LianghuiZhu commited on
Commit
dfbebe0
1 Parent(s): 53d5361

Create README.md

Browse files

[add] initial commit.

Files changed (1) hide show
  1. README.md +68 -0
README.md ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ ---
3
+ inference: false
4
+ language:
5
+ - en
6
+ tags:
7
+ - instruction-finetuning
8
+ pretty_name: JudgeLM-100K
9
+ task_categories:
10
+ - text-generation
11
+ ---
12
+
13
+
14
+ <br>
15
+
16
+ # JudgeLM Model Card
17
+
18
+ ## Model Details
19
+
20
+ JudgeLM is a judge model trained by fine-tuning Vicuna on JudgeLM-100K dataset.
21
+
22
+ - **Developed by:** [HUST](https://english.hust.edu.cn/), [BAAI](https://www.baai.ac.cn/english.html)
23
+ - **Model type:** An auto-regressive language model based on the transformer architecture.
24
+ - **License:** Non-commercial license
25
+ - **Finetuned from model:** [Vicuna](https://vicuna.lmsys.org).
26
+
27
+ ### Model Sources
28
+
29
+ - **Repository:** https://github.com/baaivision/JudgeLM
30
+ - **Paper:** https://arxiv.org/abs/2310.17631
31
+ - **Demo:** http://218.91.113.230:9004/
32
+
33
+ ## Uses
34
+
35
+ The primary use of JudgeLM is research on evaluating the performance of large language models and chatbots.
36
+ The primary intended users of the model are researchers and hobbyists in natural language processing, machine learning, and artificial intelligence.
37
+
38
+ ## How to Get Started with the Model
39
+
40
+ - Judge large language models with this model: https://github.com/baaivision/JudgeLM/tree/main/judgelm/llm_judge.
41
+ - Serve this model with the gradio: https://github.com/baaivision/JudgeLM/tree/main/judgelm/serve.
42
+
43
+ ## Training Details
44
+
45
+ JudgeLM v1.0 is fine-tuned from Vicuna-v1.3 with supervised instruction fine-tuning.
46
+ The training data is around 200K judge samples from [JudgeLM-100K dataset](https://huggingface.co/datasets/BAAI/JudgeLM-100K).
47
+ See more details in the "Fine-tuning Settings" section in the appendix of this [paper](https://arxiv.org/abs/2310.17631).
48
+
49
+ ## Evaluation
50
+
51
+ JudgeLM is evaluated on JudgeLM val set, with judgements produced by GPT-4 teacher. See more details in this [paper](https://arxiv.org/abs/2310.17631) and try it with [code](https://github.com/baaivision/JudgeLM/tree/main/judgelm/llm_judge).
52
+
53
+ ## Additional Information
54
+
55
+ ### Citation Information
56
+
57
+ ```
58
+ @article{zhu2023judgelm,
59
+ title={JudgeLM: Fine-tuned Large Language Models are Scalable Judges},
60
+ author={Lianghui Zhu and Xinggang Wang and Xinlong Wang},
61
+ year={2023},
62
+ eprint={2310.17631},
63
+ archivePrefix={arXiv},
64
+ primaryClass={cs.CL}
65
+ }
66
+ ```
67
+
68
+