lancew/Qwen3-32B-GPTQ-INT8

Benchmark

Capability Benchmark	Qwen3-32B-FP8	Qwen3-32B-INT8
aime25@mean_acc	0.4555	0.4778
gpqa_diamond@mean_acc	0.6431	0.6431
ifeval@mean_prompt_level_strict	0.8262	0.8207
live_code_bench@mean_acc	0.5498	0.5573
mmlu_pro@mean_acc	0.7816	0.7832

• 查找"stop_reason": "max_tokens"的sample，发现均为重复回答至max_token，未发现有正常回答至最长的情况
• repeats : 3
• enable_thinking: True
• evalscope version: 1.4.2

Reference

• Best Practices