lancew/Qwen3-14B-GPTQ-INT8

Benchmark

Capability Benchmark	Qwen3-14B-FP8	Qwen3-14B-INT8
aime25@mean_acc	0.4778	0.4222
gpqa_diamond@mean_acc	0.6195	0.6229
ifeval@mean_prompt_level_strict	0.8324	0.8306
live_code_bench@mean_acc	0.6423	0.643
mmlu_pro@mean_acc	0.7493	0.7551

• 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