fix: remove max_new_tokens from GRPOConfig, set batch_size=2 (multiple of num_generations)

fix: pin trl<=0.24, multi-step reward, lower LR, reduce NUM_GEN

fix: remove max_new_tokens from GRPOConfig (not supported by Unsloth patched version)

x

context

context error

fix: add missing server modules and unsloth import order

Fix imports in environment.py

Add environment + models + apps to Space

fix

Fix: make server a Python package

Fix matplotlib for Space

Add OrgOS training pipeline

initial commit