Rohan-Paul-AI

👀👀

Benchmarks are important, but it's not the end-all and be-all.

I see benchmarks like car reviews. I can read the review and see the stats of a car but I'm not going to buy it until I test drive it.

------

Are you into AI and LLMs❓ Join me on Twitter with 31.1K others, to remain on the bleeding-edge every day.

𝕏/🐦 x.com/rohanpaul_ai

.

1 year ago (edited) | [YT] | 5

Rohan-Paul-AI

Claude’s API now supports CORS requests for their JSON APIs, enabling client-side applications. 🧠

📌 So it’s now possible to call the Claude LLMs directly from a user’s browser.

✨ Anthropic was wary of this feature due to a potential security risk: embedding API keys in client code could lead to theft and unauthorized use. However, they've recognized valid applications, like internal tools for trusted users or a "bring your own key" approach where users provide their own credentials for client-side apps.

👨‍🔧 You can now add the following HTTP request header to enable CORS support for the Anthropic API, which means you can make calls to Anthropic’s models directly from a browser:

`anthropic-dangerous-direct-browser-access: true`

Example code to use this new feature with vanilla JavaScript

------

More info on this PR - github.com/anthropics/anthropic-sdk-typescript/pul…

-------

Are you into AI and LLMs❓ Connect with me on Twitter to remain on the bleeding-edge every day.

👉 x.com/rohanpaul_ai

1 year ago (edited) | [YT] | 12

Rohan-Paul-AI

I love Twitter. 🩷

What can be more thrilling than when your childhood hero follows you? 🔥🔥🔥

1 year ago | [YT] | 8

Rohan-Paul-AI

This feels really nice. My Twitter Account saw some reasonable growth over the last 4-5 months, to 14.5K followers as of today.

Last month I got monetized, and today Twitter just paid me for content creation.

Something I’d do regardless! 😊 Thanks to @elonmusk

Checkout - x.com/rohanpaul_ai

1 year ago | [YT] | 14

Rohan-Paul-AI

Very interesting excerpt from the Kolmogorov-Arnold Networks (KAN) Paper on "Catastrophic forgetting" in regular MLP based Neural Network vs in Kolmogorov-Arnold Networks 🤔

"Catastrophic forgetting is a serious problem in current machine learning. When a human masters a task and switches to another task, they do not forget how to perform the first task. Unfortunately, this is not the case for neural networks. When a neural network is trained on task 1 and then shifted to being trained on task 2, the network will soon forget about how to perform task 1.

A key difference between artificial neural networks and human brains is that human brains have functionally distinct modules placed locally in space. When a new task is learned, structure re-organization only occurs in local regions responsible for relevant skills, leaving other regions intact.

Most artificial neural networks, including MLPs, do not have this notion of locality, which is probably the reason for catastrophic forgetting.

We show that KANs have local plasticity and can avoid catastrophic forgetting by leveraging the locality of splines. The idea is simple: since spline bases are local, a sample will only affect a few nearby spline coefficients, leaving far-away coefficients intact (which is desirable since faraway regions may have already stored information that we want to preserve). By contrast, since MLPs usually use global activations, e.g., ReLU/Tanh/SiLU etc., any local change may propagate uncontrollably to regions far away, destroying the information being stored there."

1 year ago | [YT] | 11

Rohan-Paul-AI

5T tokens FineWeb dataset just dropped
@huggingface


It's a 275GB dataset with cleaned and deduplicated data under an Open Data Commons license.

We all see the difference the 15T tokens pre-training made for LLaMA-3 and now everyone can have it .

1 year ago | [YT] | 7

Rohan-Paul-AI

And finally, LLAMA 3 is out officially

1 year ago | [YT] | 3

Rohan-Paul-AI

#Llama3 is out on Azuremarketplace

1 year ago | [YT] | 10

Rohan-Paul-AI

Multiple processes (1 per GPU) to encode sentences in parallel with SentenceTransformer, to distribute the embedding process across multiple GPUs

Gives a near linear speed-up when encoding large text collections.

The relevant method is `start_multi_process_pool()`, which starts multiple processes that are used for encoding.

1 year ago | [YT] | 17

Rohan-Paul-AI

Reached 10K followers on Twitter 🙏

👉 twitter.com/rohanpaul_ai

Checkout if you are interested in daily technical nuggets on Artificial Intelligence & Large Language Models

1 year ago | [YT] | 12