Skip to content

fix(ci): pin linear CUDA-graph sizing distribution for gpt_grpo_*_8b_throughput #5242

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

Closed
Connor-XY wants to merge 1 commit into from
Closed

fix(ci): pin linear CUDA-graph sizing distribution for gpt_grpo_*_8b_throughput #5242

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
4 changes: 4 additions & 0 deletions ...onal_tests/test_cases/gpt/gpt_grpo_tp4_pp1_dp2_8b_cudagraphs_throughput/model_config.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,10 @@ MODEL_ARGS:
--inference-dynamic-batching-unified-memory-level: 1
--inference-dynamic-batching-buffer-size-gb: 20
--ckpt-format: torch_dist
# Pin the pre-#3509 (linear) CUDA-graph sizing distribution. #3509 switched the
# default to exponential + a mixed-prefill grid, which raised peak memory and
# tripped this test's mem-allocated-bytes guardrail (~69GB vs ~61GB golden, +13.6%).
--inference-dynamic-batching-cuda-graph-sizing-distribution: linear
--seq-length: 1024
--inference-max-seq-length: 1024
--load: ${CHECKPOINT_LOAD_PATH}/model/qwen3-8b-dist
Expand Down
4 changes: 4 additions & 0 deletions tests/functional_tests/test_cases/gpt/gpt_grpo_tp4_pp1_dp2_8b_throughput/model_config.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,10 @@ MODEL_ARGS:
--inference-dynamic-batching-unified-memory-level: 1
--inference-dynamic-batching-buffer-size-gb: 20
--ckpt-format: torch_dist
# Pin the pre-#3509 (linear) CUDA-graph sizing distribution. #3509 switched the
# default to exponential + a mixed-prefill grid, which raised peak memory and
# tripped this test's mem-allocated-bytes guardrail (~69GB vs ~61GB golden, +13.6%).
--inference-dynamic-batching-cuda-graph-sizing-distribution: linear
--seq-length: 1024
--inference-max-seq-length: 1024
--load: ${CHECKPOINT_LOAD_PATH}/model/qwen3-8b-dist
Expand Down
Loading