# deepseek-ai/FlashMLA

> DeepSeek's library of optimized attention kernels (sparse and dense) for running DeepSeek-V4.1 inference on NVIDIA GPUs and Huawei Ascend NPUs, with FP8/FP4 KV cache support.

- Magnitude: 2.3 out of 10 — Steady
- Stars: 13,151 total · +3 stars measured 2026-10-22, 00:42–04:12 UTC, ≈ 14 by evening
- Star trust: star growth looks organic
- Category: Language models · Language: C++ · License: MIT · Created: 2025-02-21 · Last push: 2026-10-15
- GitHub: https://github.com/deepseek-ai/FlashMLA · Page: https://gitnova.dev/en/r/deepseek-ai/FlashMLA

## Useful for

- Run sparse MLA prefill and decoding benchmarks on B200 or Ascend 950
- Integrate FP8/FP4 KV cache attention kernels into a DeepSeek-V4.1 inference pipeline
- Benchmark the fused norm-RoPE-attn-RoPE-cast kernel to reduce small-kernel overhead

## Why it’s here

- Star-counter measurements on 2026-10-22 (UTC), 00:42–04:12: 13148 → 13151 stars (+3). This is the change over that interval.
- Estimated end-of-day forecast: about +14 stars, using observed gains and the previous day.
- Over the last two days the pace is 3.4× that of the previous week and a half.
- GitHub Trending C++ today: #5, +35 stars.
- About 12 forks a day — people are taking the code.

## Star trust

Star growth looks organic.

Star-trust labels are heuristics based on the repository’s behavior, not a check of every stargazer.

## Numbers

- Forks: 1,222
- Issues and pull requests: 234
- Watchers: 106
- Average over the last week: 13 per day
- Usual pace: 6 per day
- Stars in the last hour (measured): 2

## Stars per day, last 30 days (oldest → newest, today is partial)

2026-09-23 … 2026-10-22: 0, 1, 0, 0, 3, 3, 32, 26, 13, 4, 5, 2, 5, 3, 8, 7, 7, 4, 2, 5, 3, 6, 3, 4, 4, 4, 2, 49, 13, 3

## Magnitude by day, last 2 days

10-21 2.6, 10-22 2.3

## Spotted in now

- GitHub Trending C++ today: #5, +35 stars

## Similar by description

1. **deepseek-ai/DeepInfer** — 9.8 · Breakout · Language models · Python · +533 stars measured 2026-10-22, 00:37–04:07 UTC, ≈ 3,065 by evening
   Offline inference framework for running DeepSeek models at large scale: optimized kernels, MoE/Attention parallelism, deterministic sampling and structured outputs. Aimed at teams deploying DeepSeek-V4.1-Flash on NVIDIA SM100 or Huawei…
   Full card: https://gitnova.dev/en/r/deepseek-ai/DeepInfer.md

---

Magnitude (0–10) measures how fast and how unusually interest in a repository is growing right now. It is not a quality score.
Days are UTC. “So far today” is a fact; “expected by the end of the day” is a forecast.
Summaries and use cases are written by an LLM (DeepSeek V4.1 Flash) from the README and may be inaccurate: verify specific claims (benchmarks, speed, hardware) in the repository itself.
Data as of 2026-10-22 04:18 UTC, updated every 30 minutes.