[PyT] Reduce test sizes in fused attn fp8 vs fp16 to avoid OOM #3020

cyanguwa merged 3 commits into

NVIDIA:mainNVIDIA/TransformerEngine:mainfrom

vedaanta:vedaanta/te-fp8-vs-f16-shrink-b1vedaanta/TransformerEngine:vedaanta/te-fp8-vs-f16-shrink-b1Copy head branch name to clipboard