💻DeepSeek-R1
Open-Source Reasoning, at GPT-o1 Level
⭐ Hugely popular: 92k stars, gaining about 37 a week
View on GitHub ↗repo profile
momentum
durability
since we covered it
why it's a big deal
- Provides open reasoning model weights under an MIT license, so teams can run and build on GPT-o1-level reasoning commercially rather than only calling a closed API.
- Ships six distilled variants from 1.5B to 70B built on Qwen2.5 and Llama, letting users pick a size that fits their own hardware instead of only the 671B full model.
- Documents concrete usage settings, including temperature 0.5 to 0.7, no system prompt, and a forced think prefix, giving operators a known-good starting configuration.
under the hood
- Written in Python around a mixture-of-experts model with 671B total parameters, 37B activated per token, and a 128K context window, derived from DeepSeek-V3-Base.
- DeepSeek-R1-Zero is trained with large-scale reinforcement learning and no supervised fine-tuning, while DeepSeek-R1 adds cold-start data and two further SFT stages to cut repetition, poor readability, and language mixing.
- Full models run through the DeepSeek-V3 setup rather than Hugging Face Transformers, while the distilled models are served with vLLM or SGLang.
Radar summary, generated from the project's public sources
star history
- PR#1 57k 2025-02-05
- now 92k + 35k since first covered
curve is sampled from GitHub's star history, plus our own daily readings since we covered it; the dashed stretch is before we first covered it, the solid line since. figures at coverage are the numbers we printed then (approx.), current count is live.
understory
Better known than its recent output, coasting a little on attention.
- output, commits & releases
- clout, star velocity
output = commits & releases; clout = star velocity, both 0 to 100 monthly indices; the gap where output runs above clout is the understory. The understory →
covered in
-
Open-Source Reasoning, at GPT-o1 Level
similar projects
compare these →- ☸️ EXO
Python · leaner, 47k stars
Run frontier AI locally.
47k ACTIVE - 🔗 Fabric
Go · leaner, 43k stars
Fabric is an open-source framework for augmenting humans using AI. It provides a modular system for solving specific problems using a crowdsourced set of AI prompts that can be used anywhere.
43k ACTIVE - 🖥️ openscreen
TypeScript · leaner, 40k stars
Create stunning demos for free. Open-source, no subscriptions, no watermarks, and free for commercial use. An alternative to Screen Studio.
40k OFF THE RADAR
comments
Sign in with GitHub to add your blip on DeepSeek-R1.