meta-llama_Meta-Llama-3-8B W6A6 (lambda_factor=5, seed=1)

Selected as best per-scheme using fakequant_wt103_ppl (lower is better).

Metrics (PPL)

split dense ppl fakequant ppl
wikitext-103 test 16.535811 20.450397
lambada validation 34.048287 41.175088
Downloads last month
2
Safetensors
Model size
8B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support