Trloom: run TRL trainers (SFT, DPO, GRPO) from one YAML config

Developer saqlain2204 released Trloom v0.1.3, an open-source tool that reads a single YAML config and builds the matching Hugging Face TRL trainer for SFT, DPO, or GRPO, mapping the `training` block to SFTConfig, DPOConfig, or GRPOConfig. Trloom is installed via `pip install trloom` and run with `trloom validate grpo.yaml` and `trloom run grpo.yaml`, with a `--modal` flag to execute on a Modal GPU such as an A100. The tool supports LoRA (lora_r 16, lora_alpha 32), reward functions like accuracy_reward, and W&B reporting, and the author is soliciting feedback on the config format and recipe requests.

Hi all, I built Trloom, an open-source tool built on top of TRL. It reads one YAML file and builds the matching TRL trainer, so the training block maps to SFTConfig / DPOConfig / GRPOConfig. Here is GRPO on Qwen2.5-0.5B with LoRA and accuracy reward: method: grpo model: model name or path: Qwen/Qwen2.5-0.5B-Instruct dtype: bfloat16 use peft: true lora r: 16 lora alpha: 32 dataset: path: HuggingFaceH4/Polaris-Dataset-53K train split: train reward funcs: - accuracy reward training: output dir: ./outputs/grpo-modal learning rate: 1.0e-5 per device train batch size: 1 gradient accumulation steps: 4 bf16: true report to: wandb wandb: enabled: true project: trloom-grpo modal: enabled: true gpu: A100 secrets: huggingface, wandb pip install trloom , then trloom validate grpo.yaml and trloom run grpo.yaml add --modal to run on a Modal GPU . Repo: GitHub - saqlain2204/trloom: YAML-driven end-to-end fine-tuning on Hugging Face TRL — one config for model, dataset, W&B, and Modal GPUs. · GitHub https://github.com/saqlain2204/trloom Docs: TRLoom https://saqlain2204.github.io/trloom/ It’s early v0.1.3 . Happy to add recipes people ask for, and feedback on the config format is very welcome.