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.