trunk/e1994aea9ce307fb82feeb7524ab2c5d36771e4f: [ROCm] Bump AOTriton to 0.14b (#197747)
- PyTorch: 1271 events in the last 90 days
- PyTorch: 1257th Release in the last 90 days
- Previous: earlier the same day · ciflow/torchtitan/197821: [UPDATE] Update
What happened
Notable new features compared with 0.13b: New flyc backend: FlyDSL SDPA fp16/bf16 kernels for gfx950 and gfx1201, at parity with the Triton kernels and substantially faster on most head dimensions See AOTriton 0.14b Release Notes for measured performance improvments API BREAKING, ABI STABLE VarlenBits replaces the VarlenType enum Q and K now choose their sequence addressing independently, along three axes: tensor sh…
Summary assembled by rule from the sources below