List your company

Autoregressive Decoding Performance on XSum (val)

1,736.88Peak VRAM (MB)PrunePath-Triton
1,721.70321,824.14661,926.592,029.0334May 27, 2026

Evaluation results

MethodLinks
PrunePath-TritonBackbone=GPT-2 Medium, KV-cache=true, Implementation=Triton2026.051,736.88292.5912.4880.15
Dense GPT-2 MediumBackbone=GPT-2 Medium, KV-cache=true2026.052,116.3306.1513.0576.6