Issues

• [Bug] DeepGEMM: 32-bit scale indices wrap in pack_fp32_into_ue8m0 (reads) and transpose_fp32 (writes)

  deepseek-ai/DeepGEMM#490 · Filter by author xiaoqiangheye opened on Nov 1, 2026

• [RFC] Native triangular SYRK for batched_syrk (Muon Newton-Schulz)

  deepseek-ai/DeepGEMM#487 · Filter by author MeKaustubh07 opened on Oct 29, 2026

• [Green Context MoE] Introduce GC MoE

  deepseek-ai/DeepGEMM#484 · Filter by author Bruce-Lee-LY opened on Oct 26, 2026

• Please enable private vulnerability reporting / add a security contact

  deepseek-ai/DeepGEMM#481 · Filter by author Sx4DoW opened on Oct 23, 2026

• [Bug] SM100 HC-prenorm TMEM deallocation fault

  deepseek-ai/DeepGEMM#479 · Filter by author henrylhtsang opened on Oct 21, 2026

• [Bug] SM90 fp8_einsum("bhr,hdr->bhd", a, b, d, c) silently drops c and returns a @ b.T

  deepseek-ai/DeepGEMM#476 · Filter by author arxgy opened on Oct 17, 2026

• Update CUTLASS to 4.8.0 for SM107 (Rubin) support

  deepseek-ai/DeepGEMM#461 · Filter by author lisawym opened on Sep 28, 2026
  1

• [Bug] non-deterministic results in sm90 tf32_hc_prenorm_gemm

  deepseek-ai/DeepGEMM#449 · Filter by author leevan opened on Sep 16, 2026
  4

• [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

• 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 HangruiCao2000 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