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 #14611] [0.17.5][macOS Apple Silicon] model runner unexpectedly stopped (EOF/exit status 2) on /api/generate #55980

New Issue
Open
opened 2026-04-29 10:06:10 -05:00 by GiteaMirror · 8 comments
GiteaMirror commented 2026-04-29 10:06:10 -05:00
Owner
Copy Link

Originally created by @ksw9722 on GitHub (Mar 4, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14611

What happened

On macOS (Apple Silicon), Ollama API health endpoints are reachable, but inference calls intermittently fail with:

  • {"error":"model runner has unexpectedly stopped, this may be due to resource limitations or an internal error"}
  • HTTP 500 from /api/generate
  • server log shows post predict ... EOF and llama runner terminated: exit status 2

This happened repeatedly during local automation workloads.

Environment

  • Ollama: 0.17.5
  • OS: macOS 26.2 (arm64)
  • Hardware: Apple M4 Pro, 64GB RAM
  • API endpoint: http://127.0.0.1:11434
  • Example model: gpt-oss:20b

Reproduction

  1. Start a clean server: 
    pkill -9 -f 'ollama runner' || true
pkill -9 -f '/usr/local/bin/ollama serve' || true
pkill -9 -f '/opt/homebrew/bin/ollama serve' || true
ollama serve
  2. Confirm API health: 
    curl -sS http://127.0.0.1:11434/api/version
  3. Call generate: 
    curl -sS http://127.0.0.1:11434/api/generate -d '{
  "model":"gpt-oss:20b",
  "prompt":"hi",
  "stream":false,
  "options":{"num_ctx":512,"num_predict":8}
}'
  4. Sometimes response is: 
    {"error":"model runner has unexpectedly stopped, this may be due to resource limitations or an internal error, check ollama server logs for details"}

Relevant logs

From runner/server logs around failure:

time=2026-03-04T17:10:17.873+09:00 level=ERROR source=server.go:1611 msg="post predict" error="Post \"http://127.0.0.1:53947/completion\": EOF"
[GIN] 2026/03/04 - 17:10:17 | 500 |  6.983919458s |       127.0.0.1 | POST     "/api/generate"
time=2026-03-04T17:10:17.873+09:00 level=ERROR source=server.go:303 msg="llama runner terminated" error="exit status 2"

There is also a native crash dump section with register dump in the same failure window.

Expected behavior

/api/generate should return a normal completion or a stable, actionable error without runner termination.

Notes

  • This still occurred after restarting Ollama and rebooting macOS.
  • We also reproduced with conservative options (num_ctx low, num_predict low).
Originally created by @ksw9722 on GitHub (Mar 4, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14611 ## What happened On macOS (Apple Silicon), Ollama API health endpoints are reachable, but inference calls intermittently fail with: - `{"error":"model runner has unexpectedly stopped, this may be due to resource limitations or an internal error"}` - HTTP 500 from `/api/generate` - server log shows `post predict ... EOF` and `llama runner terminated: exit status 2` This happened repeatedly during local automation workloads. ## Environment - Ollama: `0.17.5` - OS: macOS 26.2 (arm64) - Hardware: Apple M4 Pro, 64GB RAM - API endpoint: `http://127.0.0.1:11434` - Example model: `gpt-oss:20b` ## Reproduction 1. Start a clean server: ```bash pkill -9 -f 'ollama runner' || true pkill -9 -f '/usr/local/bin/ollama serve' || true pkill -9 -f '/opt/homebrew/bin/ollama serve' || true ollama serve ``` 2. Confirm API health: ```bash curl -sS http://127.0.0.1:11434/api/version ``` 3. Call generate: ```bash curl -sS http://127.0.0.1:11434/api/generate -d '{ "model":"gpt-oss:20b", "prompt":"hi", "stream":false, "options":{"num_ctx":512,"num_predict":8} }' ``` 4. Sometimes response is: ```json {"error":"model runner has unexpectedly stopped, this may be due to resource limitations or an internal error, check ollama server logs for details"} ``` ## Relevant logs From runner/server logs around failure: ```text time=2026-03-04T17:10:17.873+09:00 level=ERROR source=server.go:1611 msg="post predict" error="Post \"http://127.0.0.1:53947/completion\": EOF" [GIN] 2026/03/04 - 17:10:17 | 500 | 6.983919458s | 127.0.0.1 | POST "/api/generate" time=2026-03-04T17:10:17.873+09:00 level=ERROR source=server.go:303 msg="llama runner terminated" error="exit status 2" ``` There is also a native crash dump section with register dump in the same failure window. ## Expected behavior `/api/generate` should return a normal completion or a stable, actionable error without runner termination. ## Notes - This still occurred after restarting Ollama and rebooting macOS. - We also reproduced with conservative options (`num_ctx` low, `num_predict` low).
GiteaMirror commented 2026-04-29 10:06:13 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 4, 2026):

