flash-attn2: avoid incorrect out allocation

[drbh]

This PR removes an unneeded allocation of the out tensor which causes shape errors in some cases

[drbh]

/kernel-bot build-and-stage flash-attn2

[github-actions]

Build request processed.

Command: /kernel-bot build-and-stage flash-attn2
Mode: build and stage
Target branch: pr-913
PR head SHA: d3fafb05d468975e3a599a6300f3c9671d4628b4
Workflows: build.yaml, build-mac.yaml, build-windows.yaml

Dispatched (1):

• flash-attn2 (build.yaml): https://github.com/huggingface/kernels-community/actions/runs/26825117397

[drbh]

/kernel-bot build-and-merge

[github-actions]

Invalid command. Use /kernel-bot <build|build-and-stage|merge-and-upload|release> <kernel1> [kernel2 ...] [--branch <target_branch>].

[drbh]

/kernel-bot build-and-merge flash-attn2

[github-actions]

Invalid command. Use /kernel-bot <build|build-and-stage|merge-and-upload|release> <kernel1> [kernel2 ...] [--branch <target_branch>].

[drbh]

/kernel-bot merge-and-upload flash-attn2

[github-actions]

Build request failed.

Command: /kernel-bot merge-and-upload flash-attn2
Mode: merge, build and upload
Target branch: main
PR head SHA: d3fafb05d468975e3a599a6300f3c9671d4628b4
Workflows: build.yaml, build-mac.yaml, build-windows.yaml

Failure: Failed to merge PR before build/upload. Check mergeability and required checks.

[drbh]

/kernel-bot release flash-attn2

[github-actions]

Build request processed.

Command: /kernel-bot release flash-attn2
Mode: release (linux + mac + windows)
Target branch: ``
PR head SHA: d3fafb05d468975e3a599a6300f3c9671d4628b4
Workflows: `build.yaml, build-mac.yaml, build-windows.yaml`

Dispatched (1):

• flash-attn2 (build.yaml): https://github.com/huggingface/kernels-community/actions/runs/26883383263

[github-actions]

Build request processed.

Command: /kernel-bot release flash-attn2
Mode: release (linux + mac + windows)
Target branch: ``
PR head SHA: d3fafb05d468975e3a599a6300f3c9671d4628b4
Workflows: `build.yaml, build-mac.yaml, build-windows.yaml`

Dispatched (1):

• flash-attn2 (build.yaml): https://github.com/huggingface/kernels-community/actions/runs/26883467167