swe-gym

Run the SWE-Gym training pipeline — skill learning, SFT, GRPO, or the full pipeline. Use when the user wants to train skills, fine-tune models, generate tasks, or run the gym pipeline.

Run GRPO reinforcement learning training on gym tasks. Use when the user wants to train a model with RL rewards from test suites, format checks, or hybrid scoring.

Run hyperparameter optimization for LoRA fine-tuning using Optuna. Use when the user wants to find optimal LoRA hyperparameters before full training.

Run Unsloth SFT/DPO/CPT fine-tuning. Use when the user wants to train, fine-tune, or adapt a model with LoRA, full fine-tune, DPO, continued pretraining, or QAT using Unsloth.

Combined Unsloth training hub — SFT, GRPO, HPO, DPO, CPT, QAT, and export. Use when the user wants to train, optimize, or export a model and you need to choose or chain the right training approach.