Set OLLAMA_DEBUG=1 and post the full log.

<!-- gh-comment-id:3995970317 --> @rick-github commented on GitHub (Mar 4, 2026): Set `OLLAMA_DEBUG=1` and post the full log.
GiteaMirror commented 2026-04-29 10:06:14 -05:00
Author
Owner
Copy Link

@ksw9722 commented on GitHub (Mar 4, 2026):

Thanks — I enabled debug and captured a fresh repro run.

Environment

  • macOS 26.2 (Apple M4 Pro, 64 GB)
  • Ollama 0.17.5

Reproduction Request

curl -sS http://127.0.0.1:11434/api/generate -d '{
  "model":"gpt-oss:20b",
  "prompt":"hi",
  "stream":false,
  "options":{"num_ctx":512,"num_predict":8}
}'

Observed Error

{"error":"model runner has unexpectedly stopped, this may be due to resource limitations or an internal error, check ollama server logs for details"}

Relevant Debug Lines

time=2026-03-04T17:10:17.873+09:00 level=ERROR source=server.go:1611 msg="post predict" error="Post \"http://127.0.0.1:53947/completion\": EOF"
[GIN] 2026/03/04 - 17:10:17 | 500 | 6.98s | 127.0.0.1 | POST "/api/generate"
time=2026-03-04T17:10:17.873+09:00 level=ERROR source=server.go:303 msg="llama runner terminated" error="exit status 2"

There is also a native crash dump section in the same window (register dump + goroutine dump). If useful, I can upload a full sanitized debug chunk.

<!-- gh-comment-id:3996004011 --> @ksw9722 commented on GitHub (Mar 4, 2026): Thanks — I enabled debug and captured a fresh repro run. ## Environment - macOS 26.2 (Apple M4 Pro, 64 GB) - Ollama 0.17.5 ## Reproduction Request ```bash curl -sS http://127.0.0.1:11434/api/generate -d '{ "model":"gpt-oss:20b", "prompt":"hi", "stream":false, "options":{"num_ctx":512,"num_predict":8} }' ``` ## Observed Error ```json {"error":"model runner has unexpectedly stopped, this may be due to resource limitations or an internal error, check ollama server logs for details"} ``` ## Relevant Debug Lines ```text time=2026-03-04T17:10:17.873+09:00 level=ERROR source=server.go:1611 msg="post predict" error="Post \"http://127.0.0.1:53947/completion\": EOF" [GIN] 2026/03/04 - 17:10:17 | 500 | 6.98s | 127.0.0.1 | POST "/api/generate" time=2026-03-04T17:10:17.873+09:00 level=ERROR source=server.go:303 msg="llama runner terminated" error="exit status 2" ``` There is also a native crash dump section in the same window (register dump + goroutine dump). If useful, I can upload a full sanitized debug chunk.
GiteaMirror commented 2026-04-29 10:06:14 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 4, 2026):

Full log.

<!-- gh-comment-id:3996022794 --> @rick-github commented on GitHub (Mar 4, 2026): Full log.
GiteaMirror commented 2026-04-29 10:06:15 -05:00
Author
Owner
Copy Link

@ksw9722 commented on GitHub (Mar 4, 2026):

Uploaded full sanitized debug logs (OLLAMA_DEBUG=1 context) here:\n\nhttps://gist.github.com/ksw9722/acf0dbdb7898cf7b8c8b94518ec84fe6\n\nThis includes:\n- manual-serve.log\n- server.log tail\n- full repro serve.log with crash window\n\nIf you want, I can also provide a minimal subset focused only on the exact crash timestamp window.

