Remove M3 Coding and Agentic Benchmark Results (BF16 baseline) table
Browse files
README.md
CHANGED
|
@@ -129,16 +129,6 @@ vllm serve nvidia/MiniMax-M3-NVFP4 \
|
|
| 129 |
|
| 130 |
### Evaluation
|
| 131 |
|
| 132 |
-
**M3 Coding and Agentic Benchmark Results (BF16 baseline, as reported by MiniMax):**
|
| 133 |
-
|
| 134 |
-
| **Benchmark** | **M3 (BF16)** |
|
| 135 |
-
|---|---|
|
| 136 |
-
| SWE-Bench Pro | 59.0% |
|
| 137 |
-
| Terminal Bench 2.1 | 66.0% |
|
| 138 |
-
| SWE-fficiency | 34.8% |
|
| 139 |
-
| KernelBench Hard | 28.8% |
|
| 140 |
-
| MCP Atlas | 74.2% |
|
| 141 |
-
|
| 142 |
**NVFP4 Quantization Accuracy (vs. FP8 baseline):**
|
| 143 |
|
| 144 |
| **Precision** | **GPQA Diamond** | **AA-LCR** | **τ²-Telecom** | **MMMU-Pro** | **SciCode** |
|
|
|
|
| 129 |
|
| 130 |
### Evaluation
|
| 131 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 132 |
**NVFP4 Quantization Accuracy (vs. FP8 baseline):**
|
| 133 |
|
| 134 |
| **Precision** | **GPQA Diamond** | **AA-LCR** | **τ²-Telecom** | **MMMU-Pro** | **SciCode** |
|