@g-hano thanks for your PR. I've used your code to use DeepSeek-R1 with Ollama. It's outputting the Reasoning, but, not the Content. Does it work fine on your end? If I run it with o3-mini it outputs the Content correctly.

BTW, I had to make some changes in the pyproject.toml to make it work with my system-wide Ollama v 0.5.7

Ok, I realized that the output of ollama/deepseek-r1* models doesn't exactly follow the structure at _extract_think_content function, by making the below edit, I was able to make it work.

def _extract_think_content(self, text: str) -> tuple[str, str, bool]:
    """
    Extract reasoning from the <think> block and use everything after the </think> tag as content.
    """
    # Extract reasoning from the <think>...</think> block.
    reasoning_match = re.search(r'<think>(.*?)</think>', text, flags=re.DOTALL)
    reasoning = reasoning_match.group(1).strip() if reasoning_match else ""
    
    # Find the position of the closing </think> tag.
    end_think_tag = "</think>"
    end_index = text.find(end_think_tag)
    
    # If found, content is everything after the </think> tag; otherwise, use the whole text.
    if end_index != -1:
        content = text[end_index + len(end_think_tag):].strip()
    else:
        content = text.strip()
    
    # You can decide what to do with is_irrelevant.
    # Here we simply set it to False, meaning the content is relevant.
    is_irrelevant = False

    return content, reasoning, is_irrelevant

Ok, I realized that the output of ollama/deepseek-r1* models doesn't exactly follow the structure at _extract_think_content function, by making the below edit, I was able to make it work.

def _extract_think_content(self, text: str) -> tuple[str, str, bool]:
    """
    Extract reasoning from the <think> block and use everything after the </think> tag as content.
    """
    # Extract reasoning from the <think>...</think> block.
    reasoning_match = re.search(r'<think>(.*?)</think>', text, flags=re.DOTALL)
    reasoning = reasoning_match.group(1).strip() if reasoning_match else ""
    
    # Find the position of the closing </think> tag.
    end_think_tag = "</think>"
    end_index = text.find(end_think_tag)
    
    # If found, content is everything after the </think> tag; otherwise, use the whole text.
    if end_index != -1:
        content = text[end_index + len(end_think_tag):].strip()
    else:
        content = text.strip()
    
    # You can decide what to do with is_irrelevant.
    # Here we simply set it to False, meaning the content is relevant.
    is_irrelevant = False

    return content, reasoning, is_irrelevant

But with your code the method always returns false for is_irrelevant. I had no troubles with my code actually so I copy pasted the client.py Im running to the repo. Should work now.

@g-hano thank you for your contribution

This is released in v0.0.4