Bhavya Medishetty

icon company AMD

Results 8 comments of Bhavya Medishetty

Request to update flash_attention in deepspeed inference

@loadams I am running stable diffusion with deepspeed inference.

update triton_ops.py from triton/python/tutorials/06-fused-attention.py

@microsoft-github-policy-service agree company="AMD"

tl.dot updated https://github.com/openai/triton/issues/1098

> @bmedishe have you tested this with a stable diffusion model? It causes DeepSpeed to hang for me. @mrwyattii Hi , I see a hang on AMD GPU too, what...

[ROCm] temporary workaround till __double2half support enabled in HIP

@jithunnair-amd @amathews-amd @jeffdaily

Sow ms7 enable test reference numerics small all except complex

These tests were skipped from PR [Various OpInfo architecture improvements](https://github.com/pytorch/pytorch/pull/75951) few tests belonging to this package were enabled for rocm in [ROCM: Enable few more tests for ROCM](https://github.com/pytorch/pytorch/pull/77669) Before PR...

Sow ms7 enable test reference numerics small all except complex

jenkins retest this please

Sow ms7 enable test reference numerics small all except complex

All the above tests pass in upstream CI https://github.com/pytorch/pytorch/actions/runs/3465630790/jobs/5788831616

Sow ms7 enable test reference numerics small all except complex

> @bmedishe , Can you raise this change to upstream pytorch/pytorch and another PR into rocm5.5_internal_testing branch @pruthvistony upstream PR which is merged https://github.com/pytorch/pytorch/pull/89007 will file a PR on rocm5.5_internal_testing...