Skip to content

Misc. bug: Starting llama-server.exe with --host 0.0.0.0 results in server not working, ok if only running over 127.0.0.1, for --models-dir feature. #17862

Description

@alan-l

Name and Version

Module: llama-server.exe (b7315)

Operating systems

Windows

Which llama.cpp modules do you know to be affected?

llama-server

Command line

llama-server.exe --host 0.0.0.0 --port 8080 --n-gpu-layers 0 --models-max 1 --models-dir "C:\Models\MistralAI"

Problem description & steps to reproduce

Params: --host 0.0.0.0 --port 8080 --n-gpu-layers 0 --models-max 1 --models-dir "C:\Models\MistralAI"
Error:

Starts ok:
    init: using 6 threads for HTTP server
    main: starting router server, no model will be loaded in this process
    start: binding port with default address family
    main: router server is listening on http://0.0.0.0:8080
    main: NOTE: router mode is experimental
    main:       it is not recommended to use this mode in untrusted environments

Later after selecting model from webgui dropdown:
    [60024] main: model loaded
    [60024] main: server is listening on http://0.0.0.0:60024
    [60024] main: starting the main loop...
    [60024] srv  setup_child_: notifying router server (port=8080) that model Ministral-3-3B-Instruct-2512-GGUF is ready
    [60024] srv  setup_child_: failed to notify router server: Could not establish connection

First Bad Commit

Started when this feature was first introduced. I didn't report it earlier because I thought you guys may catch it during testing for later versions.

Relevant log output

Metadata

Metadata

Assignees

Labels

bugSomething isn't workingmedium severityUsed to report medium severity bugs in llama.cpp (e.g. Malfunctioning Features but still useable)regressionA regression introduced in a new build (something that was previously working correctly)serverserver/api

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions