Koda-v1
Koda-v1 is a coding and conversational model created by Addy. This is the standard bfloat16 merged checkpoint of the experimental 100-step CPU QLoRA fine-tune of Qwen/Qwen2.5-Coder-1.5B-Instruct on HuggingFaceH4/Bespoke-Stratos-17k. It is packaged without a bitsandbytes quantization_config so standard Hugging Face Transformers and compatible inference providers can load it directly.
This is an experimental checkpoint, not a converged production model. Review generated code before execution and retain the Qwen license, notices, and usage conditions.