forked from ggml-org/llama.cpp
-
Notifications
You must be signed in to change notification settings - Fork 5
Open
Description
Name and Version
version: 7990 (5abd9666d)
built with AppleClang 17.0.0.17000404 for Darwin arm64
Operating systems
Mac
GGML backends
Metal
Hardware
Apple M4 Max, 40 GPU cores, 64G Unified Memory
Models
Huggingface unsloth/Qwen3-Coder-Next:MXFP4_MOE
Problem description & steps to reproduce
With some MCP tools enabled (namely Chrome Devtools MCP) and using Opencode, llama-server instantly crashes with initial prompt.
Automatic parser generation failed: [json.exception.type_error.302] type must be string, but is array
llama.cpp/common/chat.cpp:1260: Unable to generate parser for this template.
I dug into LLDB backtraces and found out that the crashing point is
| auto type = param_schema.value("type", "object"); |
type be "nullable" and expresses it as ["<type>", null] where the autoparser expects to be string. Unhandled type error occurs.
I didn't followed up how the autoparser's strategy will be when nullability is allowed, so I didn't try to implement a fix on my own. Maybe just recognize if it's nullable then silently ignore it?
First Bad Commit
No response
Relevant log output
Logs
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels