[PR #454] Add SM90 NVFP4+FP8 MegaMoE support #460

New issue

Open

opened 2026-10-26 21:18:06 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 21:18:06 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepGEMM/pull/454
Author: @lengrongfu
Created: 9/21/2026
Status: 🔄 Open

Base: nv_dev ← Head: feat/sm90-nvfp4-fp8-megamoe

📝 Commits (1)

d5925de Add SM90 NVFP4+FP8 MegaMoE support

📊 Changes

18 files changed (+4828 additions, -33 deletions)

View changed files

📝 csrc/apis/mega.hpp (+197 -18)
📝 csrc/jit/handle.hpp (+5 -2)
➕ csrc/jit_kernels/heuristics/sm90_nvfp4_mega_moe.hpp (+185 -0)
➕ csrc/jit_kernels/impls/sm90_nvfp4_mega_moe_fused.hpp (+259 -0)
📝 deep_gemm/__init__.py (+2 -0)
📝 deep_gemm/include/deep_gemm/comm/barrier.cuh (+20 -0)
📝 deep_gemm/include/deep_gemm/common/math.cuh (+9 -1)
➕ deep_gemm/include/deep_gemm/impls/sm90_nvfp4_mega_moe_fused.cuh (+2310 -0)
📝 deep_gemm/include/deep_gemm/layout/mega_moe.cuh (+19 -4)
📝 deep_gemm/include/deep_gemm/ptx/tma.cuh (+7 -2)
➕ deep_gemm/include/deep_gemm/quantization/nvfp4_dequant.cuh (+236 -0)
📝 deep_gemm/include/deep_gemm/scheduler/mega_moe.cuh (+350 -0)
📝 deep_gemm/mega/__init__.py (+115 -0)
➕ deep_gemm/quantization_nvfp4.py (+185 -0)
📝 scripts/generate_pyi.py (+6 -6)
➕ scripts/run_ncu_mega_moe_sm90.sh (+94 -0)
➕ tests/bench_nvfp4_mega_moe_sm90.py (+211 -0)
➕ tests/test_nvfp4_mega_moe_sm90_correctness.py (+618 -0)

📄 Description

Add support use nvfp4 mega_moe in sm90.

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/deepseek-ai/DeepGEMM/pull/454 **Author:** [@lengrongfu](https://github.com/lengrongfu) **Created:** 9/21/2026 **Status:** 🔄 Open **Base:** `nv_dev` ← **Head:** `feat/sm90-nvfp4-fp8-megamoe` --- ### 📝 Commits (1) - [`d5925de`](https://github.com/deepseek-ai/DeepGEMM/commit/d5925deb53a26895efad49395831f0bcf30d13b1) Add SM90 NVFP4+FP8 MegaMoE support ### 📊 Changes **18 files changed** (+4828 additions, -33 deletions) <details> <summary>View changed files</summary> 📝 `csrc/apis/mega.hpp` (+197 -18) 📝 `csrc/jit/handle.hpp` (+5 -2) ➕ `csrc/jit_kernels/heuristics/sm90_nvfp4_mega_moe.hpp` (+185 -0) ➕ `csrc/jit_kernels/impls/sm90_nvfp4_mega_moe_fused.hpp` (+259 -0) 📝 `deep_gemm/__init__.py` (+2 -0) 📝 `deep_gemm/include/deep_gemm/comm/barrier.cuh` (+20 -0) 📝 `deep_gemm/include/deep_gemm/common/math.cuh` (+9 -1) ➕ `deep_gemm/include/deep_gemm/impls/sm90_nvfp4_mega_moe_fused.cuh` (+2310 -0) 📝 `deep_gemm/include/deep_gemm/layout/mega_moe.cuh` (+19 -4) 📝 `deep_gemm/include/deep_gemm/ptx/tma.cuh` (+7 -2) ➕ `deep_gemm/include/deep_gemm/quantization/nvfp4_dequant.cuh` (+236 -0) 📝 `deep_gemm/include/deep_gemm/scheduler/mega_moe.cuh` (+350 -0) 📝 `deep_gemm/mega/__init__.py` (+115 -0) ➕ `deep_gemm/quantization_nvfp4.py` (+185 -0) 📝 `scripts/generate_pyi.py` (+6 -6) ➕ `scripts/run_ncu_mega_moe_sm90.sh` (+94 -0) ➕ `tests/bench_nvfp4_mega_moe_sm90.py` (+211 -0) ➕ `tests/test_nvfp4_mega_moe_sm90_correctness.py` (+618 -0) </details> ### 📄 Description Add support use nvfp4 mega_moe in sm90. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi added the

pull-request

label

2026-10-26 21:18:06 +00:00

bsakaguchi referenced this issue

2026-10-26 21:18:08 +00:00

[PR #460] [Fix] Bound cuBLASLt descriptor lifetime on errors #464