@rashmib
Mom, Data Science/Machine Learning/Deep Learning/NLP, Teaching Fellow @ Harvard, Kaggle Competition Master https://kaggle.com/rashmibanthia ✨ Try customized Kaggle Feed ➡️ https://bsky.app/profile/did:plc:mbjdssrrxzllb2g2rq7px4pq/feed/aaadft2egbdoe
curl -fsSL https://deb.nodesource.com/setup_20.x | sudo -E bash - && \ sudo apt-get install -y nodejs && \ sudo npm install -g @anthropic-ai/claude-code && \ export PATH=/usr/bin:$PATH
Claude Code on Google Colab (commands in alt text). Claude works directly with Google Drive
This looks new and seems well put together - check it out -
www.anthropic.com/learn
Claude Code integration with GitHub
Wow Claude 4.0 😍 , we are in a different world !
Check out the demo on Project Astra
🚀🚀 so cool 😎
deepmind.google/technologies...
Wow ! What tools did they use ?
+1 O3 speaks authoritatively (occasionally wrong too)
I don't normally write advice columns but this question I got deserved its own post: With all the hype around Generative AI, how can you, as a developer, communicate the nuances and advocate for new and modular solutions that are better, easier and cheaper?
explosion.ai/blog/modular...
Does anyone know why are the meta-llama/Llama-3.1-8B-Instruct requests on @hf.co are getting denied ? I see lot of posts on the community discussion tab as well ?
Watch this video on AI Agents - youtu.be/tx5OapbK-8A?... This is actually without the hype ! (Hear what he says about Devin 😆 at 17:30
ChatGPT Desktop App , projects feature always uses 4o model and there's no way to change it for the project 🤷♀️
cuDF-pandas (%load_ext cudf.pandas) with Rapids ... work similarly and super cool to see we will be able to speed up scikit-learn
Yes and less work than creating a notebook 😅
I recently shared a Claude Artifact on Kaggle www.kaggle.com/competitions...
I really like Claude Artifacts - it uses React, so viz is better and so easy to create (oh and available in free version too) support.anthropic.com/en/articles/...
I was using aistudio.google.com with Gemini 2.0 Flash and default settings on the right.
I am not able to upload .ipynb file - I keep getting
"The current model doesn't support files of this type."
Which model works for you and should I change any settings ?
I wanted the model to explain code in a .ipynb file (over 21,000 tokens, according to Gemini).
It refused the .ipynb file, but I changed the extension to .txt, and Gemini 2.0 Flash explained it and answered all my questions well.
Impressed with Gemini 2.0 Flash, I gave the same prompt to 4o, O3-Mini, Deepseek R1, and Claude 3.7 - all either denied or hallucinated. Gemini 2.0 Flash provided the answers.
GPT4.5 preview on lmarena.ai
GPT-4.5 preview on lmarena.ai
Congratulations 🎉🎉🎉
Claude 3.7 Sonnet - Did "Ah, I see the issue" become "I see the issue" ? 🤣
Congratulations 🎉
I tried Marino notebooks yesterday first time. It looks really nice - dependency graphs, variables list super cool.
How do people share these notebooks, without having another person to install Marino ?
Model size 0.18 MB - Tensorflow GPU memory usage 22GB 🤬🤬🤬
Thank you 🙏 🙏
+1 for the “from scratch “ , I don’t understand it either otherwise !
I haven't tried Huggingface GRPO trainer, They have so many different trainers huggingface.co/docs/trl/grp... Anyways, Thank you for this !
I tried it , however I had even less memory (24GB) than L4 so I couldn't do higher batch size.
I think this intuition by Claude seems good - share.specstory.com/stories/e32f...
Model gets loaded also twice, so its kinda slow.
Thanks, I'll change the clip ratio.. I plan to try on my local 4090 - was yours on L4 ?
