82 lines
3.7 KiB
Markdown
82 lines
3.7 KiB
Markdown
---
|
||
frameworks:
|
||
- Pytorch
|
||
license: Apache License 2.0
|
||
base_model:
|
||
- Qwen/Qwen2.5-14B-Instruct
|
||
tasks:
|
||
- text-generation
|
||
language:
|
||
- en
|
||
- cn
|
||
tags:
|
||
- Chat
|
||
- Instruct
|
||
- fine-tuned
|
||
---
|
||
|
||
# Rombos-LLM-V2.6-Qwen-14b High-performance finetuned
|
||
|
||
- ***同系合集 [Rombos-LLM-Qwen2.5](https://www.modelscope.cn/collections/Rombos-LLM-Qwen25-4dc8690ad7f541)***
|
||
|
||
## Download
|
||
|
||
SDK下载
|
||
```bash
|
||
#安装ModelScope
|
||
pip install modelscope
|
||
```
|
||
```python
|
||
#SDK模型下载
|
||
from modelscope import snapshot_download
|
||
model_dir = snapshot_download('okwinds/Rombos-LLM-V2.6-Qwen-14b')
|
||
```
|
||
Git下载
|
||
```
|
||
#Git模型下载
|
||
git clone https://www.modelscope.cn/okwinds/Rombos-LLM-V2.6-Qwen-14b.git
|
||
```
|
||
|
||
# 模型简介
|
||
|
||
本模型转载于 huggingface 上的 [rombodawg/Rombos-LLM-V2.6-Qwen-14b](https://huggingface.co/rombodawg/Rombos-LLM-V2.6-Qwen-14b)
|
||
|
||
作者介绍到：本模型是一个基于 [Qwen/Qwen2.5-14B](https://www.modelscope.cn/models/Qwen/Qwen2.5-14B) 的 Continuous Fine-tuning 版本，并通过 [TIES-Merging](https://ar5iv.labs.arxiv.org/html/2306.01708) 方法将合并了adapter的 [Qwen/Qwen2.5-14B-Instruct](https://www.modelscope.cn/models/qwen/qwen2.5-14b-instruct) 与 [Qwen/Qwen2.5-14B](https://www.modelscope.cn/models/Qwen/Qwen2.5-14B) 进行了 TIES Merge。这使得本模型的性能优于原始的 base 模型以及 pretrain 后 fine-tune 的 Instruct 模型。
|
||
|
||
Model tree for [okwinds/Rombos-LLM-V2.6-Qwen-14b](https://www.modelscope.cn/models/okwinds/Rombos-LLM-V2.6-Qwen-14b) </br>
|
||
Base model - [Qwen/Qwen2.5-14B](https://www.modelscope.cn/models/Qwen/Qwen2.5-14B) </br>
|
||
<span style="font-weight: bold;">╰</span>─ Finetuned - [Qwen/Qwen2.5-14B-Instruct](https://www.modelscope.cn/models/Qwen/Qwen2.5-14B-Instruct) </br>
|
||
<span style="font-weight: bold;">╰</span>─ Finetuned - [本模型 okwinds/Rombos-LLM-V2.6-Qwen-14b](https://www.modelscope.cn/models/okwinds/Rombos-LLM-V2.6-Qwen-14b) </br>
|
||
|
||
# Benchmark
|
||
|
||
以下摘自 2024-10-29 的 Huggingface 上 [Leadboard](https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard) 排名数据
|
||
|
||
本模型以 14B 参数量在总榜排名第49。截图：
|
||
|
||
<img src="https://www.modelscope.cn/models/okwinds/Rombos-LLM-V2.6-Qwen-14b/resolve/master/HF-Leadboard-Total-2024-10-29.png" />
|
||
|
||
在 10-32B 参数量模型中排名第7。截图：
|
||
|
||
<img src="https://www.modelscope.cn/models/okwinds/Rombos-LLM-V2.6-Qwen-14b/resolve/master/HF-Leadboard-10-32B-2024-10-29.png" />
|
||
|
||
在 10-20B 参数量模型中排名第1。
|
||
|
||
<img src="https://www.modelscope.cn/models/okwinds/Rombos-LLM-V2.6-Qwen-14b/resolve/master/HF-Leadboard-10-20B-2024-10-29.png" />
|
||
|
||
截图中的数据：
|
||
|
||
| **Model** |**Average️**|**IFEval**|**BBH**|**MATH Lvl 5**|**GPQA**|**MUSR**|**MMLU-PRO**|
|
||
|-|-|-|-|-|-|-|-|
|
||
|<span style="color:#ff6600">rombodawg/Rombos-LLM-V2.6-Qwen-14b</span>|35.89|52.14|49.22|28.85|17|19.26|48.85|
|
||
|rombodawg/Rombos-LLM-V2.5-Qwen-14b|34.52|58.4|49.39|15.63|16.22|18.83|48.62|
|
||
|Tsunami-th/Tsunami-1.0-14B-Instruct |34.2|78.29|49.15|0|14.21|16.34|47.21|
|
||
|tanliboy/lambda-qwen2.5-14b-dpo-test |33.52|82.31|48.45|0|14.99|12.59|42.75|
|
||
|jpacifico/Chocolatine-14B-Instruct-DPO-v1.2 |33.3|68.52|49.85|17.98|10.07|12.35|41.07|
|
||
|TheTsar1209/qwen-carpmuscle-v0.2 |33.11|52.57|48.18|25|14.09|12.75|46.08|
|
||
|microsoft/Phi-3-medium-4k-instruct |32.67|64.23|49.38|16.99|11.52|13.05|40.84|
|
||
|TheTsar1209/qwen-carpmuscle-v0.1 |32.59|56.22|48.83|21.15|12.53|10.15|46.67|
|
||
|v000000/Qwen2.5-Lumen-14B |32.2|80.64|48.51|0|10.4|10.29|43.36|
|
||
|Qwen/Qwen2.5-14B-Instruct |32.18|81.58|48.36|0|9.62|10.16|43.38|
|
||
|v000000/Qwen2.5-14B-Gutenberg-1e-Delta |32.11|80.45|48.62|0|10.51|9.38|43.67|
|
||
|internlm/internlm2_5-20b-chat |32.08|70.1|62.83|0|9.51|16.74|33.31| |