Kolada

all in one e-commerce web design & hosting | digital platform management

[jhey ʕ•ᴥ•ʔ @jh3yy]

CSS 👏

[Rohan Paul @rohanpaul_ai]

Open Interpreter's new release looks great for locally running LLMs

The tool lets LLMs run code (Python, Javascript, Shell, and more) locally. You can chat with Open Interpreter through a ChatGPT-like interface in your terminal by running $ interpreter after installing.

Just do `pip install open-interpreter`

and then just run

`- interpreter --local` sets up fast, local LLMs.

Congrats @hellokillian 👏

[Kolada @KoladaApp]

@pydantic 👀

[IroncladDev @IroncladDev]

Rust :(

[LangChain @LangChain]

🔥 Firefunction-v2 🔥: Llama 3 fine-tuned for tool calling / agents

Firefunction-v2 is a new open weights model from @FireworksAI_HQ fine-tuned for tool calling. Built on Llama 3-70b, Firefunction-v2 marries strong instruction following capabilities with tool calling that is on par with GPT4-o.

This is a great option for building agentic applications with an open source / weight LLM.

We show how to use this model for tool calling / building agents, and benchmark its tool calling ability ourselves vs GPT4-o on a SQL challenge.

Video:
https://t.co/ZNHgcVI4Rp
Docs:
https://t.co/PKh38AWT4R
Cookbook:
https://t.co/NSJI3J3NdY
Blog: https://t.co/B1Jhr9Pjnq

[Andrej Karpathy @karpathy]

Nice new read on tokenization!
You've heard about the SolidGoldMagikarp token, which breaks GPT-2 because it was present in the training set of the Tokenizer, but not the LLM later.

This paper digs in in a lot more depth and detail, on a lot more models, discovering a less extreme version of the above - partially-trained tokens in both open/closed models. You have to be careful with a lot of small details and implications - weight sharing, constants in residual streams, weight-decays, regex splitting patterns, BPE, UTF-8, etc.

TLDR Tokenization remains a major pain and a large LLM attack surface. Including these partially-trained tokens in your prompts drifts the model out of distribution into undefined regions of the dynamics, areas that the model is not used to. They confuse the LLM. The paper's focus is discovery and not engineering, but it seems likely one can find "token attacks" that reliably induce target weirdness: pop-off safety, alter personality or behaviors (?), any other kind of ... otherwise undefined behavior, whatever that may look like.

Now go ask GPT-4 about _ForCanBeConverted, $PostalCodesNL, useRalative, and _typingsJapgolly :)
(or see Figure 4 of the paper at the very end for simple examples)

[KHON2 News @KHONnews]

The Aurora Borealis was spotted from the Hawaiian Islands for the first time in over a century.
https://t.co/pmw6xIYkPo

[Kolada @KoladaApp]

@str_xs https://t.co/9ZLxTMhtEP

[tetsuo @tetsuoai]

Dr. Gilbert Strang and Lex Fridman discuss the theory behind singular value decomposition (SVD), which is a very important concept in machine learning.

[LangChain @LangChain]

✂️Semantic Chunking

Semantic chunking using embedding distances between sentences to figure out how to split a document. A key decision is what threshold to split at.

We now support three methods:

📎Percentile
🟠Interquartile Range
👿Standard Deviation

Big shout out to @matthaigh27 for adding two new methods (interquartile, standard deviation), and big shout out to @GregKamradt for the original inspiration and the original method (percentile)

Notebook: https://t.co/k1uUUEdrc0

[Kolada @KoladaApp]

@str_xs I’m looking forward to small b2b solutions like this

[Kolada @KoladaApp]

@justinmfarrugia Maybe. But Emoji’s in use in code are interesting af

[Vaibhav (VB) Srivastav @reach_vb]

sqlcoder-70B is here 🚀

> beats gpt-4

*let the revolution begin!*

[Rishabh Srivastava @rishdotblog]

We just opened sourced SQLCoder-70B! It outperforms all publicly accessible LLMs for Postgres text-to-SQL generation by a very wide margin.

SQLCoder is finetuned on @AIatMeta's CodeLlama-70B model that was released yesterday on less than 20,000 hand-curated prompt completion pairs.

You can find it on @huggingface at https://t.co/abHsKzWYGW. This follows our 15B, 7B, and 34B models – and is the most capable of them all.

The model has a cc-by-sa-4 license, which means that you are free to use it as is for any use (including commercial) as long as you also open-source any changes to you make to it (i.e., if you fine-tune it further).

[Maxime Labonne @maximelabonne]

Super happy to see that I have three models on @huggingface's trending page. 🎉

Thanks everyone for the support and all the kind comments!

[tldraw @tldraw]

This week's release will include arrow scooting.

[Kolada @KoladaApp]

@dvassallo No, you do need CSS, Js, Canvas, and everything else shiny…Daniel. And you can feel free to contact me anytime. 🤝

[LangChain @LangChain]

🐀Improved RAGatouille <> LangChain Integration

We really enjoyed @bclavie's new RAGatouille library. It makes advanced retrieval techniques (like ColBERT - more on that below) really easy to use

We've worked on a tighter integration to make it *super simple* to use RAGatouille and LangChain together

After you construct your RAGatouille retriever, you can just call `.as_langchain_retriever()` and it's immediately available as a LangChain retriever that you can use in any chain!

Updated out docs here: https://t.co/dDos1bXs8S

The main use case for this (to start) is to use ColBERT models.

If you want to learn more about ColBERT, the best person to follow is @lateinteraction

He has GREAT threads like the following: https://t.co/nDoBiGaos4

Read up on ColBERT via @lateinteraction, try it out via RAGatouille with @bclavie, and then use it in LangChain to bring the right context to your chains!

[Christian Keil @pronounced_kyle]

All human knowledge can be compressed into ~40 GB.

We're not that smart. Yet.

[jhey ʕ•ᴥ•ʔ @jh3yy]

Messin' with JavaScript &lt;canvas&gt; 💫

[Alex Hormozi @AlexHormozi]

Work doesn’t discriminate on the fuel you use to get it done.

Passion is great.

But anger will do.