unsloth vs minimind
Two of the top projects in Training and fine-tuning, side by side: who leads on each score part and who each one suits, then setup, license, activity and what each review found.
unsloth
Desktop and web app to run, fine-tune and export open models
minimind
Train a 64M-parameter LLM from scratch in plain PyTorch
Verdict
unsloth ranks higher, 87 to 63. Of the 4 score parts we measure for both, unsloth leads on 3, minimind on none and 1 is even.
- Adoptionunsloth leads by 20 points, 93 to 73: a wide gap.
- Freshnessunsloth leads by 15 points, 100 to 85: a wide gap.
- Maintenanceunsloth leads by 45 points, 97 to 52: a wide gap.
- Agent-readyEven at 0.
Pick unsloth if you are
Engineers fine-tuning and locally running open LLMs on their own GPUs
- Fine-tuning covers LoRA, QLoRA, full fine-tuning, GRPO, DPO and FP8 pretraining
- Claims 2x faster training with up to 70% less VRAM, per README
Pick minimind if you are
Engineers learning LLM internals by pretraining and fine-tuning small models
- Core algorithms (LoRA, DPO, PPO/GRPO) written in plain PyTorch, not wrapped from trl or peft
- Open datasets for each stage, with mini variants (1.2GB pretrain, 1.6GB SFT) for quick runs
| What we compare | unsloth | minimind |
|---|---|---|
| Score parts, out of 100 | ||
| Adoption | 93, widely used | 73, popular |
| Freshness | 100, active | 85, active |
| Maintenance | 97, healthy | 52, fair |
| Easy to run | Not scored for this kind | Not scored for this kind |
| Agent-ready | 0, none | 0, none |
| Facts from GitHub and the README | ||
| Stars | 78k | 63.8k |
| License | Apache-2.0 (permissive) | Apache-2.0 (permissive) |
| Last commit | Oct 2026 | Sep 2026 |
| Last release | Oct 2026 | Oct 2025 |
| Language | Python | Python |
| Docker | Yes | No |
| GPU | Optional | Optional |
| arm64 or Apple Silicon | Mentioned | Not stated |
unsloth
Unsloth ships as a desktop app, a web UI (Studio) and a Python library for running and fine-tuning models. It runs GGUF and other open models locally with an OpenAI-compatible API, and trains with LoRA, QLoRA, full fine-tuning and RL methods such as GRPO and DPO. The README claims 2x faster training with up to 70% less VRAM, and exports to GGUF, FP8 and NVFP4.
Who it is for: Engineers fine-tuning and locally running open LLMs on their own GPUs
Strengths
- Fine-tuning covers LoRA, QLoRA, full fine-tuning, GRPO, DPO and FP8 pretraining
- Claims 2x faster training with up to 70% less VRAM, per README
- Exports to GGUF, FP8 and NVFP4 and serves an OpenAI-compatible API
- Runs on NVIDIA, AMD, Intel, CPU and Vulkan across Windows, macOS and Linux
Weaknesses
- Scope is broad: desktop app, Studio web UI and Core library each install differently
- Server-side tools are on by default; exposing it needs --disable-tools or a strong password
- Install relies on curl or irm piped to shell scripts
- Speed and VRAM figures are the project's own claims; hardware requirements are not listed
- GPU optional
- Docker
- Needs PyTorch, llama.cpp, audio.cpp, Cloudflare (optional remote access)
- Models: GGUF, MLX, LoRA, QLoRA, diffusion
- port 8000
minimind
MiniMind is a from-scratch implementation of a small Qwen3-style decoder-only LLM (64M dense, 198M-A64M MoE) with each training stage written in native PyTorch: pretrain, SFT, LoRA, DPO, PPO/GRPO/CISPO, agentic RL and distillation. It ships datasets, a 6,400-token tokenizer, an OpenAI-compatible server and a Streamlit chat UI. The README states one SFT epoch takes about 2 hours on a single 3090.
Who it is for: Engineers learning LLM internals by pretraining and fine-tuning small models
Strengths
- Core algorithms (LoRA, DPO, PPO/GRPO) written in plain PyTorch, not wrapped from trl or peft
- Open datasets for each stage, with mini variants (1.2GB pretrain, 1.6GB SFT) for quick runs
- Weights convert to transformers, llama.cpp, vllm and ollama; ollama and vllm usage is documented
- Checkpoint resume, DDP and DeepSpeed multi-GPU, wandb or swanlab logging
Weaknesses
- Models are tiny (64M), so output quality is far below mainstream LLMs
- README is primarily in Chinese; the English version is a separate file
- Training data is largely synthetic and distilled, with mixed source licenses
- Native MoE training is about 50% slower than dense, per the README
- GPU optional
- Needs PyTorch, transformers, streamlit (optional WebUI), swanlab or wandb (optional)
- Models: minimind-3 (64M), minimind-3-moe (198M-A64M), minimind2 series, Transformers format, llama.cpp
More in Training and fine-tuning
Building with these?
We help teams pick an open-source AI stack and run it in production. Client work never changes a rank.