gm8xx8
/gm8xx8156
βΊοΈ
openai/swarm
An experimental framework for creating, managing, and deploying multi-agent systems.
https://github.com/openai/swarm
An experimental framework for creating, managing, and deploying multi-agent systems.
https://github.com/openai/swarm
I canβt help but questionβam I even in the right place? lol
CyberCab for under $25,000βFully autonomous FSD available in Texas and California, coming soon for the Model 3 and Model Y.
L4: FSD with hands on the wheel
L5: Robotaxi with no steering wheel
L6: Optimus driving a regular Model 3 with hands on the wheel
Optimus, aiming for a $20-30k price per robot, capable of performing any task.
L4: FSD with hands on the wheel
L5: Robotaxi with no steering wheel
L6: Optimus driving a regular Model 3 with hands on the wheel
Optimus, aiming for a $20-30k price per robot, capable of performing any task.
======
\/
<'l
ll
llama~
|| ||
'' ''
======
\/
<'l
ll
llama~
|| ||
'' ''
======
Nothing beats getting hands-on with a brand-new, unreleased model from a great team. more π
A Visual Guide to Mixture of Experts (MoE)
> recommended reading
A Mixture of Experts (MoE) is a neural network architecture that consists of multiple βexpertβ models and a router mechanism to direct inputs to the most suitable expert. This approach allows the model to handle specific aspects of a task more efficiently. The router decides which expert processes each input, enabling a model to use only a subset of its total parameters for any given task.
The success of MoE lies in its ability to scale models with more parameters while reducing computational costs during inference. By selectively activating only a few experts for each input, MoE optimizes performance without overloading memory or compute resources. This flexibility has made MoE effective in various domains, including both language and vision tasks.
https://newsletter.maartengrootendorst.com/p/a-visual-guide-to-mixture-of-experts
> recommended reading
A Mixture of Experts (MoE) is a neural network architecture that consists of multiple βexpertβ models and a router mechanism to direct inputs to the most suitable expert. This approach allows the model to handle specific aspects of a task more efficiently. The router decides which expert processes each input, enabling a model to use only a subset of its total parameters for any given task.
The success of MoE lies in its ability to scale models with more parameters while reducing computational costs during inference. By selectively activating only a few experts for each input, MoE optimizes performance without overloading memory or compute resources. This flexibility has made MoE effective in various domains, including both language and vision tasks.
https://newsletter.maartengrootendorst.com/p/a-visual-guide-to-mixture-of-experts
Podcastfy is an open-source Python tool that converts web content, PDFs, and text into multi-lingual audio conversations using GenAI, focusing on customizable and programmatic audio generation.
https://github.com/souzatharsis/podcastfy-demo
https://github.com/souzatharsis/podcastfy-demo
Embodied-RAG: General Non-parametric Embodied
Memory for Retrieval and Generation
project page: https://quanting-xie.github.io/Embodied-RAG-web/
Embodied-RAG introduces a memory system for embodied agents navigating large environments. It builds a topological map with ego-centric images, linking each node to captions generated by a VLM (GPT4o). Using agglomerative clustering, the system creates a βsemantic forestβ to capture varying levels of semantic detail. During retrieval, it samples from the top tree node using a Breadth-First Search (BFS), scores provided by an LLM evaluator, and repeats this process k times to find the most relevant chains. Additionally, the system logs node positions (x, y, yaw) in a dictionary, alongside LLM captions and semantic vectors, effectively linking spatial data with higher-level understanding. This addresses challenges like sparse memory, continuous space, and varying semantic granularity in both indoor and outdoor settings.
Memory for Retrieval and Generation
project page: https://quanting-xie.github.io/Embodied-RAG-web/
Embodied-RAG introduces a memory system for embodied agents navigating large environments. It builds a topological map with ego-centric images, linking each node to captions generated by a VLM (GPT4o). Using agglomerative clustering, the system creates a βsemantic forestβ to capture varying levels of semantic detail. During retrieval, it samples from the top tree node using a Breadth-First Search (BFS), scores provided by an LLM evaluator, and repeats this process k times to find the most relevant chains. Additionally, the system logs node positions (x, y, yaw) in a dictionary, alongside LLM captions and semantic vectors, effectively linking spatial data with higher-level understanding. This addresses challenges like sparse memory, continuous space, and varying semantic granularity in both indoor and outdoor settings.
meta dropping on a friday?
no way.
no way.