Mathematical Reasoning on GSM8K Instruct (test)
76.12AccuracyFull precision
Evaluation results
| Method | Links | |
|---|---|---|
| Full precisionModel=Llama-3 Instruct 8B, Precision=Full2025.06 | 76.12 | |
| UniQuan_FScheme=BCQ, Bit-width=4-bit, Model=Llama-3 Instruct 8B2025.06 | 72.38 | |
| FlexRoundScheme=UQ, Bit-width=4-bit, Model=Llama-3 Instruct 8B2025.06 | 70.66 | |
| OmniQuantScheme=UQ, Bit-width=4-bit, Model=Llama-3 Instruct 8B2025.06 | 59.16 | |
| UniQuan_FScheme=BCQ, Bit-width=3-bit, Model=Llama-3 Instruct 8B2025.06 | 58.73 | |
| FlexRoundScheme=UQ, Bit-width=3-bit, Model=Llama-3 Instruct 8B2025.06 | 54.13 | |
| RTNScheme=UQ, Bit-width=4-bit, Model=Llama-3 Instruct 8B2025.06 | 49.43 | |
| OmniQuantScheme=UQ, Bit-width=3-bit, Model=Llama-3 Instruct 8B2025.06 | 2.78 | |
| RTNScheme=UQ, Bit-width=3-bit, Model=Llama-3 Instruct 8B2025.06 | 0.08 | |
| ALTERNATINGScheme=BCQ, Bit-width=4-bit, Model=Llama-3 Instruct 8B2025.06 | 0 | |
| ALTERNATINGScheme=BCQ, Bit-width=3-bit, Model=Llama-3 Instruct 8B2025.06 | 0 |