Nobody is talking about this right now but Google dropped a CRAZY model interpretation graph tool, to enable you to better understand your models better.
Check it out, link 👇
540x Faster than GPT-4
100x Longer Sequences than GPT-4
And, better performance on Long Sequence tasks than GPT-4.
Multi-Modal Mamba is going to change the LLM game on a scale you couldn't possibly imagine in the flash of a lightning.
[Get Access Now]
This prompt is OVERPOWERED.
With this Tree of Thoughts 🌴 prompt, you can split your agent into multiple personalities // experts that debate internally for a problem.
It’s great for math and science or can even be used for more nefarious purposes
Introducing Mamba Swarm 🤖 🤖 🤖 🤖
The first ever implementation of a swarm of Mambas for real-world language modeling. There are various custom fusion methods such as weighted average, the absmax, softmax, and more.
LLMs will never be the same...
Lmao
@PropheticAI
changed the visibility of their posts because I implemented their model fast.
This is a lesson.
Don’t close source your AI models or they’ll get implemented open source anyways 😂
I just launched Swarms on ProductHunt 🚀
It's a simple python package that enables you to build, deploy, and scale "Swarms" of LLMs, you can get started in 5 lines of code, check it out:
Follow it and pls leave a review if you have tried it 🙏
I'm very excited to announce the Pytorch Model Implementer GPT that helps you create bleeding edge models in pytorch 😊 🤖
This has been my personal companion helping me implement 100s of AI papers for acouple months now.
Get Started:
Jamba
A Hybrid Transformer-Mamba Language Model
present Jamba, a new base large language model based on a novel hybrid Transformer-Mamba mixture-of-experts (MoE) architecture. Specifically, Jamba interleaves blocks of Transformer and Mamba layers, enjoying the benefits of
I'm done.
I can't implement so many papers all the time, I need you.
I'm releasing a list of papers that need to be implemented to democratize SOTA AI for all Humans and advance Humanity.
Join the Agora community to learn more:
Updates on AlphaFold3 OS Implementation 🔥 100
✅ Genetic Diffusion Module
✅ PairFormer: Implementation from scratch
✅ MSA Embedding
And, join the discord to help:
I’ve been thinking about setting up hacker houses in SF, El Segundo, NYC, and Miami to connect people from different locations together.
Kinda like little monasteries you can come and stay at if you’re in the city.
Should we make this happen?
I’m looking to create dope banners to promote open source, if you’re a graphic designer and want to advance Humanity dm me.
And, if you know someone tag them
Announcing Neo Sapiens, the next stage of AI.
This project introduces the first-ever self-replicating hierarchical swarms of autonomous agents in production for real-world tasks.
Get started here:
How does AlphaFold3 work exactly?
Let's dive into the model architecture and training strategy of this new historic protein prediction model.
We're implementing AF3 open source right now.
Agora will be giving away 50-A100 GPUs with 0 time-limits to researchers working on cool stuff.
Steps to Qualify:
-> Join Agora:
-> Share your project in the "Creator Showcase" channel
-> Like, Retweet, and comment on this post.
You have 1 week.
Announcing, WhiteRock
The first ever fully-automated VC fund transforming venture capital. Built at the HackAIThon by Craft Ventures, Agent Ops, and others.
I'd like to share my first ever 3d object creation model!
It's the pytorch implementation of the new paper Gamba, an end-to-end amortized 3D reconstruction model from single-view image
Paper:
img -> 3d object using Mamba!
Announcing Open Gemini, the open source implementation of Google's fantastic Multi-Modal Model.
Multi-Modal models like this need to be democratized or we face real existential risk.
Google spent 1BILLION DOLLARS on Gemini, and I'm almost done with my implementation that I made in around 2+ hours of work for free.
Investors what are you doing this is SUPER HORRENDOUS 😢 😢
It’s official.
The Swarm Corporation office is now in Mountain View 🛸👾👾👾
This will be our office until 1B revenue.
If you want to join the team, join the discord and send PRs to the Swarm repo!
The leaderboard for the HomeRobot challenge is now open at :
- create a docker image containing your agent
- planning or learning, use your favorite strategy
- we plan to run the top 3 on real hardware
- potentially win a robot from
@hellorobotinc
This is fucking ridiculous.
Faster, higher, and stronger Multi-Modal models [PaLM2-VAdapter]
PaLM2-VAdapter outperforms SOTA with 30∼70% fewer parameters.
My open source Implementation is ready for training ✅ 😊
I'm very proud to announce HLT, Humanoid Locomotion Transformer.
This is the first ever open source implementation of a transformer for humanoid robotics.
This an implementation of the paper: "Real-World Humanoid Locomotion with Reinforcement Learning"
It's insane how a vast majority of the chinese AI research labs are open sourcing all their research with code and weights and everything while Western research labs close source everything.
What's going on here?
