Ivan Shcheklein
@shcheklein.bsky.social
CTO and co-founder - building tools to make ML engineers happy at DataChain.ai
Reposted by Ivan Shcheklein
There's a trap I see AI/ML teams fall into with video, audio, and multimodal data.
🎥🎧👽
I wrote a blog post about it (with memes). See 🧵
🎥🎧👽
I wrote a blog post about it (with memes). See 🧵
September 4, 2025 at 6:08 PM
There's a trap I see AI/ML teams fall into with video, audio, and multimodal data.
🎥🎧👽
I wrote a blog post about it (with memes). See 🧵
🎥🎧👽
I wrote a blog post about it (with memes). See 🧵
Reposted by Ivan Shcheklein
Wrote up my initial impressions of the new Google Gemini 2.0 Flash model - it's really good, and the streaming mode (where you can stream video and audio to it and get audio streamed right back) is pure science-fiction https://simonwillison.net/2024/Dec/11/gemini-2/
Gemini 2.0 Flash: An outstanding multi-modal LLM with a sci-fi streaming mode
Huge announcment from Google this morning: Introducing Gemini 2.0: our new AI model for the agentic era. There’s a ton of stuff in there (including updates on Project Astra and …
simonwillison.net
December 11, 2024 at 8:27 PM
Wrote up my initial impressions of the new Google Gemini 2.0 Flash model - it's really good, and the streaming mode (where you can stream video and audio to it and get audio streamed right back) is pure science-fiction https://simonwillison.net/2024/Dec/11/gemini-2/
Reposted by Ivan Shcheklein
Should HF do more agent stuff? If so, what would be useful?
November 23, 2024 at 4:08 PM
Should HF do more agent stuff? If so, what would be useful?
Reposted by Ivan Shcheklein
I have made a feed of my favorite GenAI ppl from X previously but in case you want them as a starter pack there you go go.bsky.app/Gmj2bgJ
November 20, 2024 at 9:49 PM
I have made a feed of my favorite GenAI ppl from X previously but in case you want them as a starter pack there you go go.bsky.app/Gmj2bgJ
Reposted by Ivan Shcheklein
Ever noticed that the attention mechanism in transformers is essentially a two-layer MLP? 🤔
A(q, K, V) = V @ softmax(K / √d @ q)
Weights: K / √d and V
nonlinearity: softmax
💡This offers fresh insights into KV cache compression research 🧵(1/3)
A(q, K, V) = V @ softmax(K / √d @ q)
Weights: K / √d and V
nonlinearity: softmax
💡This offers fresh insights into KV cache compression research 🧵(1/3)
November 20, 2024 at 9:55 AM
Ever noticed that the attention mechanism in transformers is essentially a two-layer MLP? 🤔
A(q, K, V) = V @ softmax(K / √d @ q)
Weights: K / √d and V
nonlinearity: softmax
💡This offers fresh insights into KV cache compression research 🧵(1/3)
A(q, K, V) = V @ softmax(K / √d @ q)
Weights: K / √d and V
nonlinearity: softmax
💡This offers fresh insights into KV cache compression research 🧵(1/3)
Reposted by Ivan Shcheklein
TIL about a Jupyter notebook alternative. Some features sound very exciting:
* no hidden state, deterministic execution
* execute as a Python script, parametrized by CLI args
* git-friendly: notebooks are stored as .py files
Have you tried it? How is it? github.com/marimo-team/...
* no hidden state, deterministic execution
* execute as a Python script, parametrized by CLI args
* git-friendly: notebooks are stored as .py files
Have you tried it? How is it? github.com/marimo-team/...
GitHub - marimo-team/marimo: A reactive notebook for Python — run reproducible experiments, execute as a script, deploy as an app, and version with git.
A reactive notebook for Python — run reproducible experiments, execute as a script, deploy as an app, and version with git. - GitHub - marimo-team/marimo: A reactive notebook for Python — run repr...
github.com
November 20, 2024 at 1:32 PM
TIL about a Jupyter notebook alternative. Some features sound very exciting:
* no hidden state, deterministic execution
* execute as a Python script, parametrized by CLI args
* git-friendly: notebooks are stored as .py files
Have you tried it? How is it? github.com/marimo-team/...
* no hidden state, deterministic execution
* execute as a Python script, parametrized by CLI args
* git-friendly: notebooks are stored as .py files
Have you tried it? How is it? github.com/marimo-team/...