Webpage summarizer with ollama and CLI

This projects generates summaries from documents loaded from URL.

This is largely based on https://github.com/rishabkumar7/langchain-ollama (thanks! 🎩).

Changes I've made:

• update ollama invoking to fit current langchain documentation
• add passing arguments to the script from the command line
• add appending detailed output to a .jsonl (jsonlines) file
• add parsing out deepseek's chain of thought (delimited by <think></think> markers)

Advantages of access through CLI: you can write an external bash script and call the summarizing module as many times as you like.

Requirements

Install Ollama and download some open-source models (I tested this with llama3.2, deepseek-r1:14b, deepseek-r1:8b, deepseek-r1:1.5b).

Set up a virtual environment and install the dependencies from requirements.txt.

Example usage

python main.py -i 'https://en.wikipedia.org/wiki/Interlocutor_(linguistics)' -o example_output.jsonl -l deepseek-r1:1.5b

Here's a pretty-printed result of the above command:

{
  "url": "https://en.wikipedia.org/wiki/Interlocutor_(linguistics)",
  "model_used": "deepseek-r1:1.5b",
  "chain_of_thought": "<think>\nOkay, I need to summarize this text concisely. The reference materials discuss \"interlocutor\" and \"routelers,\" which are related to language use and quotation techniques. Here's how I can approach it:\n\n1. Start with the main topic.\n2. Mention key terms like interlocutor and routelers.\n3. Note their role in communication, particularly quotation.\n\nSo, a possible summary would be: The reference materials discuss concepts like \"interlocutor\" and \"routelers,\" examining their roles in language use and quoting techniques.\n</think>",
  "summary": "\n\n**Summary:** The references discuss key terms such as \"interlocutor\" and \"routelers,\" focusing on their roles in language use and quotation techniques."
}

The result of a few runs with different models is saved in example_output.jsonl in this repo.

TODOs

1.

• Add a switch to save the date when the summary was retrieved.
• Add a switch so you can turn on and off the filtering of deepseek's chain of thought

2.

The script works despite printing the following errors:

USER_AGENT environment variable not set, consider setting it to identify your requests. (always)

None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. (sometimes)

Token indices sequence length is longer than the specified maximum sequence length for this model (1480 > 1024). Running this sequence through the model will result in indexing error (with some models)

I haven't tested Rashib's original repo, but I think it would have the same issues.