--- base_model: Qwen/Qwen2.5-Coder-0.5B-Instruct base_model_relation: adapter library_name: peft pipeline_tag: text-generation tags: - lora - peft - code datasets: - sahil2801/CodeAlpaca-20k --- # NavCoder LoRA v1 A coding-focused LoRA fine-tune of Qwen/Qwen2.5-Coder-0.5B-Instruct. ## Training configuration - Dataset: CodeAlpaca-20k - Training examples: 1,800 - Validation examples: 180 - LoRA rank: 8 - LoRA alpha: 16 - LoRA dropout: 0.05 - Learning rate: 2e-4 - Epochs: 1 - GPU: NVIDIA T4 ## Results - Final reported training loss: 0.678341 - Evaluation loss: 0.588350 Coding correctness benchmarks have not yet been run. ## Usage Load the adapter with PEFT and the original Qwen base model.