DAAUUUMMMM! Deep Seek R1 - 4bit on a single Mac Studio 512gb.
18.26 Tokens per second with MLX.
Took over a minute to load the model but I sped that up. Generation was great!
thanks @awnihannun mlx is the future.
The best-received part of my talk at GenAI London and a common question for RAG:
How do you do RAG over tables?
1. Extraction is key
Convert tables into HTML. This works best for providing context to the LLM. @UnstructuredIO is awesome for this.
🧵👇
I zero-dayed @GroqInc's API.
I was eager to use Groq in my apps, but the API is in private beta—meaning anything beyond the website was off-limits...
Or.. I could reverse-engineer the frontend to build a @langchain runnable.
Let's talk about the exploit in the thread.
👇 🧵
RAG is "garbage in, garbage out." People build RAG apps but spend all their time on the LLM & prompt engineering, not on the data.
Think data processing is unsexy? Think again. @UnstructuredIO + @langchain are 🔥🔥🔥🦵🦵👯.
Unstructured is offering $100 to try them out.
👇🧵
I have had the privilege to use LangSmith and from @langchain over the last month or so and it's downright amazing...
With LangSmith now going GA I wanted to share a few things that I have absolutely loved about it.
And shoot me a DM if you want to know more
🧵 👇
Stop making your kids share.
My kid's daycare teacher hates sharing, and she's right. Making kids share is selfish of the parent and sets kids up to be entitled and possessive.
🧵👇
Using consensus driven workflows with multiple LLMs can dramatically increase accuracy. Using @langchain and LangGraph makes building multi agent collaboration straight forward.
We just wrapped up @langchain Interrupt, and here are my 10 key takeaways!
1️⃣ Agents are Here - I'm definitely riding a post-conference high. The energy was electric, and everyone was deeply engaged. The conference showcased real-world agent implementations happening now, such
Powers been out for about 5 hours and there's no cell At our cabin. But we do have 150 down and 20 up with a 40ms ping plugged into a cigarette lighter in the car.
Starlink mini is incredible
This is 100% true. The learning curve on o1-pro is real. You have to know how to ask or you get
Me: what's the capital of France?
O1-pro:
*traveling to europe*
*eating a croissant*
*Admiring the Mona Lisa*
*Speaking in French*
Thought for 3minutes and 40seconds
O1-pro: Paris
🚨 New Video: Tutorial 🚨
It takes less than 30 minutes to deploy your own LLM using @ollama, @Hetzner_Online and @Tailscale for a fully VPN private coding experience that works with @cursor_ai
Save money and be in control of your stack!
youtu.be/0NdvGUk_fns
