ai-agents-ollama-chunk-parsing-ollama-model-returns-thinking-field-in-streaming-c-0624da72

THEN Implement a stream modifier callback in the LiteLLM proxy router that detects and transforms chunks containing 'thinking' field, ensuring they are parsed correctly. Alternatively, create a custom Ollama Modelfile to adjust the model's output template to omit the 'thinking' field.

async def my_stream_modifier(chunk: str):
    try:
        data = json.loads(chunk)
        if 'thinking' in data and data['thinking']:
            data['response'] = data.get('thinking', '') + data.get('response', '')
            del data['thinking']
        return json.dumps(data)
    except:
        return chunk