@Blockworks_
This is horrible. A USD-backed stable coin would have unimaginable consequences such as targeted inflation, targeted credit inflation, and other horrible horrible effects.
This is not it whatsoever.
BREAKING 🚨 ‼️
Imagine fully automated hospitals running at 100% efficiency. Check out this new paper on a hospital swarm:
PAPER: Agent Hospital: A Simulacrum of Hospital with Evolvable Medical Agents
@SamuelMullr
this is why papers who do not release code with experiments should be overlooked, as there is no way to formally test the paper's findings
Microsoft presents The Era of 1-bit LLMs
All Large Language Models are in 1.58 Bits
Recent research, such as BitNet, is paving the way for a new era of 1-bit Large Language Models (LLMs). In this work, we introduce a 1-bit LLM variant, namely BitNet b1.58, in which every single
El Segundo is insane.
Most capable people on the planet building ferociously to advance Humanity.
Build. Build. Build.
Everything else doesn’t matter.
Godspeed ⚡️
Excited to announce AI paper to code GPT
Implement AI research Papers into Pytorch, Jax, Triton, and Einops or framework of choice effortlessly 👾🚀
Built at the AGI house!
BREAKING:
Cubans have taken to the streets.
Tired of the communist regime and the policies ruining the countries economy, they are chanting “FREEDOM”
🇨🇺
Excited to Announce AgoraX/AIEC-140K 📝
An all-new dataset with super high High Quality AI Engineering Code Tokens totaling 140k samples!
AEIC is from the top research labs like OpenAI, Nvidia, Google, Lucidrains, and others scraped through github!
Very excited to share my implementation of Screen AI, a multi-modal model for UI and Infographics Understanding in pytorch📄 📄
✅ Model Architecture: Patching -> Vit -> embed + concat -> multi-modal encoder -> decoder
✅ ready to train right now.
The next era of Robotics is here 🤖 🤖 🤖
Announcing HRTX, a Hivemind Multi-Input and Multi-Output Transformer for Robotics.
This can model can process inputs from any number of N robots and output any Z actions across any of those robots.
Announcing, Creating a Language Model with PyTorch in 5 Minutes 🔥 🤖
Build a SOTA Language model with this simple tutorial using Pytorch in 5 min 🚀 Will be sharing a series on creating Transformers and deploying them into production soon stay tuned.
Agents don't suck.
Individual agents suck.
Multi-agent collaboration bypasses all the limits of individual agents like context window limits, single task threading, and hallucination.
Don't believe me? Check out these papers for more.
I'm very proud to announce Open QwenVL
An open source production grade ready to train Qwen VL exactly as described in the Qwen Paper.
✅ Full Multi-Modal Processing [Vit, Cross Attn]
✅ Modular and Re-Usable
✅ Built with Zeta
Get started now:
Your personal Devin in less than 80 lines of code with Swarms.
✅ Tools like Terminal ++ Browser
✅ Longterm memory
✅ Stopping conditions
✅ Any LLM [GPT-4 // Llama // etc
Get started now:
$ pip install swarms
I grew up impoverished in one of America’s poorest cities.
I never finished high school.
Nor did I go to get a PHD at university.
I taught myself by reading hundreds of books and failing again and again thousands of times in an endless loop over a decade.
You can do this
Excited to share Hierarchical Mamba 👾👾👾
This is the first hierarchical Mamba where a low level mamba will output a sequence that will then be picked up by a high level mamba.
HSSS outperforms SOTA Transformers by at least 23% on MSE.
I'm excited to announce AgoraX/VMMFC-3OK,
The first open multi-modal function calling dataset with 30k+ samples for a vast array of real-world tasks such as parallel tool usage, api calling, and multi-modal action execution.
Bigger verisons otw!
At 2,000 followers I’ll open source something that will fundamentally reshape Human civilization.
Like, retweet, and share with your family and friends
I Keep getting the same question
"How is Swarms compared with Langchain"
"Autogen"
"Griptape"
And, my only answer is
Production Grade Reliability
That's why swarms exists
To provide thee most reliable agents that actually do what you want them to do
Dont hope for GPT5 or LLAMA 3, make models better than them.
Hoping and waiting around like a puppy isn’t going to help you.
Learn PyTorch and lucidrains and Cuda.
Take control of your destiny.
[Starting next week]
I'm going to make a tutorial on how to make a transformer GPT-3 like model in less than 100 lines of code, other multimodal tutorials, and maybe even stream all day every day.
We need to accelerate learning.
The way we do normal lora won't work bitnet, bcs we have to maintain property when lora weight are merged they can still be quantized into good ternary, we need littel reformulation in forward pass, so
BitLora
Imagine buying 1,000 pounds of salad for 1$
This will soon become possible with Swarms of autonomous agents operating vertical farms.