<!-- gh-comment-id:3996023069 --> @ksw9722 commented on GitHub (Mar 4, 2026): Uploaded full sanitized debug logs (OLLAMA_DEBUG=1 context) here:\n\nhttps://gist.github.com/ksw9722/acf0dbdb7898cf7b8c8b94518ec84fe6\n\nThis includes:\n- manual-serve.log\n- server.log tail\n- full repro serve.log with crash window\n\nIf you want, I can also provide a minimal subset focused only on the exact crash timestamp window.
GiteaMirror commented 2026-04-29 10:06:15 -05:00
Author
Owner
Copy Link

@ksw9722 commented on GitHub (Mar 4, 2026):

Good point — added a file-style reference for convenience:

If preferred, I can split this into smaller per-phase files (startup / reproduce / crash window).

<!-- gh-comment-id:3996024884 --> @ksw9722 commented on GitHub (Mar 4, 2026): Good point — added a file-style reference for convenience: - **ollama-debug-full-sanitized.log** (sanitized) - Gist file: https://gist.github.com/ksw9722/acf0dbdb7898cf7b8c8b94518ec84fe6#file-ollama-debug-full-sanitized-log - Raw: https://gist.githubusercontent.com/ksw9722/acf0dbdb7898cf7b8c8b94518ec84fe6/raw/ollama-debug-full-sanitized.log If preferred, I can split this into smaller per-phase files (startup / reproduce / crash window).
GiteaMirror commented 2026-04-29 10:06:16 -05:00
Author
Owner
Copy Link

@ksw9722 commented on GitHub (Mar 6, 2026):

Update from local repro on the same machine (Apple Silicon/macOS): setting options.num_gpu=0 makes /api/generate succeed consistently (CPU path), while default GPU path can still hit runner termination.

Example payload that works for us:

{
  "model": "gpt-oss:20b",
  "prompt": "hi",
  "stream": false,
  "options": {
    "num_ctx": 512,
    "num_predict": 8,
    "num_gpu": 0
  }
}

So at least on our side this looks like a GPU-path-specific issue; num_gpu=0 is a temporary workaround.

<!-- gh-comment-id:4013556973 --> @ksw9722 commented on GitHub (Mar 6, 2026): Update from local repro on the same machine (Apple Silicon/macOS): setting `options.num_gpu=0` makes `/api/generate` succeed consistently (CPU path), while default GPU path can still hit runner termination. Example payload that works for us: ```json { "model": "gpt-oss:20b", "prompt": "hi", "stream": false, "options": { "num_ctx": 512, "num_predict": 8, "num_gpu": 0 } } ``` So at least on our side this looks like a GPU-path-specific issue; `num_gpu=0` is a temporary workaround.
GiteaMirror commented 2026-04-29 10:06:17 -05:00
Author
Owner
Copy Link

@alexgeek commented on GitHub (Apr 1, 2026):

Just an FYI if anyone else is stuck with this, I removed -DGGML_METAL_HAS_BF16 and the models are running again. Haven't found a way ti get it working with that flag still enabled.

<!-- gh-comment-id:4172511891 --> @alexgeek commented on GitHub (Apr 1, 2026): Just an FYI if anyone else is stuck with this, I removed `-DGGML_METAL_HAS_BF16` and the models are running again. Haven't found a way ti get it working with that flag still enabled.
GiteaMirror commented 2026-04-29 10:06:17 -05:00
Author
Owner
Copy Link

@alexgeek commented on GitHub (Apr 1, 2026):

Ah seems to be fixed by this https://github.com/ollama/ollama/pull/14604

<!-- gh-comment-id:4173202631 --> @alexgeek commented on GitHub (Apr 1, 2026): Ah seems to be fixed by this https://github.com/ollama/ollama/pull/14604
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#55980
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?