Feat: Megatron-LM training backend with sparse weight sync#6
Closed
jsw-zorro wants to merge 3 commits into
Closed
Conversation
Add export_hf_named_params: a streaming generator that reconstructs the global model from Megatron's TP/PP/EP/ETP/VPP layout and yields HF-named, HF-layout CPU tensors one at a time (OOM-safe for large / MoE models). The gather + mcore->HF conversion is delegated to mbridge's export_weights (the same bridge the engine already uses to load/save); this module adds the consumer concerns: CPU move, byte-bounded bucketing, and a metadata-only path for transfer-buffer sizing. This is the foundation for correct sparse weight sync under full Megatron parallelism. The design (the "delta is computed in HF byte space" invariant) is documented in docs/en/architecture/megatron-weight-sync.md. The sglang Docker image gains the Megatron compiled deps it was missing: Transformer Engine (fused LayerNorm + sequence parallelism) and apex (optional fused LayerNorm/Adam). megatron-core / mbridge were already in the base install. Validated (exact bf16 match vs the HF reference checkpoint): - Qwen3-0.6B TP=2 310 tensors, 0 mismatch - Qwen3-0.6B PP=2 311 tensors, 0 mismatch - Qwen3-0.6B TP=2 PP=2 311 tensors, 0 mismatch - Qwen3-30B-A3B TP=2 EP=2 PP=2 18867 tensors, 0 mismatch
Replace the TP-only shard-direct weight transfer with the HF-export path: - MegatronEngine.export_hf_named_params() / get_hf_weight_metadata() stream gathered HF tensors via mbridge (handles TP/PP/EP/ETP/VPP). The previous PP>1 / EP>1 NotImplementedError guards are removed. - WeightManager gains "megatron_hf_meta" mode: the transfer buffer is sized for the full HF model and offload() streams HF tensors into the inactive half on the writer rank, while the gather collectives run on all ranks in lockstep. The sender receives megatron_metadata=None and runs the plain full/delta path used by FSDP. Because the buffer now holds HF-layout bytes, the sparse delta is computed in HF space and is correct under any parallelism — fixing the latent corruption where the delta was computed in mcore layout but applied by the receiver in HF layout. - ppo_trainer wires the generator + HF metadata through. The legacy CPU shard-reassembly in the sender agent is now unused for Megatron (kept only for the deprecated megatron_metadata path). Validated (buffer roundtrip == HF reference, bit-exact): - Qwen3-0.6B TP=2 310 tensors, 0 mismatch, 1.19 GB - Qwen3-0.6B TP=2 PP=2 311 tensors, 0 mismatch, 1.50 GB
Add examples/math/qwen3-8b-megatron-delta — the FSDP qwen3-8b-m2po-delta recipe with the trainer engine switched to the Megatron backend (backend: megatron, tensor_parallel_size: 4). Identical data, algorithm, and weight-transfer path, so it doubles as a clean FSDP-vs-Megatron A/B. End-to-end (4 RaaS + 4 trainer TP=4, delta TCP): 100 training steps, weight_transfer/delta_sparsity ~0.92 (HF-space delta), and a rising task reward (first-20 mean 0.55 -> later windows ~0.61-0.63).
jsw-zorro
force-pushed
the
feat/megatron-weight-sync
branch
from
d52b6ad to
b6d0ce3
Compare
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.
Adds Megatron-LM as a training backend with correct sparse (delta) weight
synchronization to RaaS under full parallelism (TP/PP/EP/ETP/VPP).
What
export_hf_named_params: streaming Megatron→HF per-tensor weight export(OOM-safe), delegating the gather + conversion to mbridge.
WeightManagerHF-export mode: the transfer buffer holds HF-layout bytes,so the sparse delta is computed in HF space — correct under any
parallelism. Removes the PP>1/EP>1 restrictions.
docker/Dockerfile.sglang: adds Transformer Engine + apex (the Megatroncompiled deps).
examples/math/qwen3-8b-megatron-delta(Qwen3-8B, TP=4, delta).docs/en/architecture/megatron-weight-sync.md.Validation
Qwen3-30B-A3B MoE TP2/EP2/PP2 (18,867 tensors, 0 mismatch).