Commonsense Reasoning on HellaSwag TinyLlama-1.1B-Chat v1.0
50.87HellaSwag AccuracyFP32
Evaluation Results
| Method | Links | |
|---|---|---|
| FP32Precision=Full-precision2026.05 | 50.87 | |
| Quant.npuPrecision=w4a8, Weight Quantization=4-bit, Granularity=per-channel2026.05 | 48.93 | |
| MoblieQuantPrecision=w4a8, Weight Quantization=4-bit, Granularity=per-channel2026.05 | 47.87 |