Lattice Labs
Infrastructure for training and serving open models efficiently.
Repositories · 39
SciML/ReservoirComputing.jl
Reservoir computing utilities for scientific machine learning (SciML)
sylvaticus/BetaML.jl
Beta Machine Learning Toolkit
lone-cloud/gerbil
A desktop app for running Large Language Models locally.
google/grain
Library for reading and processing ML training data.
ml-tooling/best-of-jupyter
🏆 A ranked list of awesome Jupyter Notebook, Hub and Lab projects (extensions, kernels, tools). Updated weekly.
meshery-extensions/tcslabs-academy
Meshery Academy for TCS Labs
OpenSTEF/openstef
Automated Machine Learning pipelines. Builds the Open Short Term Energy Forecasting package.
marc-shade/Ollama-Workbench
A comprehensive platform for managing, testing, and leveraging Ollama AI models with advanced features for customization, workflow automation, and collaborative development.
News · 9
DeepSeek open sources DSpark, a new framework to speed up LLM inference by up to 85%
<p>Even as the geopolitical conversation around AI continues to grow more fraught following the<a href="https://venturebeat.com/technology/anthropic-blocks-all-public-access-to-claude-fable-5-mythos-5-following-us-government-order-what-enterprises-should-do"> U.S. government's actions to limit the new models from Anthropic</a> and <a href="https://venturebeat.com/technology/openai-unveils-gpt-5-6-sol-terra-and-luna-models-but-only-accessible-to-limited-preview-partners-for-now-per-us-gov">O
Predicting model behavior before release by simulating deployment
OpenAI introduces Deployment Simulation, a method to predict AI model behavior before deployment using real conversation data to improve safety and evaluation accuracy.
My reasons to run local models
<!-- SC_OFF --><div class="md"><ul> <li>I can finetune any model on any dataset I want.</li> <li>I can use techniques like speculative decoding and other sota approaches to get the max tps</li> <li>The llm provides like anthropic and openai are not getting access to my data</li> <li>The hardware is reusable for vision text speech, and I can run any blend of models for free as much as I want</li> <li>I can curate any dataset/content that I want without worrying about the costs</li> <li>I like wat
Major lab releases open weights for a frontier model
The release includes the base and instruct checkpoints under a permissive license.
Hypothetically speaking...
<!-- SC_OFF --><div class="md"><p>Would it not be possible to create crowd sourced, truly open sourced distilled LLMs with a simple wrapper around command line based AI services that exist today?</p> <p>I'm imagining a layer that goes around whatever application people currently use for coding/AI boyfriend that collects your inputs and associates them with your outputs. With enough volunteers doing this, you could create huge data sets. I understand training these models requires massive compute
IEEE Rolls Out Large Language Models Virtual Training Course
<img src="https://spectrum.ieee.org/media-library/a-middle-aged-black-man-taking-a-virtual-coding-class-in-his-home-office.jpg?id=66951841&width=1245&height=700&coordinates=0%2C156%2C0%2C157" /><br /><br /><p><a href="https://spectrum.ieee.org/recursive-self-improvement" target="_self">Large language models</a> have moved out of the research lab and into engineers’ daily workflow. LLMs serve as reasoning engines that can orchestrate complex tasks including identifying vulnerabilities
Is it agentic enough? Benchmarking open models on your own tooling
Instead of decentralized training effort we should build the “One dataset”
<!-- SC_OFF --><div class="md"><p>There are many threads here calling for united LLM training run of a new open model. Mainly, after govt. stunt of banning commercial frontier models. And also due to the lack of small-medium open-weight models releases lately.</p> <p>I genuinelly believe at some point we’ll have “SETI for LLM”. But not anytime soon, not this year. It requires a serious primary research of a training algorhytms over high latency network(s).</p> <p>What I believe be much more valu