tests/mimo-7B: switch ROCm path to async 4+4 to avoid 8-rank RCCL hang and no gradient accumulation fusion#1615
Open
sreerohi wants to merge 2 commits into
Open
Conversation
- Topology: drop --colocate, use train_async.py with 4 train + 4 rollout GPUs (TP=1, CPU-offload optimizer). - 8-rank Megatron world group deadlocks in RCCL all_gather_object inside _get_param_groups on MI350; splitting into two 4-rank groups sidesteps it, and TP=1 + CPU-offload fits the 7B optimizer state per GPU. - Args: --attention-backend auto, --no-gradient-accumulation-fusion, --update-weights-interval 2, --sglang-disable-custom-all-reduce. - --no-gradient-accumulation-fusion dodges a hipBLASLt BGRADB+accumulate wgrad failure on gfx950 in TE's LayerNormLinear backward - Env: SGLANG_SET_CPU_AFFINITY=0, RAY_EXPERIMENTAL_NOSET_HIP_VISIBLE_DEVICES=1, HIP_VISIBLE_DEVICES=0..7. Depends on PR radixark#1118 (command_utils ROCm env handling) and PR radixark#1153 (honor --no-gradient-accumulation-fusion).
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Rocm testcase changes: Drop --colocate, use train_async.py with 4 train + 4 rollout GPUs (TP=1, CPU-offload optimizer).
Supersedes tests/mimo-7B: switch ROCm path to async 4+4 to avoid 8-rank RCCL hang and no gradient accumulation fusion #1163