ollama - 💡(How to fix) Fix Ollama 0.21.0 fails to initialize Metal on Apple M5 / macOS 26.2 while 0.18.0 works [1 participants]

ggml_metal_init: the device does not have a precompiled Metal library - this is unexpected ggml_metal_library_init: using embedded metal library ... MPPTensorOpsMatMul2dImpl.h ... error: static_assert failed due to requirement '__tensor_ops_detail::__is_same_v<bfloat, half>' "Input types must match cooperative tensor types" error: static_assert failed due to requirement '__tensor_ops_detail::__is_same_v<half, bfloat>' "Input types must match cooperative tensor types" ggml_metal_init: error: failed to initialize the Metal library ggml_backend_metal_device_init: error: failed to allocate context llama_init_from_model: failed to initialize the context: failed to initialize Metal backend panic: unable to create llama context

TL;DR

Downgrade to Ollama 0.18.0 to fix the local model loading issue on Apple M5 with macOS 26.2.

Guidance

• The issue is likely caused by a compatibility problem between Ollama 0.21.0 and the Metal initialization on Apple M5 with macOS 26.2.
• To verify the issue, run the repro command ollama run "llama3.2:latest" "Return exactly this JSON and nothing else: {\"ok\":true}" --format json --nowordwrap and check the ~/.ollama/logs/server.log for Metal initialization errors.
• Downgrading to Ollama 0.18.0 has been confirmed to fix the issue, so this can be used as a temporary workaround.
• The issue appears to be specific to the newer runtime on Apple M5 and the current Metal toolchain, so it may be worth monitoring for updates to Ollama or the Metal toolchain that could resolve the issue.

Notes

The root cause of the issue is not explicitly stated, but it appears to be related to changes in the Metal initialization between Ollama 0.18.0 and 0.21.0.

Recommendation

Apply workaround: Downgrade to Ollama 0.18.0, as this has been confirmed to fix the issue and allows for local model loading to work as expected.