pitchoune 0.52.8

Load & save data & chat with LLMs

Pitchoune

A Python library on top of Polars to load & save (normal or streaming mode) data from files & chat with LLM.

Features

• Load / save dataframe :

  • Handle files types : xlsx, csv, tsv, jsonl
  • Possibility to register your own file types (see custom_io_example)

• Read / write stream :

  • Handle files types : jsonl (soon csv/tsv)
  • Call the decorated function for each row of the streamed file
  • Inject each column as a parameter of the function
  • Provide as parameters the current_row and total_row
  • Can handle recovery (in case of error or process stopping) with the parameter "recover_from". Just set it to the output file path and it will use the right row for the read_stream function.
  • Multiple outputs (see multiple_outputs_example.py)

• Current workdir for filepaths : set PITCHOUNE_WORKDIR env variable

• token replacement in filepaths

• conf: token replacement in filepaths (read value from key in .conf file)

• handle .conf file in PITCHOUNE_WORKDIR

• Save to a file with human check option enabled :

  • Open the file with the default editor
  • Then wait for the user to review it & save it

• Use chat :

  • Ollama chat (see chat_example)
  • OpenAI chat (you must then set OPENAI_API_KEY env variable)

Todo

• Testing