Search results

[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

does flash_mla_with_kvcache work only in paged mode?

Status: Open.

#126 In deepseek-ai/FlashMLA;

· vince62s opened on Dec 16, 2025

Can MHA be used in the DSA prefill stage?

Status: Open.

#125 In deepseek-ai/FlashMLA;

· starwang1024 opened on Dec 9, 2025

support for rtx 6000 sm120

Status: Open.

#124 In deepseek-ai/FlashMLA;

· fernandaspets opened on Dec 3, 2025

Build error on cuda13 arm64

Status: Open.

#121 In deepseek-ai/FlashMLA;

· icavanyu opened on Nov 16, 2025

Sparse Decode Mla kernel opt plan On blcakwell

Status: Open.

#120 In deepseek-ai/FlashMLA;

· lw921014 opened on Nov 9, 2025

Built FlashMLA for Windows and Nvidia sm_120 (workstation/50s) cards

Status: Open.

#119 In deepseek-ai/FlashMLA;

· IISuperluminaLII opened on Oct 30, 2025

how to test sparse attention in test_flash_mla_decoding

Status: Open.

#116 In deepseek-ai/FlashMLA;

· jerry-dream-fu opened on Oct 24, 2025

No code for sm100 dense mla.

Status: Open.

#113 In deepseek-ai/FlashMLA;

· engineer1109 opened on Oct 17, 2025