Use Ollama with GGUF in-place

<!-- SC_OFF --><div class="md"><p>With the <a href="https://www.reddit.com/r/LocalLLaMA/s/3mi8ohC5nN">recent change</a> leading to -hf downloaded models being moved and saved as

<!-- SC_OFF --><div class="md"><p><a href="https://github.com/ollama/ollama/releases/tag/v0.30.0-rc15">https://github.com/ollama/ollama/releases/tag/v0.30.0-rc15</a></p>

<!-- SC_OFF --><div class="md"><p>Hi all I had a quick question while we wait for llama.cpp MTP implementation, have any of y'all tried Gemma4 MTP models on ollama and or

<!-- SC_OFF --><div class="md"><p>Anthropic opened Cowork for Bedrock/Vertex/Azure providers and also Custom Inference Endpoints.</p> <p>However, connecting it to a local proxy

<!-- SC_OFF --><div class="md"><p>Been using ClaudeCode CLI with Opus 4.6 and many MCP's and honestly its addicting. </p> <p>Just tell it what to build and it does everything —

<!-- SC_OFF --><div class="md"><p>I can't run the recently released 1-bit Bonsai-4B.gguf <a href="https://huggingface.co/prism-ml/Bonsai-4B-gguf/tree/main">model</a> in llama.cpp.

<!-- SC_OFF --><div class="md"><p>Beginner here, now I have a good GPU and ollama using docker. Pulled the Gemma4 weights and was able to add it to cursor using ngrok. </p>

<!-- SC_OFF --><div class="md"><p>Noob here, Running Qwen3.6 35B A3B in LM Studio on a 3080 10GB + Ryzen 5 3600 on Windows 10.</p> <p>Tried some unsloth quants with identical