Search: “inference”

LOCAL AI MODELS ARE CATCHING UP TO FRONTIER MODELS WAY FASTER THAN ANYONE EXPECTED this guy ran qwen 3.6 27B locally on a base macbook pro M4 with 24GB of memory quantized and stripped of safety guard

Today @poolsideai is releasing Laguna M.1 & Laguna XS.2, our latest generation models and first public models We started Poolside because we believed that to build truly capable coding agents, you nee

Doing the right thing might take time, but it’s worth the wait when tens of thousands of users’ production workloads depend on it. And yes, DeepSeek V4 is coming to @FireworksAI_HQ soon 👀🚀 The DeepSee

@FireworksAI_HQ deeply believes in delivering the frontier quality. We will spend all the effort giving our users and the broad community the best OSS model quality. Deepseek V4 API up anytime now aft

Introducing our new work: “Learning to Orchestrate Agents in Natural Language with the Conductor” accepted at #ICLR2026 https://arxiv.org/abs/2512.04388 What if we trained an AI not to solve problems

Keras Kinetic has a new alpha release: v0.0.2! Including a new docs website: http://kinetic.readthedocs.io Kinetic is my favorite new release from the Keras team: a super simple Modal-like API to run

Llamacpp is a tool for running large language models locally on personal computers, offering free access without relying on cloud services. The technology emphasizes speed, security, and computational

This post from Ollama highlights the collaborative nature of open-source software development, suggesting that the value of OSS projects lies not just in the code itself but in the community and relat

@badlogicgames Yup: https://x.com/antirez/status/2048344770234249588 The GGUF tool calling template is wrong but I'm uploading a new GGUF file. Otherwise there is the right template file in one of the

Swyx shared a video of himself singing 'Sign of the Times' by Harry Styles while wearing a Base Ten shirt, which he found entertaining enough to watch repeatedly. The post expresses enthusiasm about t

Simon Willison inquires about running DeepSeek-V4-Flash on Mac hardware, specifically asking about feasibility across different RAM configurations from 512GB down to smaller amounts. This reflects dis

Launching pyptx — a Python DSL for writing NVIDIA PTX kernels. One PTX instruction = one Python call. Write pure PTX in Python. Direct Hopper + Blackwell support: wgmma, TMA, tcgen05, mbarriers. JAX +

A new Dflash drafting model based on Qwen 3.6 with 27 billion parameters has been released on Hugging Face, available at z-lab/Qwen3.6-27B-DFlash. This model likely implements speculative decoding or

Snipd hosted a user meetup in San Francisco where attendees shared how they use the podcast clipping app, including one user who creatively remixes podcast segments into music tracks. The event highli

Fireworks AI posted a question to their audience asking what features or capabilities they would like to see developed next, with a specific mention of full parameter tuning as a potential option. Thi

4⃣4⃣4⃣4⃣ Introducing DeepSeek V4 Pro, a long-context model with hybrid attention, three reasoning modes, and SOTA coding performance. AI natives can now use DeepSeek V4 Pro on Together AI and benefit

Introducing DeepSeek V4 Pro, a long-context model with hybrid attention, three reasoning modes, and SOTA coding performance. AI natives can now use DeepSeek V4 Pro on Together AI and benefit from reli

@jsconfeu @cramforce talk now live! @vercel is doing incredible AI engineering work and it comes direct from the C-suite, this should be more common but it is not, even today https://x.com/cramforce/s

Clem Delangue posted about someone flying to work with the llama.cpp team on developing next-generation local AI capabilities, framed as meta commentary on headlines related to the endeavor. The post

Our teams have been busyyy! Here are some key updates from the past week: — @GoogleCloud unveiled a suite of AI innovations at our Cloud Next event, including our eighth generation TPUs (TPUt for infe

This has been the most anticipated event in OSS and it doesn't disappoint. Congratulations to @deepseek_ai team. DSV4 Pro is now available on @togethercompute and we'll be adding a lot of capacity beh

1. We believe in iterative deployment; although GPT-5.5 is already a smart model, we expect rapid improvements. Iterative deployment is a big part of our safety strategy; we believe the world will be

Together AI achieved a 10,000x increase in monthly token throughput over one year, scaling from 30 billion to 300 trillion tokens per month. This milestone demonstrates significant growth in their API

ai researchers rocket scientists 🤝 managing burn https://x.com/shyamalanadkat/status/2047419602267988344?s=46 researchers will start giving an ai a vague hunch, compute budget, and access to tools; it

This post demonstrates a DFlash implementation integrated with llama-cpp, showcasing a speculative decoding setup that uses Qwen 3.6-27B as the main model with a smaller draft model (dflash-draft-q4_k

we burned through 5 billion tokens in 48h. turns out giving everyone unlimited access to the most expensive model on the planet is not a great business strategy 🙈 So now we have 2 free opus sessions/d

zero shot Kimi K2.6, go try it out its a good model sir! this is @Kimi_Moonshot running on @togethercompute, @opencode harness prompt below👇 Media Introducing Kimi K2.6 from @Kimi_Moonshot, a multimod

This post appears to be a humorous take on AI optimization and technology trends, likely referencing 'pixel-maxxing' (maximizing pixel-level performance or visual quality) in contrast to 'catfishing'

Highlights: 👉 80.2% SWE-Bench Verified and 89.6% LiveCodeBench v6 👉 Agent Swarm executes up to 4,000 coordinated steps 👉 Native text, image, and video input with 79.4% MMMU-Pro 👉 Production-ready on t

How far are we from agents that can self-generate world knowledge? The work proposes an outcome-based reward that measures how much an agent's self-generated world knowledge actually improves its task

Introducing Kimi K2.6 from @Kimi_Moonshot, a multimodal agentic model with Agent Swarm scaling to 300 sub-agents and long-horizon coding stability. AI natives can now use Kimi K2.6 on Together AI and

🦤 LeWorldModel: Learning Physics from Pixels — Stable World Models with Just Two Losses World models: 1️⃣ DINO-WM: pretrained ViT encoder (from ImageNet) → features → predictor. But encoder is frozen,

I tested @huggingface ml-intern, given the prompt 'Fine-tune a Segment Anything Model (SAM) on a useful medical dataset. Train the model, and provide a comprehensive tutorial in a Jupyter Notebook fil

Kimi K2.6 demonstrates strong long-horizon coding in complex engineering tasks: Kimi K2.6 successfully downloaded and deployed the Qwen3.5-0.8B model locally on a Mac. By implementing and optimizing m

This post references @dexhorthy quoting the Z/L continuum concept at an AIE Miami event, suggesting the idea is gaining traction in AI discussions. The Z/L continuum appears to be a framework or model

Kimi has been the most popular model on Fireworks, both out of the box and as a fine-tuning base (including Composer 2) Now Kimi K2.6 is live with huge jumps (10+%) in coding, long-running agents and

Swyx praised a team's livestream event for its excellent production quality and engaging content, noting that it attracted hundreds of live viewers on YouTube. The post reflects positive sentiment abo

shut the f up AIE beat TED???? a somber technical talk about security advisories and maintainer burnout beat the happy storytelling lobster on blazer one on the channel with 27 million subscribers???

Nous Research announced integration or collaboration between Hermes (their AI agent framework) and Ollama (a local LLM runtime platform), enabling users to run Hermes agents with locally-hosted langua

NEW paper from Apple. Interesting idea: 'Attention to Mamba'. The paper introduces a two-stage recipe for cross-architecture distillation from Transformers into Mamba. Naive distillation collapses tea

This post likely discusses how to run the Hermes language model using Ollama, an open-source tool for running large language models locally. It probably provides instructions or insights on setting up

Ollama is a local model-serving engine that enables users to run large language models on their own hardware without relying on cloud services. The post appears to highlight Ollama's integration with

LLM agents loop, drift, and get stuck on hard reasoning tasks up to 30% of the time. Current fixes are either too blunt (hard step limits) or too expensive (LLM-as-judge adding 10-15% overhead per ste

This post highlights a significant communication challenge that arises from differing attitudes toward linguistic precision, where those who prioritize exact word choice and terminology struggle to ef

// Skill Learning for Autonomous Web Agents // Web agents can navigate a page, but ask them to repeat a checkout flow they already completed, and they start from scratch every time. This work introduc

wow and a Colossus profile from Camille!! https://x.com/camillericketts/status/2045171537268052425?s=46 They say you can’t go home again. Pretty true, but not entirely. It was fascinating to return to