# JudgeLM **Repository Path**: qinzichen2023/JudgeLM ## Basic Information - **Project Name**: JudgeLM - **Description**: No description available - **Primary Language**: Unknown - **License**: Apache-2.0 - **Default Branch**: main - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 0 - **Created**: 2024-08-09 - **Last Updated**: 2024-08-09 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README # JudgeLM: Fine-tuned Large Language Models are Scalable Judges

[Lianghui Zhu](https://github.com/Unrealluver)^1,2, [Xinggang Wang](https://xwcv.github.io/)¹, [Xinlong Wang](https://www.xloong.wang/)² ¹[HUST](https://english.hust.edu.cn/), ²[BAAI](https://www.baai.ac.cn/english.html) ## Overview ![JudgeLM](./assets/judgelm_v1.1.png)

Abstract

Evaluating Large Language Models (LLMs) in open-ended scenarios is challenging because existing benchmarks and metrics can not measure them comprehensively. To address this problem, we propose to fine-tune LLMs as scalable judges (JudgeLM) to evaluate LLMs efficiently and effectively in open-ended benchmarks. We first propose a comprehensive, large-scale, high-quality dataset containing task seeds, LLMs-generated answers, and GPT-4-generated judgments for fine-tuning high-performance judges, as well as a new benchmark for evaluating the judges. We train JudgeLM at different scales from 7B, 13B, to 33B parameters, and conduct a systematic analysis of its capabilities and behaviors. We then analyze the key biases in fine-tuning LLM as a judge and consider them as position bias, knowledge bias, and format bias. To address these issues, JudgeLM introduces a bag of techniques including swap augmentation, reference support, and reference drop, which clearly enhance the judge's performance. JudgeLM obtains the state-of-the-art judge performance on both the existing PandaLM benchmark and our proposed new benchmark. Our JudgeLM is efficient and the JudgeLM-7B only needs 3 mins to judge 5K samples with 8 A100 GPUs. JudgeLM obtains high agreement with the teacher judge, achieving an agreement exceeding 90% that even surpasses human-to-human agreement. JudgeLM also demonstrates extended capabilities in being judges of the single answer, multimodal models, multiple answers, and multi-turn chat.

