Search results

Question of O_accum and LSE_accum shape

Status: Open.

#70 In deepseek-ai/FlashMLA;

· mingyangHao opened on Apr 7, 2025

flash_fwd_mla_kernel.h offset calculation 偏移量计算问题

Status: Open.

#67 In deepseek-ai/FlashMLA;

· ChengruiZhang opened on Mar 11, 2025

Ampere architecture FlashMLA bring-up

Status: Open.

#66 In deepseek-ai/FlashMLA;

· pzhao-eng opened on Mar 11, 2025

Why add permanent false condition(sum_lse != sum_lse) for determining global_lse?

Status: Open.

#64 In deepseek-ai/FlashMLA;

· LearnerInGithub opened on Mar 6, 2025

Question about usage of __launch_bounds__( )

Status: Open.

#63 In deepseek-ai/FlashMLA;

· LearnerInGithub opened on Mar 6, 2025

Question: why TMA is not used

Status: Open.

#61 In deepseek-ai/FlashMLA;

· Idonthaveaname-wq opened on Mar 5, 2025

Request for Ampere GPU Support in FlashMLA​

Status: Open.

#60 In deepseek-ai/FlashMLA;

· ehartford opened on Mar 3, 2025

Are the test performance results of tests/test_flash_mla.py accurate?

Status: Open.

#59 In deepseek-ai/FlashMLA;

· pipul opened on Mar 3, 2025

Where can I find the definition for the joint compression of K and V in this code?

Status: Open.

#58 In deepseek-ai/FlashMLA;

· houghtonweihu opened on Mar 3, 2025

FP8 Support

Status: Open.

#56 In deepseek-ai/FlashMLA;

· endurehero opened on Feb 28, 2025

这个库是对应v2 paper里提到的MLA吗？

Status: Open.

#53 In deepseek-ai/FlashMLA;

· echoht opened on Feb 28, 2025

Why head_size must be 576 while q_head_dim of DeepSeek-V3 is only 192?

Status: Open.

#49 In deepseek-ai/FlashMLA;

· WangNorthSea opened on Feb 27, 2025