage GPT-4 to automatically assess the quality of responses (rated on a scale of 1 to 10) between a reference model (ChatGPT) and a candidate model. Subsequently, we calculate the candidate model’s performance as the percentage of the total score it achieves compared to the reference model.
<p align="center"> <img width="330" height="500" src="https://github.com/YJiangcm/Lion/blob/master/pics/open_ended_generation.png"> </p>Results for Reasoning Dataset
<p align="center"> <img width="600" height="670" src="https://github.com/YJiangcm/Lion/blob/master/pics/reasoning.png"> </p>Citation
Please cite our paper if you use the code in this repo.
@inproceedings{jiang-etal-2023-lion,
title = "Lion: Adversarial Distillation of Proprietary Large Language Models",
author = "Jiang, Yuxin and
Chan, Chunkit and
Chen, Mingyang and
Wang, Wei",
editor = "Bouamor, Houda and
Pino, Juan and
Bali, Kalika",
booktitle = "Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing",
month = dec,
year = "2023",
address = "Singapore",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.emnlp-main.189",
doi = "10.18653/v1/2023.emnlp-main.189",
pages = "3134--3154",
}
Disclaimer
⚠️ Lion is intended and licensed for research use ONLY. Commercial use is strictly prohibited. The content produced by any version of Lion is influenced by uncontrollable variables such as randomness, and therefore, the accuracy of the output cannot be guaranteed by this project. This project does not accept any legal liability for the content of the model output, nor does it assume responsibility for any losses incurred due to the use of associated resources and output results.