chat: qwen3_0_6b_fp8-hf: gsm8k_accuracy: 3.12 race-high_accuracy: 6.25 qwen3-0_6b-fp8-turbomind: gsm8k_accuracy: 40.62 race-high_accuracy: 65.62 base: qwen3-8b-base-turbomind: gsm8k_accuracy: 50.00 GPQA_diamond_accuracy: 15.62 race-high_accuracy: 62.50 winogrande_accuracy: 71.88 qwen3-8b-base-vllm: gsm8k_accuracy: 50.00 GPQA_diamond_accuracy: 21.88 race-high_accuracy: 59.38 winogrande_accuracy: 68.75 qwen3-8b-base-hf: gsm8k_accuracy: 50.00 GPQA_diamond_accuracy: 18.75 race-high_accuracy: 59.38 winogrande_accuracy: 53.12