[cuda backend] skip fully-masked KV blocks calculation in SDPA #3525

	name: Docathon Labels Sync

	on:
	pull_request_target:
	types: [opened, synchronize, edited]
	branches: [main]

	jobs:
	check-labels:
	if: github.repository_owner == 'pytorch'
	runs-on: ubuntu-latest
	permissions:
	issues: write
	pull-requests: write
	steps:
	- name: Check out the repo
	uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
	with:
	fetch-depth: 1
	- name: Set up Python
	uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
	with:
	python-version: 3.x
	- name: Install dependencies
	run: \|
	pip install requests==2.32.3
	pip install PyGithub==2.3.0
	- name: Run Python script
	env:
	GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
	run: python ./.github/scripts/docathon-label-sync.py ${{ github.event.pull_request.number }}

Provide feedback