ollama - 💡(How to fix) Fix occasional crashes when trying ocr with qwen3.5:9b [4 comments, 2 participants]

Error 500: Error with Ollama API for testimage.jpg using model qwen3.5:9b: model runner has unexpectedly stopped, this may be due to resource limitations or an internal error, check ollama server logs for details (status code: 500) Mär 10 08:36:48 kai-BeyondMax-Series ollama[3744]: HW Exception by GPU node-1 (Agent handle: 0x753cf86eff60) reason :GPU Hang Mär 10 08:36:49 kai-BeyondMax-Series ollama[3744]: time=2026-03-10T08:36:49.152+01:00 level=ERROR source=server.go:1610 msg="post predict" error="Post "http://127.0.0.1:44471/completion\": EOF"

Fix Plan

The issue seems to be related to GPU resource limitations or internal errors. To fix this, we can try the following steps:

• Increase the GPU memory allocation for the Ollama API
• Implement retry logic for the API calls to handle temporary GPU hang issues
• Optimize the model configuration to reduce GPU usage

Example Code

Here's an example of how you can implement retry logic in Python:

import time
import requests

def retry_api_call(url, data, retries=3, delay=1):
    for i in range(retries):
        try:
            response = requests.post(url, json=data)
            response.raise_for_status()
            return response.json()
        except requests.exceptions.RequestException as e:
            if i < retries - 1:
                time.sleep(delay)
            else:
                raise e

# Example usage:
url = "http://127.0.0.1:44471/completion"
data = {"model": "qwen3.5:9b", "image": "testimage.jpg"}
response = retry_api_call(url, data)

Configuration Changes

You can also try increasing the GPU memory allocation by setting the CUDA_VISIBLE_DEVICES environment variable or by configuring the Ollama API to use a specific GPU device.

Verification

To verify that the fix worked, you can test the API call with the retry logic and check the server logs for any error messages. You can also monitor the GPU usage and memory allocation to ensure that the issue is resolved.