Failed to parse grammar - opencode

When I try to connect opencode with the docker model I only get this answer:
Failed to parse grammar

My open code configuration is:

"dmr": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Docker Model Runner",
      "options": {
        "baseURL": "http://localhost:12434/engines/v1"
      },
      "models": {
        "ai/qwen3-coder": {
          "name": "Qwen3 Coder"
        }
      }
    }

I use a macbook pro and have installed Docker model with version: Docker Model Runner version v1.0.2

I also tried it using my new gigabyte AI TOP ATOM.

"dmr": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Docker Model Runner",
      "options": {
        "baseURL": "http://localhost:12434/engines/v1"
      },
      "models": {
        "ai/qwen3-coder": {
          "name": "Qwen3 Coder"
        }
      }
    }

With Docker Model Runner version v1.0.6

I followed the instruction based on: www.docker.com/blog/new-nvidia-spark-docker-model-runner/

I hope someone can help me out

@remcocats

Could you build and test this when it's complete @remcocats ?

ericcurtin#80

copilot doesn't always understand or fix the issue. But it's worth a shot before a deeper debug session.

I will give it a try

I have tried the fix it didn't help. But I tried something else as well. I removed the mcp's from opencode and that helped.

Interesting, thanks for trying, might be worth opening a documentation PR here:

https://github.com/sst/opencode

the output log from the fix with mcp is as follows:

