End-to-end QLoRA fine-tuning platform with live loss streaming and base-vs-fine-tuned evaluation.
A platform that turns LLM fine-tuning into a product workflow: upload a dataset, pick a base model, configure LoRA / QLoRA hyper-parameters and launch a run — all from the browser.
Training telemetry (loss, learning rate, throughput) streams live to the UI over WebSockets, so a run can be judged and stopped early instead of discovered hours later.
Once training finishes, the base and fine-tuned models answer the same prompts side by side, making the effect of the run visible before anything is deployed.
- QLoRA fine-tuning via Hugging Face Transformers, PEFT and TRL
- Live training telemetry streamed to the browser
- Side-by-side base vs fine-tuned evaluation
QLoRATransformersPEFTTRLWebSocketsPython