nielsr HF Staff commited on
Commit
a897a58
·
verified ·
1 Parent(s): 4720aa4

Improve model card with metadata and description

Browse files

Hi! I'm Niels, part of the community science team at Hugging Face.

This PR improves the model card for this repository by:
- Adding `library_name: transformers` and `pipeline_tag: text-generation` to the metadata for better discoverability and to enable the "how to use" widget.
- Adding a model description that links to the associated research paper: [Mix Data or Merge Models? Balancing the Helpfulness, Honesty, and Harmlessness of Large Language Model via Model Merging](https://huggingface.co/papers/2502.06876).
- Highlighting the core methodology used (RESM merging).

This helps users understand the context and technical details of the model.

Files changed (1) hide show
  1. README.md +17 -3
README.md CHANGED
@@ -1,14 +1,28 @@
1
  ---
2
- license: apache-2.0
3
  base_model:
4
  - mistralai/Mistral-7B-Instruct-v0.2
 
 
 
5
  ---
6
 
 
 
 
 
 
 
 
 
 
 
7
  ## Citation
8
- ```
 
9
  @article{yang2025mix,
10
  title={Mix Data or Merge Models? Balancing the Helpfulness, Honesty, and Harmlessness of Large Language Model via Model Merging},
11
  author={Yang, Jinluan and Jin, Dingnan and Tang, Anke and Shen, Li and Zhu, Didi and Chen, Zhengyu and Wang, Daixin and Cui, Qing and Zhang, Zhiqiang and Zhou, Jun and others},
12
  journal={arXiv preprint arXiv:2502.06876},
13
  year={2025}
14
- }
 
 
1
  ---
 
2
  base_model:
3
  - mistralai/Mistral-7B-Instruct-v0.2
4
+ license: apache-2.0
5
+ library_name: transformers
6
+ pipeline_tag: text-generation
7
  ---
8
 
9
+ # Mistral-7B-RESM
10
+
11
+ This model is a 3H-aligned Large Language Model (LLM) developed as part of the research paper [Mix Data or Merge Models? Balancing the Helpfulness, Honesty, and Harmlessness of Large Language Model via Model Merging](https://huggingface.co/papers/2502.06876).
12
+
13
+ ## Model Description
14
+
15
+ Mistral-7B-RESM is based on [Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2) and was optimized across three critical dimensions: **Helpfulness, Honesty, and Harmlessness (3H optimization)**.
16
+
17
+ The model was created using a novel parameter-level merging strategy called **RESM** (**R**eweighting **E**nhanced task **S**ingular **M**erging). RESM addresses challenges in 3H-aligned LLM merging, such as preference noise accumulation and layer sparsity adaptation, through outlier weighting and sparsity-aware rank selection strategies.
18
+
19
  ## Citation
20
+
21
+ ```bibtex
22
  @article{yang2025mix,
23
  title={Mix Data or Merge Models? Balancing the Helpfulness, Honesty, and Harmlessness of Large Language Model via Model Merging},
24
  author={Yang, Jinluan and Jin, Dingnan and Tang, Anke and Shen, Li and Zhu, Didi and Chen, Zhengyu and Wang, Daixin and Cui, Qing and Zhang, Zhiqiang and Zhou, Jun and others},
25
  journal={arXiv preprint arXiv:2502.06876},
26
  year={2025}
27
+ }
28
+ ```