# Run a 4B LLM model locally on your mac > Source: > Published: 2026-08-12 09:39:32+00:00 | Run a 4B LLM model locally on your mac, this has been tested on MacOS Tahoe | | | Step 1: | | | Lookup "llamafile github" or go to https://github.com/mozilla-ai/llamafile | | | Download the latest release, as of this it is llamafile v0.10.5 | | | Step 2: | | | Go to https://huggingface.co/models | | | Search for qween3 | | | Click on pramodlohra/Qween3_4B_thinking_finetune | | | Go to Files and versions | | | Download qwen3-4b-thinking-2507.Q4_K_M.gguf | | | Step 3: | | | Create a folder called qwen | | | Copy llamafile v0.10.5 and qwen3-4b-thinking-2507.Q4_K_M.gguf into the folder | | | Step 4: | | | In this folder, create a file called run_qwen.command | | | Write the following lines in it | | | #!/bin/bash | | | cd "$(dirname "$0")" | | | ./llamafile-0.10.5 --server --model ./qwen3-4b-thinking-2507.Q4_K_M.gguf | | | Save the run_qwen.command file | | | Step 5: | | | Run chmod for llamafile llamafile-0.10.5 and run_qwen.command | | | chmod 755 llamafile-0.10.5 | | | chmod 755 run_qwen.command | | | Step 6: | | | Remove the quarantine flag from running this in the terminal | | | xattr -d com.apple.quarantine run_qwen.command | | | xattr -d com.apple.quarantine llamafile-0.10.5 | | | You can check whether quarantine is still present with: | | | xattr -l run_qwen.command | | | xattr -l llamafile-0.10.5 | | | If com.apple.quarantine doesn't appear, you're good. | | | NOTE: Don't disable Gatekeeper globally with commands such as sudo spctl --master-disable. | | | There's absolutely no need for that here; removing the quarantine attribute from these specific files is much safer. | | | Now you are ready to click on the run_qwen.command file and run the model locally | | | If everything is done correctly, this is what would show up in the terminal. | | | % /Users/santa/Downloads/qwen/run_qwen.command ; exit; | | | 0.00.000.673 I cmn common_param: common_params_print_info: verbosity = 3 (adjust with the `-lv N` CLI arg) | | | 0.00.000.921 W srv llama_server: ----------------- | | | 0.00.000.926 W srv llama_server: CORS is set to allow all origins ('*') and no API key is set | | | 0.00.000.926 W srv llama_server: this can be a security risk (cross-origin attacks) | | | 0.00.000.927 W srv llama_server: more info: https://github.com/ggml-org/llama.cpp/pull/25655 | | | 0.00.000.927 W srv llama_server: ----------------- | | | 0.00.001.045 W srv llama_server: sandbox: disabled in GPU mode | | | 0.00.002.429 I srv load_model: loading model './qwen3-4b-thinking-2507.Q4_K_M.gguf' | | | 0.00.429.537 W load: control-looking token: 128247 '' was not control-type; this is probably a bug in the model. its type will be overridden | | | 0.04.590.135 I srv load_model: initializing, n_slots = 4, n_ctx_slot = 14080, kv_unified = 'true' | | | 0.04.600.756 I srv llama_server: model loaded | | | 0.04.600.778 I srv llama_server: listening on http://127.0.0.1:8080 | | | You can start using the model on 127.0.0.1:8080, Enjoy! |