Post Scarcity is close…
We just need a push
Announcing, Open Source Implementation of Pairformer 🧬 🧬
Pairformer is used in AlphaFold3 to model the interactions between ligands, atoms, and more.
Help us Implement AlphaFold3 in Agora!
I'm delighted to share that I've implemented a radically simple Jamba in pytorch, to help you get started with joint Mamba + Transformer architectures!
And, if you like building models join Agora, the open source AI research lab!
[Swarms][N2]
A Swarm is any group of 2 or more agents working together to accomplish a task.
In a swarm, communication between the agents is the biggest limiting factor between the agents achieving trying to achieve a task.
AI Research Team 🤖
A 3 agent team that works together to implement an AI research paper in code. Amazing swarm for easily extracting the algorithms from a paper and using them!
�� GPT-4 + Claude
✅ Multi-Agent Orchestration
✅ SOP Prompting
One of the most underrated papers of all time: "Assembly theory explains and quantifies selection and evolution"
Provides a simple formula called The Assembly Index to predict evolution in any kind, life or technology. It's insane. check it out
I’m excited to announce I’m headed back to SF tomorrow for a month with
@thomasschulzz
Solaris Residency Program 😆
Let’s get dinner, work on autonomous agents, and advance Humanity 💯🤖🤖🤖
Vision Mamba: Efficient Visual Representation Learning with Bidirectional State Space Model
paper page:
Recently the state space models (SSMs) with efficient hardware-aware designs, i.e., Mamba, have shown great potential for long sequence modeling.
300+ free AI Models, 6+ years of open source AI, most important AI engineer in Human history
Please sponsor or donate to him so he may continue open source AI research otherwise he won't be in open source anymore and humanity be set back 100 years.
Announcing Atom ⚛️, a suite of finetuned LLMs for atomically precise function calling 🧪
✅ Massive function calling dataset of over 20M samples.
✅ First Model: Atom-Z-Tiny - Zephr trained on 100k samples
✅ Vision function calling coming soon
Wandb
Just Implemented the new Zamba joint mamba-transformer paper.
The architecture is:
6 mamba -> attention -> mlp -> linear -> 6 mamba -> attention -> mlp -> linear -> finally 6 mamba
The model architecture is new and performance is exceptional.
Links below:
Zeta is overpowered.
I can implement any paper in 15-30 minutes with modular plug in and play blocks
-> Transformers
-> FFNs
-> SSMs/ Mamba
-> Activations
-> Embedding functions
-> Biases
Updates on AlphaFold3 OS Implementation 🔥 100
✅ Genetic Diffusion Module
✅ PairFormer: Implementation from scratch
✅ MSA Embedding
And, join the discord to help:
AoT is the most underrated prompt technique EVER.
It searches the space of all possible solutions to a problem in one step. Not multiple steps like chain of thoughts or tree of thoughts and others!
Just copy and paste with your task:
Let's discuss why individual LLM agents fall short. Despite their advanced capabilities, they face five major problems: hallucinations, limited context windows, single-threaded processing, lack of collaboration, and huge size/cost.
@yednapg
I'm Kye 👋 I created my first neural network at 11 years old and started my ASI company
@apac_ai
some time ago.
@apac_ai
s goal is create artificial super intelligence to advance the human species.
It's insane how poorly built Langchain, BabyAGI, Autogen are built.
They all produce results that are not desire-able in any capacity.
All of these agent frameworks are doomed to death.
If you can't produce outcomes people want and need you're screwed.
Interstellar is one of those movies that can change your life.
It collapses your world view into the size of a small ball and shows you temporary your burdens are.
It shows you what actually matters which is advancing Humanity and building a family
Meet Reka Core, our best and most capable multimodal language model yet. 🔮
It’s been a busy few months training this model and we are glad to finally ship it! 💪
Core has a lot of capabilities, and one of them is understanding video --- let’s see what Core thinks of the 3 body
Billion Dollar App Ideas 💡
- Open source peer to peer Uber.
- Open source Calendly
- Open source Loom
- Open source peer to peer Food Truck discovery app
- Open source payment system like Cashapp
1/15
The history books will remember this day.
We're thrilled to announce Agora's partnership with
@AWS
and
@Nvidia
.
This alliance will supercharge our mission to advance humanity through open-source, multi-modality AI research.
#Agora
#AWS
#Nvidia
#AI
#OpenSource
❌ Never went to college
❌ Never finished high school
❌ Didn’t come from money [ my family came to America on a boat from Cuba ]
❌ Built many failed businesses
❌ lost 90% of my networth
Why do we fall?
To learn how to pick ourselves up.
Keep pushing forward
Ai researchers and academics don't know how to code.
It's very inefficient, poorly optimized, and just low quality.
Which is why I made Zeta, to make AI engineering as simple as possible:
Every time I try to use some simple install instructions for research code from github, I get frustrated. Even with conda etc, it seems like AI code is much less reproducible than you would think.