github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #15862] Metal shader compile fails on macOS 26.3.1 — MPPTensorOpsMatMul2dImpl static_assert (bfloat/half) on Apple M5 #72167

New Issue
Open
opened 2026-05-05 03:35:10 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-05-05 03:35:10 -05:00
Owner
Copy Link

Originally created by @mcp10 on GitHub (Apr 28, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15862

Environment

  • Hardware: MacBook Pro (Mac17,2), Apple M5
  • macOS: 26.3.1 (a) — build 25D771280a (Rapid Security Response)
  • Ollama: tested on 0.21.2 (stable) and 0.22.0-rc1 — same failure

Symptom

POST /api/chat returns 500 Internal Server Error: llama runner process has terminated. Affects every model loaded so far (gemma4:26b, llama3.2:3b).

Root cause (from logs)

On-the-fly Metal library compilation fails against macOS 26.3.1's MetalPerformancePrimitives headers, which now enforce strict type-matching on cooperative tensors:

ggml_metal_init: picking default device: Apple M5
ggml_metal_init: the device does not have a precompiled Metal library - this is unexpected
ggml_metal_init: will try to compile it on the fly
ggml_metal_library_init: error: Error Domain=MTLLibraryErrorDomain Code=3
  MPPTensorOpsMatMul2dImpl.h:3266: static_assert failed due to requirement
    '__tensor_ops_detail::__is_same_v<bfloat, half>'
    "Input types must match cooperative tensor types"
  MPPTensorOpsMatMul2dImpl.h:3267: static_assert failed due to requirement
    '__tensor_ops_detail::__is_same_v<half, bfloat>'
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

Embedded GGML shader source is instantiating mpp::tensor_ops::matmul2d with mismatched <half, bfloat> operands; the new MPP headers reject this.

Notes

  • Models that previously worked (cached compiled shaders) broke after the macOS 26.3.1 RSR purged Metal's per-device shader cache, forcing a recompile that now fails.
  • Reproduces on a fresh install of 0.22.0-rc1 (verified locally).
Originally created by @mcp10 on GitHub (Apr 28, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15862 ### Environment - Hardware: MacBook Pro (Mac17,2), Apple **M5** - macOS: **26.3.1 (a)** — build `25D771280a` (Rapid Security Response) - Ollama: tested on **0.21.2** (stable) and **0.22.0-rc1** — same failure ### Symptom `POST /api/chat` returns `500 Internal Server Error: llama runner process has terminated`. Affects every model loaded so far (`gemma4:26b`, `llama3.2:3b`). ### Root cause (from logs) On-the-fly Metal library compilation fails against macOS 26.3.1's `MetalPerformancePrimitives` headers, which now enforce strict type-matching on cooperative tensors: ``` ggml_metal_init: picking default device: Apple M5 ggml_metal_init: the device does not have a precompiled Metal library - this is unexpected ggml_metal_init: will try to compile it on the fly ggml_metal_library_init: error: Error Domain=MTLLibraryErrorDomain Code=3 MPPTensorOpsMatMul2dImpl.h:3266: static_assert failed due to requirement '__tensor_ops_detail::__is_same_v<bfloat, half>' "Input types must match cooperative tensor types" MPPTensorOpsMatMul2dImpl.h:3267: static_assert failed due to requirement '__tensor_ops_detail::__is_same_v<half, bfloat>' 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 ``` Embedded GGML shader source is instantiating `mpp::tensor_ops::matmul2d` with mismatched `<half, bfloat>` operands; the new MPP headers reject this. ### Notes - Models that previously worked (cached compiled shaders) broke after the macOS 26.3.1 RSR purged Metal's per-device shader cache, forcing a recompile that now fails. - Reproduces on a fresh install of 0.22.0-rc1 (verified locally).
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#72167
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?