ollama - 💡(How to fix) Fix [Bug]: Metal compilation error on Apple M5 (macOS Tahoe 26.3.1) - static_assert failed [3 comments, 4 participants]

The following error is captured in ~/.ollama/logs/server.log during the model load attempt:

ggml_metal_library_init: error: Error Domain=MTLLibraryErrorDomain Code=3
In file included from /System/Library/Frameworks/MetalPerformancePrimitives.framework/Headers/__impl/MPPTensorOpsMatMul2dImpl.h:3266:5: 
error: static_assert failed due to requirement '__tensor_ops_detail::__is_same_v<bfloat, half>' 
"Input types must match cooperative tensor types"

The runner crashes shortly after:

time=2026-04-14T14:42:25.972-03:00 level=ERROR source=server.go:316 msg="llama runner terminated" error="exit status 2"
[GIN] 2026/04/14 - 14:42:25 | 500 |  2.125581542s |       127.0.0.1 | POST     "/api/generate"

TL;DR

The issue can be resolved by updating the Ollama version to support the stricter Metal compiler requirements on Apple M5 hardware with macOS Tahoe.

Guidance

• Verify that the error is caused by the Metal library initialization by checking the server.log for the ggml_metal_library_init error message.
• Attempt to bypass the GPU with export OLLAMA_LLM_LIBRARY=cpu to confirm that the issue is related to the Metal library.
• Consider updating the Ollama version to a newer release that may include fixes for the Metal compiler compatibility issues on Apple M5 hardware.
• If updating Ollama is not possible, try to find alternative models that are compatible with the current Metal library version.

Example

No code snippet is provided as the issue is related to a compatibility problem between the Ollama version and the Metal library on Apple M5 hardware.

Notes

The current Ollama version (0.20.7) may not be compatible with the stricter Metal compiler requirements on Apple M5 hardware with macOS Tahoe. Updating to a newer version of Ollama may resolve the issue.

Recommendation

Apply workaround: Update Ollama to a version that supports the stricter Metal compiler requirements on Apple M5 hardware with macOS Tahoe, if available. This is recommended because the current version (0.20.7) is incompatible with the Metal library on this specific hardware and OS combination.