Fail to call Qwen3-Coder model with tool calling enabled

[ontecAI]

Before submitting your bug report

• I've tried using the "Ask AI" feature on the Continue docs site to see if the docs have an answer
• I'm not able to find a related conversation on GitHub discussions that reports the same bug
• I'm not able to find an open issue that reports the same bug
• I've seen the troubleshooting guide on the Continue Docs

Relevant environment info

- OS: Windows 11
- Continue version: 1.2.10
- IDE version: VS Code
- Model: Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8
- config:
  
  - name: qwen3-coder-30b
    provider: vllm
    model: Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8
    apiBase: http://ai01-dx-mr300-b04.dev1.xxx.at:7777/v1/
    apiKey: xxxx
    roles:
      - apply
      - chat
      - edit
    capabilities:
      - tool_use

  - name: qwen3-coder-30b-litellm
    provider: openai
    model: qwen3-coder-30b-fp8
    apiBase: https://litellm.dev1.xxx.at/v1/
    apiKey: xxxx
    roles:
      - apply
      - chat
      - edit
    capabilities:
      - tool_use
  
  OR link to agent in Continue hub:

Description

[@continuedev] error: Error streaming response: list index out of range {"context":"llm_stream_chat","model":"Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8","provider":"vllm","useOpenAIAdapter":true,"streamEnabled":true,"templateMessages":true}

[Extension Host] Error: Error streaming response: list index out of range
at parseDataLine (c:\Users\mbuchner.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:131119:15)
at parseSseLine (c:\Users\mbuchner.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:131136:33)
at streamSse (c:\Users\mbuchner.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:131151:37)
at process.processTicksAndRejections (node:internal/process/task_queues:105:5)
at async Vllm2._streamChat (c:\Users\mbuchner.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:244427:26)
at async Vllm2._streamComplete (c:\Users\mbuchner.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:244338:26)
at async Vllm2.streamChat (c:\Users\mbuchner.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:243942:30)
at async llmStreamChat (c:\Users\mbuchner.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:722153:17)
at async Wd.handleMessage [as value] (c:\Users\mbuchner.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:760618:27)

To reproduce

Chat with the model - think it happens when tool calling ...

vllm runtime - docker compose

services:
  vllm_openai:
    image: vllm/vllm-openai:v0.11.0
    restart: unless-stopped
    runtime: nvidia
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [ gpu ]
    volumes:
      - /opt/projects/.cache:/root/.cache/huggingface # the model is saved here
    ipc: host
    environment:
      HUGGING_FACE_HUB_TOKEN: hf_xxxx
    ports:
      - 7777:8000
    networks:
      - traefik-public
    command:
      - --model
      - "Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8"
      - --api-key
      - "xxx"
      - --gpu-memory-utilization
      - "0.90"
      - --kv-cache-dtype
      - "fp8"
      - --enable-chunked-prefill
      - --swap-space
      - "16"
      - --trust-remote-code
      - --disable-log-request
      - --tensor-parallel-size
      - "1"        # should be same as the number of GPUs
      - --enable-auto-tool-choice       # enable tools
      #- --tool-call-parser
      #- "qwen3_xml"    # see vllm docs - recomended setting for qwen3 coder
      - --max-num-seqs
      - "4"              # batch size for throughput - decrease = lesser memory (as huge context size set it to 16 - default 128)

Log output

[@continuedev] error: Error streaming response: list index out of range {"context":"llm_stream_chat","model":"Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8","provider":"vllm","useOpenAIAdapter":true,"streamEnabled":true,"templateMessages":true}

[Extension Host] Error: Error streaming response: list index out of range
	at parseDataLine (c:\Users\mbuchner\.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:131119:15)
	at parseSseLine (c:\Users\mbuchner\.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:131136:33)
	at streamSse (c:\Users\mbuchner\.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:131151:37)
	at process.processTicksAndRejections (node:internal/process/task_queues:105:5)
	at async Vllm2._streamChat (c:\Users\mbuchner\.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:244427:26)
	at async Vllm2._streamComplete (c:\Users\mbuchner\.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:244338:26)
	at async Vllm2.streamChat (c:\Users\mbuchner\.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:243942:30)
	at async llmStreamChat (c:\Users\mbuchner\.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:722153:17)
	at async Wd.handleMessage [as value] (c:\Users\mbuchner\.vscode\extensions\continue.continue-1.2.10-win32-x64\out\extension.js:760618:27)

[goodnessnwajichukwu]

Hey team, Continue.dev looks impressive! I noticed some features aren’t explained clearly in the docs, especially for users new to AI coding assistants.
I specialize in documentation for dev tools happy to help improve clarity and flow. Want a quick outline?

[axe-wood]

you can try this.

• name: Qwen-coder-flash
  provider: openrouter
  apiBase: https://dashscope.aliyuncs.com/compatible-mode/v1
  apiKey: xxxxx
  model: qwen3-coder-flash

[emcpoh]

I have the same issue with Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8 + vLLM in both VS Code and JetBrains DataGrip. Does anyone have a workaround other than using OpenRouter?

[karoldydo]

Hello,

I was thinking about reporting this issue, but I wasn’t sure whether it was a model problem or a Continue issue, so I’ll provide a bit of context:

Environment:

• GMKTEC EVO-X2 with AMD Ryzen AI Max+ 395 processor, 128 GB RAM, and 2 TB SSD
• Ubuntu 24.04 LTS Server
• ROCm 7.1 for AMD GPU acceleration

Software:

• WebUI (latest version)
• Ollama (latest version)

Both of them are publicly accessible via Cloudflare Tunnel, and I’m using Ollama over HTTPS.

Continue config:

---
name: hf.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q8_0
version: 1.1.0
schema: v1
models:
  - name: hf.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q8_0
    provider: ollama
    model: hf.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q8_0
    apiBase: https://llm.mydomain.com
    apiKey: sk-XXX
    defaultCompletionOptions:
      contextLength: 262144
    roles:
      - apply
      - chat
      - edit
    capabilities:
      - tool_use

Tested models:

• qwen3-coder:30b
• qwen3-coder:30b-a3b-q8_0
• hf.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q8_0

Behavior:

• The agent performs its tasks and generates code, but does not apply changes.
• Manual changes do not reset Continue’s state, and after applying changes I can still see the “Accept/Reject” buttons.
• The agent spinner next to the task never stops.

Tested in IDEs:

• Cursor
• Visual Studio Code

Videos (OneDrive):

• https://1drv.ms/f/c/a2a08a4f067ae07f/EgzDLg4rI-NMgQMo9TmqQoIBhRNZKjTmxhGHpM3QGvBrUQ?e=9Y5aFJ

[winmutt]

I have the same issue with any of the qwen3-coder models and ollama. Continue will not read files, edit files or create files. Using cline and its working fine.