This model is deprecated and should not be used. It suffers from output collapse and produces degenerate/repetitive outputs. Use the RFT version instead.
A supervised fine-tuned LoRA adapter on Mistral-7B-Instruct-v0.3 for Roblox Luau code generation. This was an early experiment that did not produce stable outputs.
Do Not Use
This model frequently collapses into repetitive or degenerate outputs. The successor RFT model fixes this through reinforcement fine-tuning with hybrid reward scoring.
