# Tencent-Hunyuan/Simple-Attention-Sparsification

> SAS is a gated sparse-attention mechanism that learns per-query which context blocks to attend to, optimizing context ranking end-to-end with the language modeling loss. The repo provides the paper's training and evaluation code.

- Magnitude: 0.0 out of 10 — Steady
- Stars: 60 total · +0 stars on 2026-09-26 (UTC), ≈ 0 by evening
- Star trust: star growth looks organic
- Category: Language models · Language: Python · Created: 2026-09-10 · Last push: 2026-09-14
- GitHub: https://github.com/Tencent-Hunyuan/Simple-Attention-Sparsification · Page: https://gitnova.dev/en/r/Tencent-Hunyuan/Simple-Attention-Sparsification

## Useful for

- Train sparse attention on Qwen3-4B/8B/14B with the OpenR1-Math-220k dataset
- Evaluate the model on MATH, GPQA-Diamond, AIME24/25 and LongBench-E via sglang-blocksparse
- Run BFCL function-calling and VitaBench tool-use benchmarks with sparse attention

## Why it’s here

- The repository is 16 days old and already has 60 stars.

## 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
- Issues and pull requests: 0
- Watchers: 0
- Average over the last week: 0 per day
- Usual pace: 4 per day

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

2026-10-05 … 2026-11-03: 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 11 days

09-16 1.0, 09-17 0.8, 09-18 0.4, 09-19 0.4, 09-20 0.4, 09-21 0.0, 09-22 0.0, 09-23 0.0, 09-24 0.0, 09-25 0.0, 09-26 0.0

---

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 2026-09-26 10:16 UTC, updated every 30 minutes.