Support Flashinfer-cutedsl nvfp4 grouped masked gemm

[mpgemm]

Motivation

支持 flashinfer-cutedsl NVFP4 FusedMoE 计算。

Modifications

1.新增 /moe/flashinfer_cutedsl_moe.py 用于引入 flashinfer-cutedsl nvfp4 group gemm 支持 FFN 计算。

2.在 /quantization/nvfp4.py 新增 ModelOptNvFp4FusedMoECuteDSL 以支持Nvfp4FusedMoE 计算。

3.新增测试文件 tests/layer/test_cutedsl_moe.py 和 tests/layers/test_nvfp4_fusedmoe.py

Usage or Command

Paddle Flashinfer 和 nvidia-cutlass-dsl 存在问题，导入时需要修改 python3.10/site-packages/flashinfer 和 nvidia-dsl。
总结出了三个问题，1个nvidia-dsl和2个flashinfer。

1：nvidia_cutlass_dsl/python_packages/cutlass/torch.py 将 torch.device 改成 "torch.device"。
2：flashinfer/utils.py. get_compute_capability函数下面改成：
@functools.cache
def get_compute_capability(device: torch.device) -> Tuple[int, int]:
return torch.cuda.get_device_capability(device)
if device.type != "cuda":
raise ValueError("device must be a cuda device")
return torch.cuda.get_device_capability(device.index)
注：如果遇到device的问题，将 A.place 换成 A.device 可以解决大部分问题。
3：flashinfer/cute_dsl/blockscaled_gemm.py
首先 import cuda.bindings.driver as cuda
然后将 cutlass_torch.current_stream() 替换成 cuda.CUstream(torch.cuda.current_stream().stream_base.raw_stream)

---

Flashinfer-cutedsl nvfp4 grouped mask gemm 算子测试： python tests/layer/test_cutedsl_moe.py

decode误差测试：python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tests/layers/test_nvfp4_fusedmoe.py TestFusedMoE.test_decode_correctness 2>&1

Prefill测试: NVFP4_TEST_MODE=prefill NVFP4_TEST_ITERS=2 python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tests/layers/test_nvfp4_fusedmoe.py

Decode测试： NVFP4_TEST_MODE=decode NVFP4_TEST_ITERS=2 python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tests/layers/test_nvfp4_fusedmoe.py

Accuracy Tests

Checklist

• Add at least a tag in the PR title.
  • Tag list: [[FDConfig],[APIServer],[Engine], [Scheduler], [PD Disaggregation], [Executor], [Graph Optimization], [Speculative Decoding], [RL], [Models], [Quantization], [Loader], [OP], [KVCache], [DataProcessor], [BugFix], [Docs], [CI], [Optimization], [Feature], [Benchmark], [Others], [XPU], [HPU], [GCU], [DCU], [Iluvatar], [Metax]]
  • You can add new tags based on the PR content, but the semantics must be clear.
• Format your code, run pre-commit before commit.
• Add unit tests. Please write the reason in this PR if no unit tests.
• Provide accuracy results.
• If the current PR is submitting to the release branch, make sure the PR has been submitted to the develop branch, then cherry-pick it to the release branch with the [Cherry-Pick] PR tag.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[paddle-bot]

Thanks for your contribution!

[mpgemm]

support flashinfer-cutedsl nvfp4 fusedmoe， prefill仍然存在问题

[lizexu123]

你这个函数写了，但是apply里面没用到是为什么？