[megatron] fix: forward mhc_multistream to MTP and skip activation reclaim for MTP checkpoints

[HollowMan6]

What does this PR do?

Fixes two crashes that block mHC (manifold hyper-connections) + MTP training on Megatron-core (DeepSeek-V4, use_fused_mhc=False + mtp.enable=True). Both stem from verl's MTP postprocess / recomputation patches not accounting for the mHC multi-stream tensor.

Background: With mHC + MTP, the decoder returns a tuple (contracted_hidden, mhc_multistream) where mhc_multistream is the pre-contraction [s, b, n*h] multi-stream tensor. GPTModel._postprocess forwards mhc_multistream into self.mtp(...), and the MTP module branches on it (if mhc_multistream is not None) to use the multi-stream tensor as its depth input. mhc_multistream is a plain alias of the decoder output (transformer_block.py: mhc_multistream = hidden_states), and the MTP depth input is a torch.chunk view of it — i.e. shared storage, not a copy.

Bug 1 — reshape crash (forward): verl's patched _megatron_gptmodel_postprocess (mtp_patch.py) accepted mhc_multistream as a parameter but never forwarded it into self.mtp(...), so it arrived as None inside MTP. MTP then took the contracted [s, b, h] path while the mHC branch of _concat_embeddings expected [s, b, n*h], failing at:

RuntimeError: shape '[23986, 1, 4, 4096]' is invalid for input of size 98246656

Fix: forward mhc_multistream into the self.mtp(...) call, matching the upstream GPTModel._postprocess.

Bug 2 — async CUDA illegal-memory-access (backward): verl's recomputation backward patch (apply_patch_megatron_recomputation_backward, commit 04df110, a MoE residual-memory leak fix) reclaims saved activation memory by calling t.untyped_storage().resize_(0) on every checkpoint input after computing grads. With mHC + MTP this is a use-after-free: the MTP checkpoint's saved hidden_states is a torch.chunk view of the decoder's mhc_multistream (shared storage via make_viewless_tensor → _kernel_make_viewless_tensor doing out.data = inp.data), and the MTP checkpoint backward runs before the decoder's learned_output_contract backward (reverse topological order). resize_(0) therefore truncates storage the pending decoder backward still reads → async CUDA illegal-memory-access, surfacing at the next sync point (set_state inside _fork_rng.__exit__).

Fix: detect MTP-layer checkpoints by inspecting the MultiTokenPredictionLayer instance captured in ctx.run_function.__closure__ (its __qualname__ is ..._checkpointed_forward.<locals>.custom_forward, which carries no class name), and skip the resize_(0) reclaim for those. Non-MTP checkpoints keep the reclaim, preserving the original MoE leak fix.

Checklist Before Starting

• Search for similar PRs. Paste at least one query link here: ...
• Format the PR title as [{modules}] {type}: {description} (This will be checked by the CI)
  • {modules} include fsdp, megatron, veomni, sglang, vllm, rollout, trainer, ci, training_utils, recipe, hardware, deployment, ray, worker, single_controller, misc, perf, model, algo, env, tool, ckpt, doc, data, cfg, reward, fully_async, one_step_off
  • If this PR involves multiple modules, separate them with , like [megatron, fsdp, doc]
  • {type} is in feat, fix, refactor, chore, test
  • If this PR breaks any API (CLI arguments, config, function signature, etc.), add [BREAKING] to the beginning of the title.
  • Example: [BREAKING][fsdp, megatron] feat: dynamic batching

Test

For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc.

API and Usage Example

Demonstrate how the API changes if any, and provide usage example(s) if possible.

# Add code snippet or script demonstrating how to use this

Design & Code Changes

Demonstrate the high-level design if this PR is complex, and list the specific changes.

Checklist Before Submitting

Important

Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review.

• Read the Contribute Guide.
• Apply pre-commit checks: pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always
• Add / Update the documentation.
• Add unit or end-to-end test(s) to the CI workflow to cover all the code. If not feasible, explain why: ...
• Once your PR is ready for CI, send a message in the ci-request channel in the verl Slack workspace. (If not accessible, please try the Feishu group (飞书群).)
• If your PR is related to the recipe submodule, please also update the reference to the submodule commit via git submodule update --remote or cd recipe && git pull origin main.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.