[GH-ISSUE #1946] SIGSEGV: segmentation violation when shutting down server with ctrl+c #1120

Closed
opened 2026-04-12 10:52:00 -05:00 by GiteaMirror · 2 comments
Owner

Originally created by @jmorganca on GitHub (Jan 12, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/1946

[GIN] 2024/01/12 - 12:38:39 | 200 |  5.985573917s |       127.0.0.1 | POST     "/api/chat"
2024/01/12 12:38:52 ext_server_common.go:158: loaded 0 images
^Cggml_metal_free: deallocating
SIGSEGV: segmentation violation
Originally created by @jmorganca on GitHub (Jan 12, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/1946 ``` [GIN] 2024/01/12 - 12:38:39 | 200 | 5.985573917s | 127.0.0.1 | POST "/api/chat" 2024/01/12 12:38:52 ext_server_common.go:158: loaded 0 images ^Cggml_metal_free: deallocating SIGSEGV: segmentation violation ```
GiteaMirror added the bug label 2026-04-12 10:52:00 -05:00
Author
Owner

@dcasota commented on GitHub (Jan 12, 2024):

As a volunteer feedback on this - after ollama-0.1.20, ext_server_common.go isn't part of the repo anymore, right? On an ollama make build with latest source I haven't seen a segmentation violation.

Btw. multiple llm servers using the same application instance would be on top of gpu driver capability. There are multi-instance gpu limitations, e.g. for nvidia in terms of driver mode (wddm/tcc) and its availability per gpu card. A baremetal prerequirement cascade might look like [cuda-capable gpu + baremetal, os version, cuda driver type and version, gcc version + correct development packages, application].

<!-- gh-comment-id:1889063231 --> @dcasota commented on GitHub (Jan 12, 2024): As a volunteer feedback on this - after ollama-0.1.20, ext_server_common.go isn't part of the repo anymore, right? On an ollama [make build](https://github.com/jmorganca/ollama/files/13918586/build_output.txt) with latest source I haven't seen a segmentation violation. Btw. multiple llm servers using the same application instance would be on top of gpu driver capability. There are multi-instance gpu limitations, e.g. for nvidia in terms of driver mode (wddm/tcc) and its availability per gpu card. A baremetal prerequirement cascade might look like [cuda-capable gpu + baremetal, os version, cuda driver type and version, gcc version + correct development packages, application].
Author
Owner

@dhiltgen commented on GitHub (Mar 12, 2024):

Pretty sure this has been resolved via llama.cpp bumps or other fixes we've done since early Jan. I get a clean shutdown on mac with the latest code.

<!-- gh-comment-id:1992274201 --> @dhiltgen commented on GitHub (Mar 12, 2024): Pretty sure this has been resolved via llama.cpp bumps or other fixes we've done since early Jan. I get a clean shutdown on mac with the latest code.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#1120