INFO[0064] root-tool-call-multiple-with-end ::= root-tool-call-once root-tool-call-more* root-call-end  component=llama.cpp
INFO[0064] root-tool-call-once ::= bash-call | read-call | glob-call | grep-call | list-call | edit-call | write-call | task-call | webfetch-call | todowrite-call | todoread-call | skill-call | playwright-browser-close-call | playwright-browser-resize-call | playwright-browser-console-messages-call | playwright-browser-handle-dialog-call | playwright-browser-evaluate-call | playwright-browser-file-upload-call | playwright-browser-fill-form-call | playwright-browser-install-call | playwright-browser-press-key-call | playwright-browser-type-call | playwright-browser-navigate-call | playwright-browser-navigate-back-call | playwright-browser-network-requests-call | playwright-browser-take-screenshot-call | playwright-browser-snapshot-call | playwright-browser-click-call | playwright-browser-drag-call | playwright-browser-hover-call | playwright-browser-select-option-call | playwright-browser-tabs-call | playwright-browser-wait-for-call | duckduckgo-search-call | duckduckgo-fetch-content-call | backlog-get-workflow-overview-call | backlog-get-task-creation-guide-call | backlog-get-task-execution-guide-call | backlog-get-task-completion-guide-call | backlog-task-create-call | backlog-task-list-call | backlog-task-search-call | backlog-task-edit-call | backlog-task-view-call | backlog-task-archive-call | backlog-task-complete-call | backlog-milestone-list-call | backlog-milestone-add-call | backlog-milestone-rename-call | backlog-milestone-remove-call | backlog-document-list-call | backlog-document-view-call | backlog-document-create-call | backlog-document-update-call | backlog-document-search-call | context7-resolve-library-id-call | context7-get-library-docs-call  component=llama.cpp
INFO[0064] skill-arg-after-0 ::= func-skill-kv-name skill-last-arg-end  component=llama.cpp
INFO[0064] skill-arg-after-0-with-sep ::= "\n</parameter>\n" func-skill-kv-name skill-last-arg-end  component=llama.cpp
INFO[0064] skill-call ::= "<function=" "skill" ">\n" skill-arg-after-0  component=llama.cpp
INFO[0064] skill-last-arg-end ::= "\n</parameter>\n"     component=llama.cpp
INFO[0064] space ::= | " " | "\n"{1,2} [ \t]{0,20}       component=llama.cpp
INFO[0064] string ::= "\"" char* "\"" space              component=llama.cpp
INFO[0064] string-arg-val ::= ( [^\x0A] | "\x0A" ( [^<] | "<" ( [^/] | "/" ( [^p] | "p" ( [^a] | "a" ( [^r] | "r" ( [^a] | "a" ( [^m] | "m" ( [^e] | "e" ( [^t] | "t" ( [^e] | "e" ( [^r] | "r" ( [^>] | ">" ( [^\x0A] ) ) ) ) ) ) ) ) ) ) ) ) ) )*  component=llama.cpp
INFO[0064] task-arg-after-0 ::= func-task-kv-description task-arg-after-1-with-sep  component=llama.cpp
INFO[0064] task-arg-after-0-with-sep ::= "\n</parameter>\n" func-task-kv-description task-arg-after-1-with-sep  component=llama.cpp
INFO[0064] task-arg-after-1 ::= func-task-kv-prompt task-arg-after-2-with-sep  component=llama.cpp
INFO[0064] task-arg-after-1-with-sep ::= "\n</parameter>\n" func-task-kv-prompt task-arg-after-2-with-sep  component=llama.cpp
INFO[0064] task-arg-after-2 ::= func-task-kv-subagent-type task-arg-after-3-with-sep  component=llama.cpp
INFO[0064] task-arg-after-2-with-sep ::= "\n</parameter>\n" func-task-kv-subagent-type task-arg-after-3-with-sep  component=llama.cpp
INFO[0064] task-arg-after-3 ::= ( func-task-kv-session-id task-arg-after-4-with-sep ) | task-arg-after-4  component=llama.cpp
INFO[0064] task-arg-after-3-with-sep ::= ( "\n</parameter>\n" func-task-kv-session-id task-arg-after-4-with-sep ) | task-arg-after-4-with-sep  component=llama.cpp
INFO[0064] task-arg-after-4 ::= ( func-task-kv-command task-last-arg-end ) | ""  component=llama.cpp
INFO[0064] task-arg-after-4-with-sep ::= ( "\n</parameter>\n" func-task-kv-command task-last-arg-end ) | task-last-arg-end  component=llama.cpp
INFO[0064] task-call ::= "<function=" "task" ">\n" task-arg-after-0  component=llama.cpp
INFO[0064] task-last-arg-end ::= "\n</parameter>\n"      component=llama.cpp
INFO[0064] todoread-call ::= "<function=" "todoread" ">\n" ""  component=llama.cpp
INFO[0064] todoread-last-arg-end ::= "\n</parameter>\n"  component=llama.cpp
INFO[0064] todowrite-arg-after-0 ::= func-todowrite-kv-todos todowrite-last-arg-end  component=llama.cpp
INFO[0064] todowrite-arg-after-0-with-sep ::= "\n</parameter>\n" func-todowrite-kv-todos todowrite-last-arg-end  component=llama.cpp
INFO[0064] todowrite-arg-todos ::= "[" space (todowrite-arg-todos-item ("," space todowrite-arg-todos-item)*)? "]" space  component=llama.cpp
INFO[0064] todowrite-arg-todos-item ::= "{" space todowrite-arg-todos-item-content-kv "," space todowrite-arg-todos-item-status-kv "," space todowrite-arg-todos-item-priority-kv "," space todowrite-arg-todos-item-id-kv "}" space  component=llama.cpp
INFO[0064] todowrite-arg-todos-item-content-kv ::= "\"content\"" space ":" space string  component=llama.cpp
INFO[0064] todowrite-arg-todos-item-id-kv ::= "\"id\"" space ":" space string  component=llama.cpp
INFO[0064] todowrite-arg-todos-item-priority-kv ::= "\"priority\"" space ":" space string  component=llama.cpp
INFO[0064] todowrite-arg-todos-item-status-kv ::= "\"status\"" space ":" space string  component=llama.cpp
INFO[0064] todowrite-call ::= "<function=" "todowrite" ">\n" todowrite-arg-after-0  component=llama.cpp
INFO[0064] todowrite-last-arg-end ::= "\n</parameter>\n"  component=llama.cpp
INFO[0064] webfetch-arg-after-0 ::= func-webfetch-kv-url webfetch-arg-after-1-with-sep  component=llama.cpp
INFO[0064] webfetch-arg-after-0-with-sep ::= "\n</parameter>\n" func-webfetch-kv-url webfetch-arg-after-1-with-sep  component=llama.cpp
INFO[0064] webfetch-arg-after-1 ::= func-webfetch-kv-format webfetch-arg-after-2-with-sep  component=llama.cpp
INFO[0064] webfetch-arg-after-1-with-sep ::= "\n</parameter>\n" func-webfetch-kv-format webfetch-arg-after-2-with-sep  component=llama.cpp
INFO[0064] webfetch-arg-after-2 ::= ( func-webfetch-kv-timeout webfetch-last-arg-end ) | ""  component=llama.cpp
INFO[0064] webfetch-arg-after-2-with-sep ::= ( "\n</parameter>\n" func-webfetch-kv-timeout webfetch-last-arg-end ) | webfetch-last-arg-end  component=llama.cpp
INFO[0064] webfetch-arg-format ::= ("\"text\"" | "\"markdown\"" | "\"html\"") space  component=llama.cpp
INFO[0064] webfetch-call ::= "<function=" "webfetch" ">\n" webfetch-arg-after-0  component=llama.cpp
INFO[0064] webfetch-last-arg-end ::= "\n</parameter>\n"  component=llama.cpp
INFO[0064] write-arg-after-0 ::= func-write-kv-content write-arg-after-1-with-sep  component=llama.cpp
INFO[0064] write-arg-after-0-with-sep ::= "\n</parameter>\n" func-write-kv-content write-arg-after-1-with-sep  component=llama.cpp
INFO[0064] write-arg-after-1 ::= func-write-kv-filePath write-last-arg-end  component=llama.cpp
INFO[0064] write-arg-after-1-with-sep ::= "\n</parameter>\n" func-write-kv-filePath write-last-arg-end  component=llama.cpp
INFO[0064] write-call ::= "<function=" "write" ">\n" write-arg-after-0  component=llama.cpp
INFO[0064] write-last-arg-end ::= "\n</parameter>\n"     component=llama.cpp
INFO[0064]                                               component=llama.cpp
INFO[0064] llama_grammar_init_impl: failed to parse grammar  component=llama.cpp
INFO[0064] srv    send_error: task id = 0, error: Failed to parse grammar  component=llama.cpp
INFO[0064] srv  process_sing: failed to launch slot with task, id_task = 0  component=llama.cpp
INFO[0064] srv  update_slots: all slots are idle         component=llama.cpp
INFO[0064] srv          stop: cancel task, id_task = 0   component=llama.cpp
INFO[0064] srv  update_slots: all slots are idle         component=llama.cpp
INFO[0064] srv  log_server_r: request: POST /v1/chat/completions  400  component=llama.cpp

the tools works for lmstudio and ollama implementation

Are you interested in working on this? It's:

make && ./model-runner serve --host 0.0.0.0 --port 13434

to run a new DMR server to test. Would be a nice contribution.

Hi I can try, but I'm not a go specialist, my background is Java, typescript,

btw, I have narrowed it to one specific MCP server that is the backlog.md MCP server.

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "backlog": {
      "type": "local",
      "command": [ "backlog",
        "mcp",
        "start"
      ],
      "enabled": true
    }
  }
}

After some digging the issue is inside the docker-model-backend-laamacpp. It's the way how the grammer is parsed for the tooling.

That's this code:

https://github.com/docker/model-runner/tree/main/llamacpp

@remcocats

@remcocats what platform are you on? The first thing to try would be to see if this occurs with the latest llama.cpp and vLLM . Then update in DMR if it's just an update.

Also worth checking if it works ok with this model:

https://hf.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF

@remcocats what platform are you on? The first thing to try would be to see if this occurs with the latest llama.cpp and vLLM . Then update in DMR if it's just an update.

I use a nvidia spark. I now use a vllm container with ftp-Oss-120b that works now I haven’t got the time to go deeper in the investigation. My main programming language is Java and not python

Ok, please reopen if you encounter again

Navigation Menu

Repository navigation

Failed to parse grammar - opencode #534

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Failed to parse grammar - opencode #534

Description

Activity

ericcurtin commented on Dec 24, 2025

remcocats commented on Dec 24, 2025

remcocats commented on Dec 24, 2025

ericcurtin commented on Dec 24, 2025

remcocats commented on Dec 24, 2025

remcocats commented on Dec 24, 2025

ericcurtin commented on Dec 24, 2025

remcocats commented on Dec 24, 2025

remcocats commented on Dec 24, 2025

ericcurtin commented on Dec 24, 2025

ericcurtin commented on Jan 11, 2026

ericcurtin commented on Jan 11, 2026

remcocats commented on Jan 11, 2026

ericcurtin commented on Jan 11, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions