target: august mid-september 2023

wanted:

• Jupyterlab-4
• PEP-514 compatibility
• Python-3.11.5,
• bokeh-3.2
• numba-0.58 final, numpy-1.25 (in pursuite of data-apis standard )

focus:

• Jupyterlab-4
• simplify to standard: PEP-514

exploratory:

• LargeLangageModel teaching per Andrew Ng:
  • 1 hour : ChatGPT Prompt Engineering for Developers
  • 1 hour: LangChain for LLM Application Development
• LargeLangageModel on various WinPython problems: sans-Qt interface, build, clean-up, semi-automated DataScience ...
• a local LLM that could work on a gpu-less PC:
  • https://simonwillison.net/2023/May/5/mpt-7b/
  • https://github.com/ggerganov/llama.cpp
  • https://github.com/rustformers/llm
  • ... llm-0.7 from Simon Willison

probably postponed:

• practice & build LLMs toolbox from factual learning examples on WinPython
• multi-interpreters , nogil ... seems 3.13
• simplify winpython build
• jedi-0.19 big speed-up is with spyder-5.5... next cycle
• micromamba study

helping to push the Python envelop:

• ubiquity: Jupyterlab-4, VSCode-Jupyter, Jupyterlite, Python-in-Excel
• speed in general: Cython-3, mypyc, latest Python with multi-interpreters, nogil-3.13
• speed in datascience:
  • the move to "array API standard" and to "pyarrow" latest numpy compatible with numba, latest Pandas
  • rust packages accelerators : duckdb, polars, pydantic, ...
• ... ml & llm on local