TRLoom

Fine-tune LLMs with one YAML file on Hugging Face TRL

4 followers

Fine-tune LLMs with one YAML file on Hugging Face TRL

4 followers

TRLoom turns Hugging Face TRL fine-tuning into a single YAML config. Describe your model, dataset, trainer, W&B, and optional Modal GPUs — then run one command. `pip install trloom` `trloom run config.yaml` `trloom run config.yaml --modal` Built for ML engineers and researchers who want SFT, DPO, GRPO, and more without rewriting boilerplate every experiment. Open source (Apache 2.0).

Free

Launch tags:Open Source•Developer Tools•Artificial Intelligence

Launch Team

AWS Activate CreditsCredits that fuel your startup’s next milestone

Promoted

Maker

📌

Hey Product Hunt 👋 I’m Saqlain - I kept rewriting the same Hugging Face TRL setup for every SFT / DPO / GRPO run: model loading, datasets, trainers, W&B, then Modal when I needed GPUs. So I built TRLoom: one YAML file → one command. • TRL-native trainers (follows whatever your installed TRL exposes) • Hub or local datasets • W&B from YAML • Same config on Modal with --modal Try it: pip install trloom trloom run config.yaml Docs: https://saqlain2204.github.io/tr... GitHub: https://github.com/saqlain2204/t... Curious what you’d run first — local SFT, DPO, or a Modal GPU job? Feedback and issues welcome.

Report

2mo ago

Reviews