jerpint

Using a colab with printed outputs could be a good option to at the very least hint to reproducing results independently

I recommend VScode for that use case (other editors probably support this too) where you can ssh to your cluster, access all your files in the tree, and run Jupyter notebooks from VScode directly using HPC resources, very nice workflow

I wonder if this could be combined with fastHTML to have a full python html rendered client side

There’s an added objective in some of these models to make them more aesthetically pleasing based on subjective crowdsourced data that very likely contributes to this

One thing I find particularly interesting is that SOTA video understanding requires significantly less parameters than SOTA language understanding. Are we just overfitting way too much to language?

Also how long until SAM gets aligned to an LLM? Would be great to natively prompt it and not hackily chain through separate vision models

Ive used p5js to build my own Tetris game with vim bindings :)

https://www.jerpint.io/blog/tetris/

Models can easily regurgitate back training data verbatim, so anything private can be in theory accessed by anyone without proper access to that file

Awesome, thanks for sharing

Why do I need to supply an API key to try their demo? Seems absurd

I go back regligiously to this post whenever I need a quick visual refresh on how transformers work, I can’t overstate how fantastic it is

I have my own small blog , and am more and more tempted to “poison” it with made up facts about me to see if/when it gets scraped by LLMs

Still blazing fast on iPhone

I implemented a version of game of life using 2D convolutions from scipy, I’d be curious to see how it compares to GPU crunching

Nice! I did a game of life using stable diffusion recently, check it out :)

https://www.jerpint.io/blog/diffusion-gol/

Why are llama 3 weights listed as partially available ?

Also Olmo didn’t release a paper per se but a very detailed blog post / cookbook for reproduction which should count to a perfect score imo

Not yet, some models in that direction are LTU-AS

Most “serious” companies would probably rewrite it from scratch anyway

It could be both. In my case my keys were used to call OpenAI, almost certain they were leaked from my Spaces secrets

My openAI key was leaked and I noticed someone was using it, luckily the damage wasn’t nearly as bad as you. A few dollars worth of GPT4, a model none of my apps were using at the time.

I’m almost entirely certain it was leaked via secrets on HF space, I got a message a few days ago warning me some of my spaces were affected

I noticed a few weeks ago that some of my OpenAI keys got compromised, they were only active as secrets on a huggingface space. I got an email a few days ago informing me that the spaces were compromised , so I suspect this issue has been going on for at least a few weeks

Completely agree, and it’s been great to see to some extent that the gap between open source and 3rd party models is really starting to close

One point they don’t seem to spend much time on is also the difficulty in reproducing outputs in closed-source models. Setting temperature to 0 and setting seeds doesn’t always seem to be enough to get exactly the same results for a given prompt

Microsoft has the compute, so its likely to be an infrastructure problem

Have you tried CLIP image embeddings ?

I don’t think the app version is available

The imagenet results’ “main” contribution was showing that scaling the compute on large datasets worked surprisingly well. They didn’t contribute any new models as much as they showed that scaling is what helped these models learn and obtain SOTA.

It’s no surprise that industry has been pushing for more compute ever since.

How does earth get seeded?

The videos look incredible, but a lot of the captions are riddled with grammar/syntax mistakes that seem odd for a model to make of that quality.

My understanding is that tokenization is part of the bottleneck. When you break up a sentence to tokens, each token gets a vector representation. The dictionary of all tokens would be infinite if it was at the sentence level

It seems to be a python app, so probably set it up as a seperate microservice with its own REST API