# deepseek-ai/DeepSelect

> A high-performance CUDA implementation of the TopK kernel used in DeepSeek Sparse Attention (DSA) and the sampler, speeding up torch.topk by 2-20x. Targeted at developers running DeepSeek V3.2/V4 model inference on GPUs.

- Magnitude: 0.7 out of 10 — Steady
- Stars: 389 total · +0 stars on 2026-09-25 (UTC)
- Star trust: star growth looks organic
- Category: Language models · Language: Cuda · License: MIT · Created: 2026-09-09 · Last push: 2026-09-10
- GitHub: https://github.com/deepseek-ai/DeepSelect · Page: https://gitnova.dev/en/r/deepseek-ai/DeepSelect

## Useful for

- Speed up top-k token selection when sampling from a ~128K vocabulary on GPU
- Replace torch.topk in the Lightning Indexer inside DeepSeek Sparse Attention
- Run top-k with variable-length rows via the end argument and out-of-bounds fill values

## Why it’s here

- The repository is 16 days old and already has 389 stars.
- Recent forks include notable developers: @Enigmatisms (245 followers).

## 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: 29
- Issues and pull requests: 22
- Watchers: 0
- Average over the last week: 3 per day
- Usual pace: 19 per day

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

2026-09-30 … 2026-10-29: None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None

## Magnitude by day, last 13 days

09-13 2.5, 09-14 2.2, 09-15 2.1, 09-16 1.6, 09-17 1.4, 09-18 1.1, 09-19 1.1, 09-20 1.0, 09-21 1.3, 09-22 1.1, 09-23 1.4, 09-24 1.4, 09-25 0.9

---

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 2026-09-25 23:47 UTC, updated every 30 minutes.