Issues

• Update CUTLASS to 4.8.0 for SM107 (Rubin) support

  deepseek-ai/DeepGEMM#461 · Filter by author lisawym opened on Sep 28, 2026

• [Bug] non-deterministic results in sm90 tf32_hc_prenorm_gemm

  deepseek-ai/DeepGEMM#449 · Filter by author leevan opened on Sep 16, 2026
  3

• [Bug] SM120 m_grouped_fp8 contiguous kernel: illegal memory access at CUDA graph replay (eager + capture + identical-replay all clean)

  deepseek-ai/DeepGEMM#444 · Filter by author liaoruoxue opened on Sep 14, 2026
  3

• DeepGEMM JIT: CUDA 800 (NOT_PERMITTED) at jit/handle.hpp:154 kills vLLM engine on first long-context request

  deepseek-ai/DeepGEMM#443 · Filter by author RenZov-A opened on Sep 14, 2026
  1

• [nv_dev][SM120] fp8_fp4_mqa_logits fp8 path returns nondeterministic in-window wrong results for head_dim 32/64 — read-side swizzle mode doesn't match the TMA descriptors

  deepseek-ai/DeepGEMM#433 · Filter by author aganhui opened on Sep 10, 2026
  4

• [nv_dev][SM120] k_grouped_fp8_gemm_nt_contiguous returns non-deterministic wrong results — sm120 kernel is missing the tensormap drain fixed for SM90 in #343

  deepseek-ai/DeepGEMM#430 · Filter by author aganhui opened on Sep 9, 2026
  1

• sm121a: paged_mqa_logits varlen path produces illegal memory access; uniform path does not scale past batch 1 at long context

  deepseek-ai/DeepGEMM#425 · Filter by author srinathh opened on Aug 31, 2026
  3

• [Regression] 8b1392b removes SM12x pure-fp8 1d1d kernels and aliases fp8_gemm_nt to the fp4 dispatcher (silent corruption on GB10)

  deepseek-ai/DeepGEMM#417 · Filter by author maci0 opened on Aug 24, 2026
  4

• [Bug] pack_ue8m0_to_int is not CUDA Graph capture safe

  deepseek-ai/DeepGEMM#414 · Filter by author foraxe opened on Aug 24, 2026

• get_env<int> returns uninitialized data for malformed values

  deepseek-ai/DeepGEMM#412 · Filter by author OllieinCanada opened on Aug 24, 2026
  1

• [RFC] Chunk-ready symmetric-buffer protocol for streaming MegaMoE inputs

  deepseek-ai/DeepGEMM#410 · Filter by author 0z5a opened on Aug 17, 2026
  1

• CUDA Illegal Instruction in test_k_grouped_gemm_contiguous

  deepseek-ai/DeepGEMM#405 · Filter by author henny-bee opened on Aug 12, 2026
  2

• Measured data point on fixed-tile vs shape-adaptive GEMM: where a bit-exact fixed-tile kernel stops (and why we couldn't fix it)

  deepseek-ai/DeepGEMM#402 · Filter by author dancinlife opened on Aug 7, 2026

• Question About deepGEMM contributor and Collaboration

  deepseek-ai/DeepGEMM#391 · Filter by author DiegoCao opened on Jul 22, 2026

• Origin & calibration methodology of the three cost-model constants in sm120.hpp

  deepseek-ai/DeepGEMM#385 · Filter by author json0701 opened on Jul 17, 2026
  1

• [DSv4] IS W4A16(mxfp4, act bf16) on the planning list?

  deepseek-ai/DeepGEMM#374 · Filter by author heroes999 opened on Jul 6, 2026

• transform_sf_into_required_layout: no SM120/121 (arch_major=12) branch for (gran_mn=1, gran_k=32) NVFP4 expert scales -- blocks DeepSeek-V4 weight loading

  deepseek-ai/DeepGEMM#372 · Filter by author zbrad opened on Jul 6, 2026
  1

• Incompatibility between DeepEP V2 ElasticBuffer non-expanded output and DeepGEMM masked/contiguous grouped GEMM input format

  deepseek-ai/DeepGEMM#370 · Filter by author litiantian00 opened on Jul 1, 2026

• question about tmem barrier

  deepseek-ai/DeepGEMM#368 · Filter by author wukong1992 opened on Jun 26, 2026
  1

• question about fp8fp4 mega moe weights load

  deepseek-ai/DeepGEMM#367 · Filter by author wukong1992 opened on Jun 25, 2026
  1

• [BUG] Non-deterministic numerical problem of k_grouped_fp8_gemm_nt_contiguous

  deepseek-ai/DeepGEMM#365 · Filter by author FFGGSSJJ opened on Jun 23, 2026
  3

• [BUG] NVRTC build fails for sm90_tf32_hc_prenorm_gemm — cudaGridDependencySynchronize undefined (PDL intrinsic not available under NVRTC)

  deepseek-ai/DeepGEMM#359 · Filter by author zhehli688 opened on Jun 12, 2026

• [Feature request] Documented clangd setup for CUDA kernel indexing

  deepseek-ai/DeepGEMM#358 · Filter by author Rachmanino opened on Jun 12, 2026

• Does Mega MoE support RTX 50-series GPUs, and can it work in a PCIe-only setup without NVLink?

  deepseek-ai/DeepGEMM#356 · Filter by author wukong1992 opened on Jun 10, 2026
  1

• [RFC] Support FP4 (E2M1) activation dispatch in MegaMoE

  deepseek-ai/DeepGEMM#338 · Filter by author wuudiz opened on May 19, 2026