Blog

Notes from production ML work: computer vision on the factory floor, inference optimization, and fine-tuning open LLMs.

LLM Fine-tuning

Fine-tuning gpt-oss-20b with Unsloth and running it in Ollama

QLoRA settings, VRAM, GGUF export and a working Ollama Modelfile for a fine-tuned gpt-oss-20b, including the Harmony template and stop token it needs.

· 5 min read