raihan-js/roofline-decoding-results download history

raihan-js/roofline-decoding-results is a dataset on the Hugging Face Hub. In the last 30 days it was downloaded 36 times (36 in the last 7 days), and 36 times in total. It ranks #237,825 among datasets by monthly downloads.

Roofline-First Decoding: results Result files for roofline-decoding: a fused 4-bit dequantise-plus-GEMV Triton kernel for batch-1 decoding, measured against the memory-bandwidth roofline of one RTX 3060 12GB. Files roofline.json: measured copy bandwidth (323.9 GB/s) an

Open raihan-js/roofline-decoding-results on Hugging Face

Sister project: Paper Pulse, the upvote history of every Hugging Face Daily Paper.