Add ROCm support for CI test_qwen3_4B_ckpt#1621
Open
sreerohi wants to merge 1 commit into
Open
Conversation
tests/e2e/ckpt/test_qwen3_4B_ckpt.py:
Added IS_ROCM detection (os.path.exists("/opt/rocm")) and
if-else conditions in execute() for ROCm-specific args.
Uses async 4+4 mode (4 GPUs train, 4 GPUs rollout) with
train_async.py, --attention-backend auto, optimizer CPU offload,
--sglang-disable-custom-all-reduce, and ROCm env vars.
NVIDIA path remains unchanged.
miles/utils/external_utils/command_utils.py:
Added ROCm NVTE env var cleanup (unset NVTE_FUSED_ATTN,
NVTE_FLASH_ATTN, NVTE_UNFUSED_ATTN) to convert_checkpoint(),
matching the existing cleanup in execute_train().
miles/backends/training_utils/ci_utils.py:
Made KL divergence CI threshold ROCm-conditional: 1e-8 on ROCm
(torch.version.hip), 1e-10 on NVIDIA. ROCm floating point
rounding produces slightly different near-zero values.
miles/backends/megatron_utils/ci_utils.py:
Made peak GPU memory CI threshold ROCm-conditional: 40 GB on
ROCm, 20 GB on NVIDIA. The ROCm config uses TP=1, CP=1 so
each GPU holds the full model (~34 GB vs ~17 GB with TP=2, CP=2).
sreerohi
requested review from
Shi-Dong,
Zhichenzzz,
fzyzcjy,
guapisolo,
jybsuper,
maocheng23,
yueming-yuan and
yushengsu-thu
as code owners
Contributor
|
Warning Gemini encountered an error creating the review. You can try again by commenting |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Added
IS_ROCMdetection and if-else conditions inexecute()for ROCm-specific args. Uses async 4+4 mode (4 GPUs train, 4 GPUs rollout) withtrain_async.py,--attention-backend auto, optimizer CPU offload,--sglang-disable-custom-all-reduce, and ROCm env vars. NVIDIA path remains unchanged.unset NVTE_FUSED_ATTN,NVTE_FLASH_ATTN,NVTE_UNFUSED_ATTN) toconvert_checkpoint()andexecute_train()incommand_utils.py.ci_utils.py.Related to Miles CI gap Between ROCm & CUDA #1105
Supersedes Add ROCm support for CI test_qwen3_4B_ckpt #1118