All issues

Search results

[Portability][MSVC] long2 is only 64 bits on Windows but is used as a 128-bit PTX transport

Status: Open.

#219 In deepseek-ai/FlashMLA;

· rogeroberg opened on Aug 30, 2026

[Portability][MSVC] Replace compound-literal-style dependent-type initialization in dense MLA decode

Status: Open.

#218 In deepseek-ai/FlashMLA;

· rogeroberg opened on Aug 30, 2026

[Portability][MSVC] Replace __int128_t in shared-memory load/store paths 3

Status: Open.

#217 In deepseek-ai/FlashMLA;

· rogeroberg opened on Aug 30, 2026

Sparse MLA decode is not batch-invariant: a row's output depends on its batchmates (1-2 ULP), flipping argmax at near-tie tokens

Status: Open.

#212 In deepseek-ai/FlashMLA;

· apejcic opened on Aug 21, 2026

PyBind11 interface: How should function signature and CUDA error handling be handled for registered kernels?

Status: Open.

#211 In deepseek-ai/FlashMLA;

· jsramesh1990 opened on Aug 18, 2026

10 rounds of hand-written fused attention that never beat a 3-kernel cuBLAS stack — trajectory, where each round died, and a correctness trap

Status: Open.

#210 In deepseek-ai/FlashMLA;

· dancinlife opened on Aug 7, 2026

DeepSeek-Coder-V2-Lite-Instruct 16B running locally on an RTX 2050 laptop

Status: Open.

#206 In deepseek-ai/FlashMLA;

· lkdmfklsdf opened on Aug 3, 2026

flashmla dense decoding test lack fp8 kv cache case

Status: Open.

#204 In deepseek-ai/FlashMLA;

· xu78guo6 opened on Aug 3, 2026

Sparse MLA decode (V3.2 / FP8 KV): throughput cost of the B200 accuracy fix (5aa668c) scales with topk

Status: Open.

#192 In deepseek-ai/FlashMLA;

· MogicianWu opened on Jun 26, 2026

flash mla kernal是否支持同一batch下不同query的动态token数？

Status: Open.

#190 In deepseek-ai/FlashMLA;

· echo-timeless opened on Jun 17, 2026

Make FlashMLA a libtorch and cpython stable extension

Status: Open.

#180 In deepseek-ai/FlashMLA;

· janeyx99 opened on Apr 27, 2026

Do you open to add SM80 support

Status: Open.

#179 In deepseek-ai/FlashMLA;

· haosdent opened on Apr 27, 2026