Athene-V2-Chat-AWQ / README.md
radm's picture
Update README.md
414547b verified
---
license: other
license_name: nexusflowresearchlicense
license_link: >-
https://huggingface.co/Nexusflow/Athene-V2-Chat/resolve/main/Nexusflow_Research_License_.pdf
language:
- en
library_name: transformers
tags:
- RLHF
- Nexusflow
- Athene
- Chat Model
base_model:
- Qwen/Qwen2.5-72B-Instruct
---
# Athene-V2-Chat-72B: Rivaling GPT-4o across Benchmarks
- AWQ 4bit version of [Nexusflow/Athene-V2-Chat](https://huggingface.co/Nexusflow/Athene-V2-Chat)
- [Quantization code](https://docs.vllm.ai/en/latest/quantization/auto_awq.html)
## Eval AWQ version
Evaluation results on [ZebraLogic](https://github.com/WildEval/ZeroEval/blob/main/result_dirs/zebra-grid.summary.md)
```
β”‚ Model β”‚ Mode β”‚ N_Mode β”‚ N_Size β”‚ Puzzle Acc β”‚ Easy Puzzle Acc β”‚ Hard Puzzle Acc β”‚ Cell Acc β”‚ No answer β”‚ Total Puzzles β”‚ Reason Lens β”‚
β”‚ o1-preview-2024-09-12 β”‚ greedy β”‚ single β”‚ 1 β”‚ 71.4 β”‚ 98.57 β”‚ 60.83 β”‚ 75.14 β”‚ 0.3 β”‚ 1000 β”‚ 1565.88 β”‚
β”‚ claude-3-5-sonnet-20241022 β”‚ greedy β”‚ single β”‚ 1 β”‚ 36.2 β”‚ 91.07 β”‚ 14.86 β”‚ 54.27 β”‚ 0 β”‚ 1000 β”‚ 861.18 β”‚
β”‚ Llama-3.1-405B-Inst-fp8@together β”‚ greedy β”‚ single β”‚ 1 β”‚ 32.6 β”‚ 87.14 β”‚ 11.39 β”‚ 45.8 β”‚ 12.5 β”‚ 1000 β”‚ 314.66 β”‚
β”‚ Athene-V2-Chat-AWQ β”‚ greedy β”‚ single β”‚ 1 β”‚ 27.8 β”‚ 77.14 β”‚ 8.61 β”‚ 45.83 β”‚ 6.4 β”‚ 1000 β”‚ 1785.7 β”‚
β”‚ Qwen2.5-72B-Instruct β”‚ greedy β”‚ single β”‚ 1 β”‚ 26.6 β”‚ 76.43 β”‚ 7.22 β”‚ 40.92 β”‚ 11.9 β”‚ 1000 β”‚ 1795.9 β”‚
β”‚ Qwen2.5-32B-Instruct β”‚ greedy β”‚ single β”‚ 1 β”‚ 26.1 β”‚ 77.5 β”‚ 6.11 β”‚ 43.39 β”‚ 6.3 β”‚ 1000 β”‚ 1333.07 β”‚
β”‚ Athene-70B β”‚ greedy β”‚ single β”‚ 1 β”‚ 16.7 β”‚ 52.5 β”‚ 2.78 β”‚ 32.98 β”‚ 21.1 β”‚ 1000 β”‚ 391.19 β”‚
```