File size: 1,789 Bytes
fbb0935 920f20e fbb0935 920f20e b8d5c09 920f20e b8d5c09 920f20e 549e69d |
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 |
---
language:
- en
- zh
tags:
- qwen
- llama
- llama-2
---
[WIP]
Origin repository [JosephusCheung/Qwen-LLaMAfied-7B-Chat](https://huggingface.co/JosephusCheung/Qwen-LLaMAfied-7B-Chat).
This is the LLaMAfied version of [Qwen/Qwen-7B-Chat](https://huggingface.co/Qwen/Qwen-7B-Chat), recalibrated to fit the original LLaMA/LLaMA-2-like model structure.
You can use LlamaForCausalLM for model inference, which is the same as LLaMA/LLaMA-2 models.
I converted the tokenizer from tiktoken format to huggingface format, so you do not need to allow external codes when loading anymore.
The model has been edited to be white-labelled, meaning the model will no longer call itself a Qwen.
SPOILOR: Further finetuning is in progress, the current version is a work-in-progress, some knowledge may be biased and illusory due to structural changes. Will be updated very, very sooooooooooon.
PROMPT FORMAT: [chatml](https://github.com/openai/openai-python/blob/main/chatml.md)
CURRENT MMLU: 50.36
Issue: Compared to the original Qwen-Chat scoring 53.9, the MMLU score dropped slightly (-3.54) due to insufficient realignment.
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_vonjack__Qwen-LLaMAfied-HFTok-7B-Chat)
| Metric | Value |
|-----------------------|---------------------------|
| Avg. | 48.51 |
| ARC (25-shot) | 50.51 |
| HellaSwag (10-shot) | 83.65 |
| MMLU (5-shot) | 51.53 |
| TruthfulQA (0-shot) | 44.23 |
| Winogrande (5-shot) | 71.43 |
| GSM8K (5-shot) | 2.5 |
| DROP (3-shot) | 35.7 |
|