Wiki C4 PIQA ARC-E ARC-C HellaSwag Wino Avg.
Unquantized 2.82 7.18 82.81 85.31 59.64 67.49 82.00 75.45
W4G64 3.09 7.53 83.03 85.52 58.19 67.04 80.43 74.84
W3G64 4.29 8.91 82.04 83.29 54.78 64.99 78.14 72.65

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.

Downloads last month
3
Safetensors
Model size
20.3B params
Tensor type
FP16
·
F32
·
I16
·
Inference API
Unable to determine this model's library. Check the docs .

Collection including radi-cho/Meta-Llama-3.1-70B-FLUTE