--- license: llama3 --- | | Wiki | C4 | PIQA | ARC-E | ARC-C | HellaSwag | Wino | Avg. | | ----------- | ---- | ---- | ----- | ----- | ----- | --------- | ----- | ----- | | Unquantized | 2.9 | 6.9 | 82.4 | 86.9 | 60.3 | 66.4 | 80.6 | 75.3 | | W4G64 | 3.03 | 7.03 | 82.15 | 85.98 | 57.85 | 66.17 | 79.79 | 74.39 | | W3G64 | 4.15 | 8.10 | 80.74 | 83.71 | 55.29 | 64.05 | 78.45 | 72.45 | Revisions available in this repository: - `main` (W4G64, scales learned); - `nfl_w3g64` (W3G64, scales learned); Evaluations are provided for models with learned scales.
Benchmark scores (zero-shot) are computed with [`lm-evaluation-harness`](https://github.com/EleutherAI/lm-evaluation-harness).