Ollama version 0.30 has been released. Through llama.cpp, performance has been improved and GGUF model compatibility has been enhanced. Through this, Apple Silicon-based Ollama's MLX engine has become more powerful and now supports more models on more diverse hardware. | | Performance improvements on more GPUs | Faster throughput on NVIDIA hardware | With Ollama 0.30, performance on NVIDIA hardware improves up to 20%, thanks to optimizations provided by the NVIDIA and llama.cpp teams. | | | Tested with the Gemma 4 26B model using Q4_K_M quantization on NVIDIA RTX 5090. | | Broader hardware support through Vulkan | Vulkan is now enabled by default, extending Ollama's GPU acceleration to a wider range of hardware including AMD and Intel devices. Now more users can run models directly on GPU without installing vendor-specific libraries. | | Support for more models | Ollama 0.30 expands compatibility with the GGUF ecosystem to LFM and Prism model families, as well as fine-tuned models released by Unsloth, so more models can be run directly without additional configuration. | Run GGUF models from Hugging Face | To use a model, first download the GGUF file or a directory containing GGUF files. Then create a Modelfile that specifies the path to the GGUF file (or directory) using the FROM command. | | FROM ./my-model.Q4_K_M.gguf | | Next, create and run the model. | ollama create -f Modelfile my-model
ollama run my-model | | Coding agents and assistants | If a model supports tool calling, that capability is available in Ollama as well. Therefore, you can use these models with your preferred coding agents and personal assistants with a single command. | | Claude Code | ollama launch claude --model my-model | | Hermes Agent | ollama launch hermes --model my-model | | OpenClaw | ollama launch openclaw --model my-model | | To check if a GGUF file supports tool calling, use the ollama show command to look for the tools capability. | | ollama show my-model | | Acknowledgments | We would like to thank Georgi Gerganov and the llama.cpp maintenance team, as well as our hardware partners including NVIDIA, AMD, Qualcomm, and Intel. They have put in great effort to optimize the performance of the GGML ecosystem on their respective platforms. | | If you have feedback, join Ollama's Discord or contact hello@ollama.com. | |  BE |
|
|
|
I got this email.
Previously, the model downloaded from Hugging Face was converted to a format compatible with Olllama after further processing. Now it seems we can use it directly.
The version I used before was 0.24. The update is quite fast. It's been less than two months and it's already a 0.06 update.
