[Common][PyTorch] EP dispatch with unfused MXFP8 quantization

[phu0ngng]

Description

This PR adds MXFP8 support to the dispatch op of the NCCL EP path. The dispatch op is used in two places, and MXFP8 applies to both:

• Dispatch forward bfloat16 tokens are quantized to MXFP8 internally and dispatched to the target experts; recv is returned as a per-expert GroupedTensor.
• Combine backward the result-grad is scattered back to expert positions through the same (reverse) dispatch op, quantized to MXFP8, returning the expert-output grad as a per-expert GroupedTensor.

Type of change

• Documentation change (change only to the documentation, either a fix or a new content)
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Infra/Build change
• Code refactoring

Changes

PyTorch frontend (transformer_engine/pytorch/ep.py, distributed.py, csrc/extensions/ep.cpp)**

• The dispatch op quantizes bfloat16 tokens to MXFP8 internally when the buffer's dispatch_quant_recipe is set (MXFP8BlockScaling only for now); dispatch-forward recv is returned as a per-expert GroupedTensor. A pre-quantized input is rejected.
• Combine backward reuses the dispatch op to scatter the result-grad: it quantizes the grad to MXFP8 and returns the expert-output grad as a per-expert GroupedTensor. Combine forward is unchanged (high-precision).
• Recv data and block scales share a single caller-supplied (optionally symm-mem-backed) buffer, sliced into data-then-scale regions; the same convention is used for the combine backward grad buffer.

Common backend (common/ep/ep_backend.cpp, include/.../ep.h, comm_window.h)**

• Backend and public headers extended to carry block-scale buffers/windows through the dispatch primitive.

NCCL EP submodule**

• Bumped 3rdparty/nccl-extensions to the revision providing block-scaled dispatch.

Tests (tests/cpp_distributed/test_ep.cu, tests/pytorch/distributed/run_ep.py, run_test_ep.sh)**

• Added C++ distributed coverage for the MXFP8 dispatch path.
• Added PyTorch MXFP8 test passes for dispatch forward (normal, zero-copy, eager IO modes) and combine backward, gated behind a dedicated NVTE_EP_MXFP8_PASS run since the grouped path pins the per-expert alignment process-wide.

Checklist:

• I have read and followed the contributing guidelines
• The functionality is complete
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[greptile-apps]

Greptile Summary

Adds unfused MXFP8 quantization support to NCCL expert-parallel dispatch and combine backward.

• Routes MXFP8 data and block scales through common and PyTorch dispatch interfaces.
• Returns per-expert quantized outputs as grouped tensors and supports caller-provided or symmetric-memory-backed buffers.
• Extends distributed tests for normal, zero-copy, eager, and combine-backward paths.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains, and the previously reported ignore-file deletion has been fully restored in the current tree.

Important Files Changed

Filename	Overview
transformer_engine/pytorch/ep.py	Adds MXFP8 recipe handling, compact scale-buffer allocation, grouped output construction, and quantized combine-backward dispatch.
transformer_engine/pytorch/csrc/extensions/ep.cpp	Extends PyTorch bindings to validate, describe, and route MXFP8 data and scale-inverse tensors.
transformer_engine/common/ep/ep_backend.cpp	Adds NCCL descriptors and dispatch configuration for block-scaled token data and scales.
transformer_engine/pytorch/distributed.py	Adds explicit symmetric-memory pool lifecycle and cache cleanup needed by EP buffers.
tests/pytorch/distributed/run_ep.py	Adds MXFP8 dispatch and combine-backward coverage across fixed, eager, caller-buffer, and zero-copy modes.

Sequence Diagram

sequenceDiagram
  participant Caller
  participant PyEP as PyTorch EP
  participant Quant as MXFP8 Quantizer
  participant Backend as NCCL EP Backend
  participant Expert
  Caller->>PyEP: ep_dispatch(BF16 tokens)
  PyEP->>Quant: quantize data and block scales
  Quant-->>PyEP: E4M3 data + E8M0 scales
  PyEP->>Backend: dispatch data and scales
  Backend-->>Expert: per-expert GroupedTensor
  Expert->>PyEP: ep_combine(expert output)
  PyEP-->>Caller: high-precision combined result
  Caller->>PyEP: result gradient
  PyEP->>Quant: quantize gradient
  PyEP->>Backend: reverse dispatch data and scales
  Backend-->>Expert: grouped expert-output gradient

Loading

_{Reviews (10): Last reviewed commit: "Merge branch 'main' into ep_mxfp8" | Re-trigger Greptile}

[phu0ngng]

/te-ci L1 pytorch

[phu0ngng]

/te-ci L1

[phu0ngng]

/te-ci L1

[phu0ngng]

/te-ci L1

[YangFei1990]

Need to further align on API contracts before merging

[phu0ngng]

/te-ci L1

[phu0ngng]

/te-ci L1