Issues

• groupsize=128 gives wrong, non-deterministic results at Llama-7B shapes on sm_87 (Jetson Orin Nano)

  IST-DASLab/marlin#45 · Filter by author jayden1711 opened on Sep 29, 2026

• nvcc 13.1 ptxas miscompile on sm_90 drops createpolicy for cp.async cache_hint

  IST-DASLab/marlin#44 · Filter by author huangzhilin-hzl opened on May 1, 2026

• Any plan on AMD ROCm platform support?

  IST-DASLab/marlin#43 · Filter by author charyang-ai opened on Oct 29, 2025
  1

• Potential Async Data Race in cp_async4_stream Usage - Request for Validation

  IST-DASLab/marlin#42 · Filter by author Zephyreeze opened on Jul 24, 2025

• Question about the meaning of slice_count and slice_idx in global reduction condition

  IST-DASLab/marlin#41 · Filter by author ZhengChen6 opened on Jul 24, 2025

• Issue with running kernel on blackwell GPUs (RTX 5090)

  IST-DASLab/marlin#40 · Filter by author huoshenlaile opened on Mar 10, 2025
  2

• Different batch size results to different result for gptq_marlin_gemm kernel.

  IST-DASLab/marlin#38 · Filter by author rooooc opened on Dec 30, 2024

• Question about matrix A

  IST-DASLab/marlin#37 · Filter by author Zhao-Dongyu opened on Oct 23, 2024

• Marlin vs gguf

  IST-DASLab/marlin#36 · Filter by author blap opened on Sep 13, 2024
  1

• comparison with FastTransformer for BatchSize=1

  IST-DASLab/marlin#35 · Filter by author matrix97317 opened on Sep 11, 2024

• How to move the pointer of B and s in cuda kernel

  IST-DASLab/marlin#34 · Filter by author Eutenacity opened on Aug 22, 2024

• comparison with SmoothQuant

  IST-DASLab/marlin#32 · Filter by author lxww302 opened on Jul 29, 2024
  1

• RuntimeError: CUDA error: an illegal instruction was encountered when runing test.py

  IST-DASLab/marlin#30 · Filter by author MekkCyber opened on Jul 2, 2024
  2

• Trying to understand the kernel

  IST-DASLab/marlin#29 · Filter by author MekkCyber opened on Jun 27, 2024

• Why the linear input for Layer.pack "must be of type torch.half"?

  IST-DASLab/marlin#28 · Filter by author Azure-Tang opened on Jun 26, 2024
  4

• slight nondeterminism

  IST-DASLab/marlin#27 · Filter by author MichoChan opened on Jun 7, 2024
  2

• perfmance

  IST-DASLab/marlin#26 · Filter by author MichoChan opened on May 7, 2024

• Issues to generate tokens after "get_llama_marlin"

  IST-DASLab/marlin#23 · Filter by author HaoWuSR opened on Apr 19, 2024

• a_sh_rd_delta_o

  IST-DASLab/marlin#22 · Filter by author Lenan22 opened on Apr 12, 2024
  1

• Marlin slower than fp16 on larger batches

  IST-DASLab/marlin#21 · Filter by author mobicham opened on Apr 9, 2024
  2

• Questions about matrix A's layout in shared memory.

  IST-DASLab/marlin#20 · Filter by author HandH1998 opened on Apr 8, 2024

• questions about slice_col_par

  IST-DASLab/marlin#19 · Filter by author Lenan22 opened on Apr 7, 2024
  2

• [QST] Weight Format & GEMM

  IST-DASLab/marlin#18 · Filter by author jeromeku opened on Apr 1, 2024
  2

• groupsize=64 is not supported

  IST-DASLab/marlin#17 · Filter by author jameswu2014 opened on Mar 21, 2024
  1

• Do you have any plan support moe gemm?

  IST-DASLab/marlin#16 · Filter by author LinHR000 opened on Mar 13, 2024