MiniMind
Train small language models from scratch with complete PyTorch pipeline in just 2 hours
Opens jingyaogong.github.io in a new tab.
How the 7.6 was reached
- Quality of output8/10
62,855 GitHub stars, active, Apache-2.0; heavy adoption for a small-model training project, though the 64M model is naturally limited. Source: github.com
- Ease of use6/10
Pure PyTorch code covering the full pipeline, but you need a GPU and training know-how; docs found are limited. Source: jingyaogong.github.io
- Pricing value9/10
Free Apache-2.0 code; training a 64M model costs about $3 on a single 3090. Source: github.com
- API and integration quality7/10
Works with vLLM, ollama, llama.cpp, SGLang, transformers, and offers an OpenAI-compatible API for Open-WebUI and Dify. Source: jingyaogong.github.io
- Solves the problem it claims to solve8/10
Trains a small chatbot from scratch cheaply; 62,855 stars show wide adoption, though cost and time figures come from the project. Source: github.com
7.6/10 is the average of the five criterion scores above. The evidence comes from 3 different sources.Scores come from public evidence (the tool's own pages and documentation, and independent reviews); we do not test every tool hands-on. Scored 29 September 2026, on version 1 of our five-criteria scoring method.
Pricing
- Complete source code
- Training pipeline
- Multiple model architectures
- Documentation
For the full list of what each plan includes, and current prices, see github.com.
Key Features
- 64M parameter ChatBot training
- Complete pipeline from tokenizer to deployment
- RLAIF algorithms (PPO, GRPO, CISPO)
- Compatible with vLLM, ollama, llama.cpp
- OpenAI API drop-in replacement
Pros & Cons
Pros
- Ultra-low cost training (~¥3)
- Pure PyTorch implementation with no black boxes
- Complete end-to-end pipeline
- Multiple deployment options
- Active development with latest RL techniques
Cons
- Requires technical expertise in ML
- Limited to small model sizes
- Documentation primarily in Chinese
- Requires GPU hardware for training
MiniMind is an excellent educational tool for learning LLM training from scratch. While limited to smaller models, it provides a complete, transparent pipeline that's perfect for understanding the fundamentals of language model development.
Added to scored.tools on
For the MiniMind team: put this score on your website
A small image (a badge) that shows this score and links back to this page. Free for any tool, at any score. It always shows the current score, so it changes when the score does. Copy this code into your website where you want the badge:
<a href="https://scored.tools/tools/minimind/"><img src="https://scored.tools/badge/minimind.svg" alt="MiniMind on scored.tools: 7.6/10" width="160" height="40"></a>Do you work for MiniMind? Claim this page: prove you work there, then confirm the facts, correct a mistake or add a short reply. It is free and never changes the score.
Competitors to MiniMind
Other tools in the Coding category worth comparing.
