Squig Trainer
Fine-Tune AI Models on Your GPU
Fine-tune open LLMs on your own NVIDIA GPU. LoRA/QLoRA, DPO and KTO training with a guided first run, a VRAM fit check, live loss charts, resume, and one-click GGUF export with an Ollama Modelfile. Compare base and fine-tuned answers side by side. No command line required.
Screenshots
See it in action
Features
Everything you need
LoRA / QLoRA Training
Efficient fine-tuning with low-rank adaptation and 4-bit quantization, plus DPO and KTO for preference data. Pick from 80+ curated models or any compatible Hugging Face model.
Guided First Run
A setup wizard checks your GPU, previews your dataset, and shows a VRAM fit check before you start. A Quick test preset runs a short sample training in minutes.
Live Charts
Loss curves, ETA, tokens per second, and VRAM use while you train. Resume from a checkpoint if a run is interrupted.
Smart Auto-Config
Hyperparameters chosen from your model, data, and hardware.
GGUF Export
Export to GGUF with a ready-to-use Ollama Modelfile, or to SafeTensors as a LoRA adapter or merged model.
Playground Compare
Ask the same prompt of the base model and your fine-tuned model and read the answers side by side.
Pricing
Simple, fair pricing
Squig Trainer License
Start using Squig Trainer today
Join developers and creators who build with AI that respects their workflow.