Inspired by Simon Willison, I packaged whisper.cpp on PyPI, with pre-built binaries for macOS and Linux. So you can install and run Whisper with pip or uv -- nothing else required.

Get Whisper running in ~200ms with uv.

Very large context windows may extend the capabilities of LLMs because you can give them hundreds of examples on how to solve a problem (many shot learning).

This paper from Google finds significant performance boosts from many shot, even when the AI generates its own examples.

I'll be speaking at POSETTE: An Event for Postgres 🌟 🎥 #PosetteConf, a virtual conference of PostgreSQL talks happening from June 11-13.
Save the date!

citusdata.com/posette/2024

It is worth watching this CNN video from the moment Emory Econ Professor Caroline Fohlin came across the violent arrest of a protester on campus and asked the police, with shock, 'What are you doing?' That's all that prompted an officer to hurl her to the ground and handcuff her.

On today's stream, I talked about fitting conversation history in a Chat Completion API call by token counting:
youtube.com/watch?v=ix8n_g…

Inspired by Andrew Ng’s threads on agents in the past few weeks, I’m excited to share my talk from the Weights & Biases conference that outlines how to build a general context-augmented research assistant 🧑‍🔬

Naive RAG is mostly good for simple questions in a single-shot setting.…

who has an open source project they want to show off on GitHub's Open Source Friday livestream? 👀

Call for proposals for FlaskCon at PyCon US open!

flaskcon.com/2024/

Going live in a few for VS Code Day! 💃🏽

Liiky: youtube.com/live/iCDfAC4f2…

VS Code Day is happening now! youtube.com/watch?v=iCDfAC…
I'll be presenting at 12:45 PM showing a RAG app running off local PostgreSQL and local Olllama, with a FastAPI backend.

Comparing results for a simple RAG (no query rewriting/routing phrase) on phi-3 3.8B, gpt-35-turbo, and gpt-4, with same search results.
Notice the first two models recommend the harnesses, even though user asked for shoes.

There are no small language models with support for function calling, right? Especially OpenAI-API-compatible function calling?
(I've just starred all the relevant Ollama issues about it.)

FlaskCon is happening at Pycon this year! I won't be attending IRL but would still love to hear about all the talks that do happen. If you work with Flask, Quart, or related libraries, submit your talk now:
flaskcon.com/2024/

#PyCon #PyCon US #Flask

It's 6 years old, but this talk is still a fantastic overview of OAuth2 and OIDC: youtube.com/watch?v=996Oie…

Added a Llamaindex example to my Python OpenAI demos that shows how to make chat completions with either Azure OpenAI (keyless), OpenAI.com, or Ollama OpenAI endpoint.

github.com/pamelafox/pyth…

Live streams today:
10 AM PT: I'll show passwordless auth for Azure PostgreSQL
youtube.com/live/Arq3HEBs3…

10:30 AM: I'll show keyless auth for Azure OpenAI
youtube.com/watch?v=P0ZSd8…

11:00 AM: I'll watch the live stream on Langchain from Marlene Mhangami
youtube.com/watch?v=x_sgqs…