File size: 2,068 Bytes
3ebb4f2 0e2dd85 3ebb4f2 9584bb2 ff218b4 3ebb4f2 b013e7f 3ebb4f2 478ba65 5dfe497 478ba65 3ebb4f2 d78773b 3ebb4f2 d78773b 3ebb4f2 0e2dd85 d78773b 3ebb4f2 0e2dd85 3ebb4f2 d78773b 3ebb4f2 9584bb2 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 |
---
base_model:
- beomi/Llama-3-Open-Ko-8B
- ProbeMedicalYonseiMAILab/medllama3-v20
library_name: transformers
tags:
- mergekit
- merge
license: apache-2.0
datasets:
- sean0042/KorMedMCQA
---
# BioLlama-Ko-8B
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
## 🏆 Evaluation
### [kormedmcqa(한국어 의학 벤치마크)](https://huggingface.co/datasets/sean0042/KorMedMCQA)
| Model | Doctor | Nurse | Pharm | Avg |
|------------------------------------------|-------|-------|-------|-------|
| gpt-4-0613 | 75.09 | 85.86 | 83.22 | 81.39 |
| **iRASC/BioLlama-Ko-8B** | **45.26** | **63.37** | **58.47** | **55.70** |
| gpt-3.5-turbo-0613 | 41.75 | 62.18 | 56.35 | 53.43 |
| llama2-70b | 42.46 | 63.54 | 53.26 | 53.09 |
| SOLAR-10.7B-v1.0 | 37.19 | 55.54 | 54.07 | 48.93 |
| ProbeMedicalYonseiMAILab/medllama3-v20 | 37.19 | 54.68 | 50.65 | 47.51 |
| beomi/Llama-3-Open-Ko-8B | 38.95 | 53.49 | 46.09 | 46.18 |
## Merge Details
### Merge Method
This model was merged using the [DARE](https://arxiv.org/abs/2311.03099) [TIES](https://arxiv.org/abs/2306.01708) merge method using [ProbeMedicalYonseiMAILab/medllama3-v20](https://huggingface.co/ProbeMedicalYonseiMAILab/medllama3-v20) as a base.
### Models Merged
The following models were included in the merge:
* [beomi/Llama-3-Open-Ko-8B](https://huggingface.co/beomi/Llama-3-Open-Ko-8B)
### Configuration
The following YAML configuration was used to produce this model:
```yaml
models:
- model: ProbeMedicalYonseiMAILab/medllama3-v20
- model: beomi/Llama-3-Open-Ko-8B
parameters:
density: 0.8
weight: 0.45
merge_method: dare_ties
base_model: ProbeMedicalYonseiMAILab/medllama3-v20
parameters:
int8_mask: true
dtype: bfloat16
``` |