Skip to content

fix: resolve NPU OOM with default training config #620

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

Open
curnane-lab wants to merge 11 commits into
base: main
Choose a base branch
from
Open

fix: resolve NPU OOM with default training config #620

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
a293764
fix(dflash/domino): use mask_token_id from dflash_config
mingliangfu Jun 26, 2026
2e7ee2a
Merge branch 'main' into domino_npu
curnane-lab Jun 28, 2026
9aaafe5
Merge branch 'sgl-project:main' into domino_npu
curnane-lab Jun 29, 2026
57b7226
fix: resolve OOM with default training config
mingliangfu Jun 29, 2026
5132dd9
update default domino training config
mingliangfu Jun 29, 2026
cc38def
Merge branch 'main' into domino_npu
curnane-lab Jun 30, 2026
e06d4ba
Merge branch 'main' into domino_npu
curnane-lab Jun 30, 2026
e935695
fix(npu): keep default num anchors in examples
mingliangfu Jul 1, 2026
7bee8ef
style(npu): simplify OOM anchor hints
mingliangfu Jul 1, 2026
7fdc7c4
Merge branch 'main' into domino_npu
curnane-lab Jul 1, 2026
2f4d6d4
Merge branch 'sgl-project:main' into domino_npu
curnane-lab Jul 3, 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
1 change: 1 addition & 0 deletions examples/run_qwen3.5_4b_dflash_online_npu.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,7 @@ export SPECFORGE_DATA_NUM_PROC=64
ATTENTION_BACKEND=${2:-flex_attention}
NUM_GPUS=${1:-8}

# If OOM occurs on low-memory NPU devices, retry with a smaller --num-anchors value.
torchrun \
--standalone \
--nproc_per_node $NUM_GPUS \
Expand Down
3 changes: 2 additions & 1 deletion examples/run_qwen3.5_4b_domino_online_npu.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,7 @@ NUM_DEVICES=$(echo "$NPU_DEVICES" | tr ',' '\n' | wc -l)
export ASCEND_RT_VISIBLE_DEVICES=$NPU_DEVICES
export PYTORCH_NPU_ALLOC_CONF=max_split_size_mb:32

# If OOM occurs on low-memory NPU devices, retry with a smaller --num-anchors value.
torchrun \
--standalone \
--nproc_per_node "$NUM_DEVICES" \
Expand All @@ -37,7 +38,7 @@ torchrun \
--max-length 1024 \
--chat-template qwen3.5 \
--attention-backend sdpa \
--num-anchors 16 \
--num-anchors 256 \
--loss-decay-gamma 7.0 \
--log-interval 50 \
--save-interval 3000 \
Expand Down
Loading