Search results

Use pyproject.toml

Status: Open.

#172 In deepseek-ai/FlashMLA;

· sakgoyal opened on Mar 27, 2026

CUTLASS Internal Error during run on SM100 with long sequences (seq_len=1M)

Status: Open.

#171 In deepseek-ai/FlashMLA;

· For-rest2005 opened on Mar 20, 2026

dual gemm

Status: Open.

#169 In deepseek-ai/FlashMLA;

· WalrusWTQ opened on Mar 6, 2026

flash_mla.flash_mla_with_kvcache have plans to support FP8 for the q data in the future or FP8 insufficient to support model accuracy?

Status: Open.

#168 In deepseek-ai/FlashMLA;

· zhangfengwei opened on Mar 6, 2026

[Question]Questions regarding INTERLEAVE vs. SW128 Layouts for SM90 Sparse Attention Decode

Status: Open.

#166 In deepseek-ai/FlashMLA;

· pengwubj opened on Feb 25, 2026

[Question] why not using FP8 rope in sparse FP8 decoding?

Status: Open.

#165 In deepseek-ai/FlashMLA;

· lyppg opened on Feb 18, 2026

mla

Status: Open.

#164 In deepseek-ai/FlashMLA;

· TZWX-0 opened on Feb 12, 2026

[CUDA/CUTLASS] Improvements for varlen option derivation, input validation, can_implement errors, and workspace handling

Status: Open.

#161 In deepseek-ai/FlashMLA;

· red1239109-cmd opened on Feb 8, 2026

[Bug/Correctness] Hardcoded device_id=0 + missing CUDAGuard can break multi-GPU correctness (wrong hw_info / stream mismatch)

Status: Open.

#158 In deepseek-ai/FlashMLA;

· red1239109-cmd opened on Jan 25, 2026

[Question] What is MODEL1?

Status: Open.

#155 In deepseek-ai/FlashMLA;

· gary-wjc opened on Jan 21, 2026

[Question] DSA VS MLA Prefill Benchmark On H100

Status: Open.

#149 In deepseek-ai/FlashMLA;

· ZavierXing opened on Jan 15, 2026

FlashMLA for Blackwell architecture (B200)

Status: Open.

#148 In deepseek-ai/FlashMLA;

· abdul7mohsen opened on Jan 9, 2026