JudgeLM is an open platform for training, serving, and evaluating scalable large language model judges. - JudgeLM is a scalable language model judge, designed for evaluating LLMs in open-ended scenarios. It achieves an agreement exceeding 90\% that surpasses the human-to-human agreement. - JudgeLM dataset contains 100K judge samples for training and 5K judge samples for validation. All the judge samples have the GPT-4-generated high-quality judgements. JudgeLM's core features include: - The training and evaluation code for state-of-the-art LLM judges. - The broad capacities to deal with extended tasks. (e.g., judges of the single answer, multimodal models, multiple answers, and multi-turn chat) - A distributed multi-model serving system with web UI. ## News - [2023/10] We released **JudgeLM: Fine-tuned Large Language Models are Scalable Judges**. Check out the [paper](https://arxiv.org/abs/2310.17631). ## Contents - [Install](#install) - [Model Weights](#model-weights) - [Evaluation](#evaluation) - [Serving with Web GUI](#serving-with-web-gui) - [Fine-tuning](#fine-tuning) - [Citation](#citation) ## Install: From source 1. Clone this repository and navigate to the JudgeLM folder ```bash git clone https://github.com/baaivision/JudgeLM cd JudgeLM ``` 2. Install Package ```bash conda create -n judgelm python=3.10.10 -y conda activate judgelm pip3 install --upgrade pip pip3 install -e . pip install flash-attn==2.0.4 --no-build-isolation ``` ## Model Weights JudgeLM is based on LLaMA and should be used under LLaMA's [model license](https://github.com/facebookresearch/llama/blob/main/LICENSE). | Model | w/ reference? | Agreement↑ | Precision↑ | Recall↑ | F1↑ | Consistency↑ | |:------------------------------------------------------------------:|:-------------:|:----------:|:----------:|:-------:|:-----:|:------------:| | [**JudgeLM-7B**](https://huggingface.co/BAAI/JudgeLM-7B-v1.0) | ❎ | 81.11 | 69.67 | 78.39 | 72.21 | 83.57 | | [**JudgeLM-7B**](https://huggingface.co/BAAI/JudgeLM-7B-v1.0) | ✅ | 84.08 | 75.92 | 82.55 | 78.28 | 84.46 | | [**JudgeLM-13B**](https://huggingface.co/BAAI/JudgeLM-13B-v1.0) | ❎ | 84.33 | 73.69 | 80.51 | 76.17 | 85.01 | | [**JudgeLM-13B**](https://huggingface.co/BAAI/JudgeLM-13B-v1.0) | ✅ | 85.47 | 77.71 | 82.90 | 79.77 | 87.23 | | [**JudgeLM-33B** 🔥](https://huggingface.co/BAAI/JudgeLM-33B-v1.0) | ❎ | 89.03 | 80.97 | 84.76 | 82.64 | 91.36 | | [**JudgeLM-33B** 🔥](https://huggingface.co/BAAI/JudgeLM-33B-v1.0) | ✅ | 89.32 | 84.00 | 86.21 | 84.98 | 92.37 | ## Evaluation ![judge_pairs](./assets/judge_pairs_v1.0.png) ![judge_mmvet](./assets/mmvet_v1.0.png) JudgeLM can judge open-ended answers from LLMs, as well as the multimodal models. See instructions for running JudgeLM at [judgelm/llm_judge](judgelm/llm_judge). ## Serving with Web GUI ![gradio](./assets/gradio_v1.1.png) We use gradio to provide web server and UI for users to evaluate LLMs' performance at open-ended tasks. The demo can be tried [here](http://218.91.113.230:9004/). See instructions for running JudgeLM web server at [judgelm/serve](judgelm/serve). ## Fine-tuning ### Data The JudgeLM-100K dataset is available at [HuggingFace Datasets](https://huggingface.co/datasets/BAAI/JudgeLM-100K). ### Code and Hyperparameters Our code is based on [Vicuna](https://github.com/lm-sys/FastChat) with additional support for judging answer pairs. We use similar hyperparameters as the Vicuna. | Hyperparameter | Global Batch Size | Learning rate | Epochs | Max length | Weight decay | | --- | ---: | ---: | ---: | ---: | ---: | | JudgeLM-13B | 128 | 2e-5 | 3 | 2048 | 0 | ### Fine-tuning JudgeLM-7B with Local GPUs - You can use the following command to train JudgeLM-7B with 4 x A100 (40GB). Update `--model_name_or_path` with the actual path to Vicuna weights and `--data_path` with the actual path to JudgeLM data. ```bash torchrun --nproc_per_node=4 --master_port=20001 judgelm/train/train_mem.py \ --model_name_or_path="/share/project/lianghuizhu/vicuna-weights-collection-v1.3/vicuna-7b-v1.3" \ --data_path /share/project/lianghuizhu/JudgeLM-Project/JudgeLM/judgelm/data/JudgeLM/judgelm_train_100k.jsonl \ --bf16 True \ --output_dir="/home/zhulianghui/ProjectC_ChatGPT/alpaca/output/judgelm-debug-evaluator" \ --num_train_epochs 3 \ --per_device_train_batch_size 1 \ --per_device_eval_batch_size 1 \ --gradient_accumulation_steps 32 \ --evaluation_strategy no \ --save_strategy steps \ --save_steps 1000 \ --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 offload" \ --fsdp_transformer_layer_cls_to_wrap "LlamaDecoderLayer" \ --tf32 True \ --model_max_length 2048 \ --gradient_checkpointing True \ --run_name 7B-full-model \ --swap_aug_ratio 0.5 \ --ref_drop_ratio 0.5 ``` Tips: - If you are using V100 which is not supported by FlashAttention, you can use the [memory-efficient attention](https://arxiv.org/abs/2112.05682) implemented in [xFormers](https://github.com/facebookresearch/xformers). Install xformers and replace `judgelm/train/train_mem.py` above with [judgelm/train/train_xformers.py](judgelm/train/train_xformers.py). - If you meet out-of-memory due to "FSDP Warning: When using FSDP, it is efficient and recommended... ", see solutions [here](https://github.com/huggingface/transformers/issues/24724#issuecomment-1645189539). - If you meet out-of-memory during model saving, see solutions [here](https://github.com/pytorch/pytorch/issues/98823). ## Acknowledgement :heart: This project is based on Vicuna ([blog](https://vicuna.lmsys.org), [code](https://github.com/lm-sys/FastChat)), PandaLM ([paper](https://arxiv.org/abs/2306.05087), [code](https://github.com/WeOpenML/PandaLM)), LLM-Blender ([paper](https://arxiv.org/abs/2306.02561), [code](https://github.com/yuchenlin/LLM-Blender)). Thanks for their wonderful works. ## Citation The code (training, serving, and evaluation) in this repository is mostly developed for or derived from the paper below. Please cite it if you find the repository helpful. ``` @article{zhu2023judgelm, title={JudgeLM: Fine-tuned Large Language Models are Scalable Judges}, author={Lianghui Zhu and Xinggang Wang and Xinlong Wang}, year={2023}, eprint={2310.17631}, archivePrefix={arXiv}, primaryClass={cs.CL} } ```