Zero-shot Reasoning on Zero-shot Average
66AccuracyBF16 baseline
Evaluation Results
| Method | Links | |
|---|---|---|
| BF16 baselineFormat=BF16, Model Family=Llama3, Model Size=8B2026.01 | 66 | |
| MR-GPTQ / BRQFormat=MXFP4, Model Family=Llama3, Model Size=8B2026.01 | 63.4 | |
| MixQuant†Format=INT4, Model Family=Llama3, Model Size=8B2026.01 | 60.9 | |
| MixQuant*Format=INT4, Model Family=Llama3, Model Size=8B2026.01 | 60.3 | |
| BF16 baselineFormat=BF16, Model Family=Llama3, Model Size=3B2026.01 | 59.8 | |
| MixQuant†Format=MXFP4, Model Family=Qwen3, Model Size=8B2026.01 | 54.2 | |
| BF16 baselineFormat=BF16, Model Family=Qwen3, Model Size=8B2026.01 | 53.5 | |
| BF16 baselineFormat=BF16, Model Family=Llama3, Model Size=1B2026.01 | 51.8 | |
| BF16 baselineFormat=BF16, Model Family=Qwen3, Model Size=4B2026.01 | 50.9 | |
| MixQuant†Format=FP4, Model Family=Qwen3, Model Size=8B2026.01 | 49.6 | |
| BF16 baselineFormat=BF16, Model Family=Qwen3, Model Size=1.7B2026.01 | 48.4 |