Skip to content

Refactor dsv4 decode_sparse_attn; restore a2a3 ring sizes; drop sim from decode_{swa,csa,hca} #393

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
zhangqi-chen merged 11 commits into from
May 26, 2026
Merged

Refactor dsv4 decode_sparse_attn; restore a2a3 ring sizes; drop sim from decode_{swa,csa,hca} #393

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
1bfe0c8
Refactor: clean up dsv4 decode_sparse_attn local constants and buffers
zhangqi-chen May 26, 2026
b1742ea
Refactor: drop ATTN_TOKEN_TILE in dsv4 decode_sparse_attn
zhangqi-chen May 26, 2026
9b5c21e
Refactor: shrink dsv4 decode_sparse_attn scratch tensors
zhangqi-chen May 26, 2026
55a888b
Refactor: fuse merge+norm and tighten valid-shape in dsv4 decode_spar…
zhangqi-chen May 26, 2026
5d1f65b
Refactor: tighten dsv4 decode_sparse_attn syntax and naming
zhangqi-chen May 26, 2026
82f40ff
Refactor: fuse rope/proj_a/proj_b, double quant, simplify comments
zhangqi-chen May 26, 2026
c130158
Refactor: replace pl.parallel+pl.at with pl.spmd in dsv4 decode_spars…
zhangqi-chen May 26, 2026
e884207
Refactor: fuse qk_softmax + pv into one scope in dsv4 decode_sparse_attn
zhangqi-chen May 26, 2026
9a69914
Refactor: split qk_pv and merge_norm into separate pl.spmd loops
zhangqi-chen May 26, 2026
05d65ef
CI: restore larger PTO2_RING_* pool sizes for a2a3 jobs
zhangqi-chen May 26, 2026
0c8486d
Chore: drop sim platforms from dsv4 decode_{swa,csa,hca} -p choices
zhangqi-chen May 26, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions .github/workflows/ci.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -330,9 +330,9 @@ jobs:
shell: bash
env:
PYTHONPATH: ${{ github.workspace }}
PTO2_RING_DEP_POOL: 524288
PTO2_RING_TASK_WINDOW: 524288
PTO2_RING_HEAP: 1073741824
PTO2_RING_DEP_POOL: 1048576
PTO2_RING_TASK_WINDOW: 1048576
PTO2_RING_HEAP: 4294967296
run: |
set +e
if [ "${{ github.event_name }}" = "pull_request" ]; then
Expand Down
6 changes: 3 additions & 3 deletions .github/workflows/daily_ci.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -242,9 +242,9 @@ jobs:
shell: bash
env:
PYTHONPATH: ${{ github.workspace }}
PTO2_RING_DEP_POOL: 524288
PTO2_RING_TASK_WINDOW: 524288
PTO2_RING_HEAP: 1073741824
PTO2_RING_DEP_POOL: 1048576
PTO2_RING_TASK_WINDOW: 1048576
PTO2_RING_HEAP: 4294967296
PYPTO_LOG_LEVEL: error
PYPTO_WARNING_LEVEL: none
run: |
Expand Down
2 changes: 1 addition & 1 deletion models/deepseek/v4/decode_csa.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,3 @@
# Copyright (c) PyPTO Contributors.
# This program is free software, you can redistribute it and/or modify it under the terms and conditions of
# CANN Open Software License Agreement Version 2.0 (the "License").
Expand Down Expand Up @@ -399,7 +399,7 @@

parser = argparse.ArgumentParser()
parser.add_argument("-p", "--platform", type=str, default="a2a3",
choices=["a2a3", "a2a3sim", "a5", "a5sim"])
choices=["a2a3", "a5"])
parser.add_argument("-d", "--device", type=int, default=0)
parser.add_argument("--layer-id", type=int, default=0)
parser.add_argument("--seed", type=int, default=0)
Expand Down
2 changes: 1 addition & 1 deletion models/deepseek/v4/decode_hca.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,3 @@
# Copyright (c) PyPTO Contributors.
# This program is free software, you can redistribute it and/or modify it under the terms and conditions of
# CANN Open Software License Agreement Version 2.0 (the "License").
Expand Down Expand Up @@ -347,7 +347,7 @@

parser = argparse.ArgumentParser()
parser.add_argument("-p", "--platform", type=str, default="a2a3",
choices=["a2a3", "a2a3sim", "a5", "a5sim"])
choices=["a2a3", "a5"])
parser.add_argument("-d", "--device", type=int, default=0)
parser.add_argument("--layer-id", type=int, default=0)
parser.add_argument("--seed", type=int, default=0)
Expand Down
Loading
Loading