[pull] main from inclusionAI:main#38
Merged
pull[bot] merged 4 commits intoaxistore80-coder:mainfrom Apr 16, 2026
Merged
Conversation
#1181) * fix: update weights on disk for fsdp, megatron, and archon engine --------- Co-authored-by: Mohammad Asiful Hossain (A) <m00802411@china.huawei.com>
…blueprint (#1179) * refactor: mount data blueprint via ASGI and implement engine Pydantic models * fix: correct typo in set_env attribute name * test: add RPC engine validation tests and apply NonEmptyStr constraints
…load (#1182) When memory_efficient_load=True, all ranks previously created the full model on CPU via from_config(), causing CPU OOM on nodes with limited RAM (e.g. 8 workers × 64GB = 512GB on a 256GB node). Now only rank-0 loads on CPU; other ranks use meta device (zero memory cost). Weights are broadcast from rank-0 after FSDP sharding via fsdp2_load_full_state_dict with broadcast_from_rank0=True. Also defer AutoProcessor import in saver.py and recover.py to avoid importing torchvision eagerly (which can crash when torchvision version mismatches torch). Key changes: - fsdp_engine.py: non-rank-0 uses "meta" device in memory_efficient_load - fsdp_utils/__init__.py: use to_empty() for meta→device conversion - saver.py, recover.py: move AutoProcessor to TYPE_CHECKING block
) * ci: parallelize unit and integration tests across 4 GPU instances Split the CI matrix from 2 jobs (sglang, vllm) to 4 by adding a test_type dimension (unit, integration). Each variant now provisions two separate GCP runners that execute in parallel, reducing overall wall-clock time. Key changes: - Add test_type matrix dimension to provision-runner, unit-tests, cleanup jobs - Add conditional steps: unit tests run only on test_type=unit, SFT+GRPO on integration - Update runner labels and instance names to include test_type for isolation * fix(test): use correct Content-Type in data proxy rtensor tests The tests sent application/octet-stream but the body was JSON-serialized data. FastAPI passed raw bytes instead of parsing JSON, causing 'bytes' object has no attribute 'nbytes' in rtensor._store_local. Key changes: - Change Content-Type from application/octet-stream to application/json - Clear _storage_stats in test fixture to prevent cross-test leaks * fix(test): fix test_train_engine failures in dcp save/load and device model tests Add missing @torch.no_grad() to test_dcp_save_load_weights to prevent in-place param.zero_() error on grad-requiring leaf tensors, matching the existing decorator on test_hf_save_load_weights. Mock dist.get_rank in test_create_device_model_applies_use_kernels to handle the new rank check added by f34bea8 (memory_efficient_load meta device optimization) when distributed is not initialized. * fix(infra): return 400 for non-dict JSON in /data/batch endpoint Validate that raw_payload is a dict before unpacking into BatchShardRequest. Non-dict JSON bodies (e.g. arrays) raised TypeError instead of ValidationError, bypassing the 400 handler and falling through to the 500 catch-all.
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 subscribe to this conversation on GitHub.
Already have an account?
Sign in.
4 participants
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.
See Commits and Changes for more details.
Created by
pull[bot] (v2.0.0-alpha.4)
Can you help keep this open source service alive? 💖 Please sponsor : )