bug: chat-ui (Docker) crashes after reasoning-enabled QwQ model inference (ECONNREFUSED 0.0.0.0:443)

[dosomethingbyme]

Bug description

When using the QwQ model via Ollama in Chat UI with reasoning enabled (<think>...</think>), the chat-ui Docker container crashes after completing a single inference. The logs report a fetch failed: connect ECONNREFUSED 0.0.0.0:443 error, followed by a fatal Generation failed exception.

The model successfully returns the response, but then the container exits unexpectedly. This behavior is consistent and reproducible.

This issue appears to be related to a post-inference process—possibly related to moderation or summary generation logic—attempting to reach 0.0.0.0:443, which is clearly invalid.

Steps to reproduce

1. Configure .env.local with the QwQ model and reasoning enabled:

   MODELS=[{
     "name": "QwQ 32B",
     "chatPromptTemplate": "{{#each messages}}{{#ifUser}}<|im_start|>user\n{{content}}<|im_end|>\n{{/ifUser}}{{#ifAssistant}}<|im_start|>assistant\n{{content}}<|im_end|>\n{{/ifAssistant}}{{/each}}<|im_start|>assistant\n",
     "parameters": {
       "temperature": 0.7,
       "top_p": 0.9,
       "repetition_penalty": 1.1,
       "top_k": 50,
       "truncate": 3072,
       "max_new_tokens": 10240,
       "stop": ["<|im_end|>"]
     },
     "endpoints": [{
       "type": "ollama",
       "url": "http://172.30.0.1:11434",
       "ollamaName": "qwq:32b"
     }],
     "reasoning": {
       "type": "tokens",
       "beginToken": "<think>",
       "endToken": "</think>"
     }
   }]

2. Launch the chat-ui and mongo containers using docker-compose up.

3. Open the web UI at http://localhost:3000 and send a reasoning-style prompt (e.g., a logical puzzle or multistep deduction).

4. Observe that the model returns a response successfully, then the chat-ui container logs an error and exits.

Screenshots

Context

Logs

(base) renmin@renmin-ThinkStation-P920:~/dockercomposes/chat-ui$ docker logs -f chat-ui 
{"level":30,"time":1743073937640,"pid":22,"hostname":"c88c59dc2d52","msg":"Starting server..."}
Listening on http://0.0.0.0:3000
{"level":30,"time":1743073938199,"pid":22,"hostname":"c88c59dc2d52","msg":"[MIGRATIONS] Begin check..."}
{"level":30,"time":1743073938217,"pid":22,"hostname":"c88c59dc2d52","msg":"[MIGRATIONS] \"Update search assistants\" already applied. Skipping..."}
{"level":30,"time":1743073938217,"pid":22,"hostname":"c88c59dc2d52","msg":"[MIGRATIONS] \"Update deprecated models in assistants with the default model\" should not be applied for this run. Skipping..."}
{"level":30,"time":1743073938217,"pid":22,"hostname":"c88c59dc2d52","msg":"[MIGRATIONS] \"Add empty 'tools' record in settings\" already applied. Skipping..."}
{"level":30,"time":1743073938217,"pid":22,"hostname":"c88c59dc2d52","msg":"[MIGRATIONS] \"Convert message updates to the new schema\" already applied. Skipping..."}
{"level":30,"time":1743073938217,"pid":22,"hostname":"c88c59dc2d52","msg":"[MIGRATIONS] \"Convert message files to the new schema\" already applied. Skipping..."}
{"level":30,"time":1743073938217,"pid":22,"hostname":"c88c59dc2d52","msg":"[MIGRATIONS] \"Trim message updates to reduce stored size\" already applied. Skipping..."}
{"level":30,"time":1743073938217,"pid":22,"hostname":"c88c59dc2d52","msg":"[MIGRATIONS] \"Reset tools to empty\" already applied. Skipping..."}
{"level":30,"time":1743073938217,"pid":22,"hostname":"c88c59dc2d52","msg":"[MIGRATIONS] \"Update featured to review\" already applied. Skipping..."}
{"level":30,"time":1743073938217,"pid":22,"hostname":"c88c59dc2d52","msg":"[MIGRATIONS] \"Delete conversations with no user or assistant messages or valid sessions\" should not be applied for this run. Skipping..."}
{"level":30,"time":1743073938217,"pid":22,"hostname":"c88c59dc2d52","msg":"[MIGRATIONS] All migrations applied. Releasing lock"}
{"level":50,"time":1743073960409,"pid":22,"hostname":"c88c59dc2d52","err":{"type":"TypeError","message":"fetch failed: connect ECONNREFUSED 0.0.0.0:443","stack":"TypeError: fetch failed\n    at node:internal/deps/undici/undici:13502:13\n    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)\n    at async Promise.all (index 0)\n    at async Promise.all (index 0)\n    at async file:///app/build/server/chunks/index3-nzCYc910.js:152:63\ncaused by: Error: connect ECONNREFUSED 0.0.0.0:443\n    at TCPConnectWrap.afterConnect [as oncomplete] (node:net:1611:16)"},"msg":"Failed to initialize PlaywrightBlocker from prebuilt lists"}
file:///app/build/server/chunks/index3-nzCYc910.js:1053
  throw new Error("Generation failed");
        ^

Error: Generation failed
    at generateFromDefaultEndpoint (file:///app/build/server/chunks/index3-nzCYc910.js:1053:9)
    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
    at async getReturnFromGenerator (file:///app/build/server/chunks/index3-nzCYc910.js:1058:14)
    at async generateSummaryOfReasoning (file:///app/build/server/chunks/_server.ts-BrpIp4z7.js:275:15)

Specs

• OS:
• Browser:
• chat-ui commit:

Config

.env.local

MONGODB_URL=mongodb://mongo-chatui:27017
HF_TOKEN=hf_VXxxxxxxxxx
PUBLIC_ORIGIN=http://chat.aicxxx.xxx

ALLOW_INSECURE_COOKIES=true


MODELS=[{"name":"Qwen 2.5 7B","chatPromptTemplate":"{{#each messages}}{{#ifUser}}<|im_start|>user\n{{content}}<|im_end|>\n{{/ifUser}}{{#ifAssistant}}<|im_start|>assistant\n{{content}}<|im_end|>\n{{/ifAssistant}}{{/each}}<|im_start|>assistant\n","parameters":{"temperature":0.7,"top_p":0.9,"repetition_penalty":1.1,"top_k":50,"truncate":3072,"max_new_tokens":10240,"stop":["<|im_end|>"]},"endpoints":[{"type":"ollama","url":"http://172.30.0.1:11434","ollamaName":"qwen2.5:7b"}]},{"name":"Qwen 2.5 14B","chatPromptTemplate":"{{#each messages}}{{#ifUser}}<|im_start|>user\n{{content}}<|im_end|>\n{{/ifUser}}{{#ifAssistant}}<|im_start|>assistant\n{{content}}<|im_end|>\n{{/ifAssistant}}{{/each}}<|im_start|>assistant\n","parameters":{"temperature":0.7,"top_p":0.9,"repetition_penalty":1.1,"top_k":50,"truncate":3072,"max_new_tokens":10240,"stop":["<|im_end|>"]},"endpoints":[{"type":"ollama","url":"http://172.30.0.1:11434","ollamaName":"qwen2.5:14b"}]},{"name":"Qwen 2.5 32B","chatPromptTemplate":"{{#each messages}}{{#ifUser}}<|im_start|>user\n{{content}}<|im_end|>\n{{/ifUser}}{{#ifAssistant}}<|im_start|>assistant\n{{content}}<|im_end|>\n{{/ifAssistant}}{{/each}}<|im_start|>assistant\n","parameters":{"temperature":0.7,"top_p":0.9,"repetition_penalty":1.1,"top_k":50,"truncate":3072,"max_new_tokens":10240,"stop":["<|im_end|>"]},"endpoints":[{"type":"ollama","url":"http://172.30.0.1:11434","ollamaName":"qwen2.5:32b"}]},{"name":"QwQ 32B","chatPromptTemplate":"{{#each messages}}{{#ifUser}}<|im_start|>user\n{{content}}<|im_end|>\n{{/ifUser}}{{#ifAssistant}}<|im_start|>assistant\n{{content}}<|im_end|>\n{{/ifAssistant}}{{/each}}<|im_start|>assistant\n","parameters":{"temperature":0.7,"top_p":0.9,"repetition_penalty":1.1,"top_k":50,"truncate":3072,"max_new_tokens":10240,"stop":["<|im_end|>"]},"endpoints":[{"type":"ollama","url":"http://172.30.0.1:11434","ollamaName":"qwq:32b"}],"reasoning":{"type":"tokens","beginToken":"<think>","endToken":"</think>"}}]

# MODELS=[{"name":"Qwen 2.5 7B","chatPromptTemplate":"{{#each messages}}{{#ifUser}}<|im_start|>user\n{{content}}<|im_end|>\n{{/ifUser}}{{#ifAssistant}}<|im_start|>assistant\n{{content}}<|im_end|>\n{{/ifAssistant}}{{/each}}<|im_start|>assistant\n","parameters":{"temperature":0.7,"top_p":0.9,"repetition_penalty":1.1,"top_k":50,"truncate":3072,"max_new_tokens":10240,"stop":["<|im_end|>"]},"endpoints":[{"type":"ollama","url":"http://172.30.0.1:11434","ollamaName":"qwen2.5:7b"}]},{"name":"Qwen 2.5 14B","chatPromptTemplate":"{{#each messages}}{{#ifUser}}<|im_start|>user\n{{content}}<|im_end|>\n{{/ifUser}}{{#ifAssistant}}<|im_start|>assistant\n{{content}}<|im_end|>\n{{/ifAssistant}}{{/each}}<|im_start|>assistant\n","parameters":{"temperature":0.7,"top_p":0.9,"repetition_penalty":1.1,"top_k":50,"truncate":3072,"max_new_tokens":10240,"stop":["<|im_end|>"]},"endpoints":[{"type":"ollama","url":"http://172.30.0.1:11434","ollamaName":"qwen2.5:14b"}]},{"name":"Qwen 2.5 32B","chatPromptTemplate":"{{#each messages}}{{#ifUser}}<|im_start|>user\n{{content}}<|im_end|>\n{{/ifUser}}{{#ifAssistant}}<|im_start|>assistant\n{{content}}<|im_end|>\n{{/ifAssistant}}{{/each}}<|im_start|>assistant\n","parameters":{"temperature":0.7,"top_p":0.9,"repetition_penalty":1.1,"top_k":50,"truncate":3072,"max_new_tokens":10240,"stop":["<|im_end|>"]},"endpoints":[{"type":"ollama","url":"http://172.30.0.1:11434","ollamaName":"qwen2.5:32b"}]}]

docker-compose.yml (relevant parts)

services:
  mongo:
    image: mongo:latest
    container_name: mongo-chatui
    ports:
      - "27017:27017"
    volumes:
      - ./mongo-data:/data/db
    networks:
      chat-net:
        ipv4_address: 172.30.0.2

  chat-ui:
    image: ghcr.io/huggingface/chat-ui
    container_name: chat-ui
    ports:
      - "3000:3000"
    env_file:
      - .env.local
    environment:
      HF_ENDPOINT: https://hf-mirror.com
    depends_on:
      - mongo
    networks:
      chat-net:
        ipv4_address: 172.30.0.3

networks:
  chat-net:
    driver: bridge
    ipam:
      config:
        - subnet: 172.30.0.0/16

Notes

Notes

• The issue only occurs when using models with "reasoning" enabled (<think>...</think>).
• The model (QwQ 32B via Ollama) returns a valid response before the crash happens.
• The chat-ui container attempts to connect to 0.0.0.0:443, which fails and results in a crash.
• It seems related to post-processing logic, possibly PlaywrightBlocker or summary generation that depends on external resources or a hardcoded endpoint.
• The same environment works fine when reasoning is disabled or when using simpler models (e.g., without summary/think step).
• The behavior is consistent and reproducible across restarts and clean deployments.