github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #14516] MiniCPM-o 4.5: Audio input crashes with GGML_ASSERT on macOS Metal (Apple M4) #71479

New Issue
Closed
opened 2026-05-05 01:51:15 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-05-05 01:51:15 -05:00
Owner
Copy Link

Originally created by @samuelazran on GitHub (Feb 28, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14516

Description

MiniCPM-o 4.5 (openbmb/minicpm-o4.5) crashes immediately when sending audio input via the /api/chat endpoint on macOS with Apple Silicon (M4). Vision and text-only queries work correctly — the crash is specific to audio modality.

Steps to Reproduce

  1. ollama pull openbmb/minicpm-o4.5
  2. Send an audio request:
import base64, requests

with open("test.wav", "rb") as f:
    audio_b64 = base64.b64encode(f.read()).decode()

resp = requests.post("http://localhost:11434/api/chat", json={
    "model": "openbmb/minicpm-o4.5",
    "messages": [{"role": "user", "content": "What do you hear?", "audios": [audio_b64]}],
    "stream": False,
})
print(resp.status_code, resp.text[:300])

Error

HTTP 500
{"error":"llama runner process has terminated: GGML_ASSERT([rsets->data count] == 0) failed\nWARNING: Using native backtrace. Set GGML_BACKTRACE_LLDB for more info.\nWARNING: GGML_BACKTRACE_LLDB may cause native MacOS Terminal.app to crash.\nSee: https://github.com/ggml-org/llama.cpp/pull/17869"}

The crash happens with any audio file (tested synthetic sine wave, real speech WAV files at 16kHz mono).

Expected Behavior

The model should process audio input and return a text response, same as it does for image/text inputs.

Workaround

Audio works correctly via llama.cpp-omni (the fork built specifically for MiniCPM-o), which handles Metal acceleration for audio without this assertion failure. The issue appears to be in Ollama's Metal backend when processing audio embeddings.

Environment

  • Ollama version: 0.17.4
  • OS: macOS 26.2 (Build 25C56)
  • Hardware: Apple M4, 16GB unified memory
  • Model: openbmb/minicpm-o4.5:latest (6.1 GB, Q4_0)

Related

  • #14065 — MiniCPM-o 4.5 support request
  • #11798 — Audio input support for multimodal models
  • The assertion [rsets->data count] == 0 is in the Metal residency set handling code
Originally created by @samuelazran on GitHub (Feb 28, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14516 ## Description MiniCPM-o 4.5 (`openbmb/minicpm-o4.5`) crashes immediately when sending audio input via the `/api/chat` endpoint on macOS with Apple Silicon (M4). Vision and text-only queries work correctly — the crash is specific to audio modality. ## Steps to Reproduce 1. `ollama pull openbmb/minicpm-o4.5` 2. Send an audio request: ```python import base64, requests with open("test.wav", "rb") as f: audio_b64 = base64.b64encode(f.read()).decode() resp = requests.post("http://localhost:11434/api/chat", json={ "model": "openbmb/minicpm-o4.5", "messages": [{"role": "user", "content": "What do you hear?", "audios": [audio_b64]}], "stream": False, }) print(resp.status_code, resp.text[:300]) ``` ## Error ``` HTTP 500 {"error":"llama runner process has terminated: GGML_ASSERT([rsets->data count] == 0) failed\nWARNING: Using native backtrace. Set GGML_BACKTRACE_LLDB for more info.\nWARNING: GGML_BACKTRACE_LLDB may cause native MacOS Terminal.app to crash.\nSee: https://github.com/ggml-org/llama.cpp/pull/17869"} ``` The crash happens with any audio file (tested synthetic sine wave, real speech WAV files at 16kHz mono). ## Expected Behavior The model should process audio input and return a text response, same as it does for image/text inputs. ## Workaround Audio works correctly via [llama.cpp-omni](https://github.com/tc-mb/llama.cpp-omni) (the fork built specifically for MiniCPM-o), which handles Metal acceleration for audio without this assertion failure. The issue appears to be in Ollama's Metal backend when processing audio embeddings. ## Environment - **Ollama version**: 0.17.4 - **OS**: macOS 26.2 (Build 25C56) - **Hardware**: Apple M4, 16GB unified memory - **Model**: `openbmb/minicpm-o4.5:latest` (6.1 GB, Q4_0) ## Related - #14065 — MiniCPM-o 4.5 support request - #11798 — Audio input support for multimodal models - The assertion `[rsets->data count] == 0` is in the Metal residency set handling code
GiteaMirror commented 2026-05-05 01:51:18 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Feb 28, 2026):

Ollama doesn't currently support audio models.

<!-- gh-comment-id:3977670094 --> @rick-github commented on GitHub (Feb 28, 2026): Ollama doesn't currently support audio models.
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#71479
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?