Search: “model-releases”

Gaussian Universality and Its Breakdown in Tensor-Network Machine Learning

arXiv:2610.06080v1 Announce Type: cross Abstract: Gaussian-process limits are powerful in describing overparameterized machine learning models, yet their validity in structured tensor-network architec

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Gefen: Optimized Stochastic Optimizer

DGX agent

arXiv:2606.13894v3 Announce Type: replace-cross Abstract: AdamW is a default optimizer for deep learning, but its moment states add two parameter-sized buffers to training memory, increasing the cost

model-releasesarxiv-cs-cv

6 Oct 2026

Gemma 4 26B-A4B and a 37 GB Qwen3.6 MoE running in a browser tab on a 24 GB Mac — experts streamed from disk, output matches llama.cpp

This is an update. I posted LocalMind here many moons ago from another account, when it was a Gemma chat in a tab. LocalMind is a static web page that runs models on your GPU through WebGPU. It has no

Geometry-Dependent Approximation for Non-Monotone k-Submodular Maximization

arXiv:2610.04049v1 Announce Type: cross Abstract: We study nonnegative, non-monotone k-submodular maximization with kge2 labels under support constraints, and show how the certified approximation coef

GitSwarm: Decentralized Compounding Inference

arXiv:2610.04862v1 Announce Type: cross Abstract: Long-horizon problem solving and scientific research require computation to accumulate across successive attempts. Partial solutions, experimental fin

GNN-CB: A Graph Neural Network Competition Benchmark for Human and LLM Evaluation

arXiv:2610.05387v1 Announce Type: new Abstract: Large language models (LLMs) have demonstrated strong performance on coding and reasoning benchmarks; however, their ability to solve graph-structured m

Google DeepMind launches EmbeddingGemma 2, a 740M-parameter model to map code, images, video, and audio in a shared embedding space, under an Apache 2.0 license (Google)

Google: Google DeepMind launches EmbeddingGemma 2, a 740M-parameter model to map code, images, video, and audio in a shared embedding space, under an Apache 2.0 license — EmbeddingGemma 2 is the most

model-releasestechmeme

Google expands EmbeddingGemma beyond text to images, audio and video

model-releasessiliconangle

Google LLC today released EmbeddingGemma 2, an open multimodal embedding model small enough to run on a smartphone. The release takes the EmbeddingGemma line beyond text, which was all the first versi

Google is about to remove free access to Gemini Flash and Pro

model-releasesthe-verge-ai

Starting on October 9th, anyone using Google Gemini on a free plan will be limited to the Flash Lite model. Free users can currently choose from Gemini Flash Lite, Flash, and Pro, but now you'll need

google/embeddinggemma-2 · Hugging Face

EmbeddingGemma 2 is an open multimodal embedding model built by Google DeepMind which maps text (incl. code), images, video, and audio inputs—and combinations thereof—into a single, unified 768-dimens

GPT-6.1 Sol looped 'leak' hints at nested models serving architecture

Hello llamas. I am posting this because I believe that, despite it being closed source models, the discussion will bring value to the local AI community. As many of you probably heard, GPT-6 Astra is

GPU-Accelerated Bregman Douglas-Rachford Splitting for Discrete Optimal Transport

arXiv:2610.04715v1 Announce Type: cross Abstract: We present GPU-accelerated Bregman Douglas--Rachford splitting algorithm (BDRS) for discrete optimal transport problem in three input formats: an expl

GR-LIO: A Local Ground-Aware LiDAR-Inertial Odometry System Using Body-to-Ground Height

model-releasesarxiv-cs-ro

arXiv:2610.05546v1 Announce Type: new Abstract: LiDAR-inertial odometry (LIO) is widely used for state estimation in ground-based autonomous mobile robots. However, the geometric constraints provided

Graph-Based Floor Separation Using Node Embeddings and Clustering of WiFi Trajectories

arXiv:2505.08088v4 Announce Type: replace-cross Abstract: Vertical localization, particularly floor separation, remains a major challenge in indoor positioning systems operating in GPS-denied multisto

Greedy Multi-Path Block Verification for Faster Decoding in Speculative Sampling

arXiv:2602.16961v2 Announce Type: replace-cross Abstract: The goal of L-step speculative decoding is to accelerate autoregressive decoding of a target model by using a cheaper draft model to generate

Green-Routed Neural Operators: Physics Determines Where the Network Reads

arXiv:2610.05337v1 Announce Type: new Abstract: We identify a mismatch between the physical role of transport fields in many PDEs and their usual role in neural operators: PDEs use them to select read

Grounded in Time: A Multi-Source Dataset and Benchmark for Temporal Grounding in Robotic Manipulation

arXiv:2610.04255v1 Announce Type: cross Abstract: Robotic manipulation often requires inferring task-relevant states from past interactions when the current observation alone is insufficient to determ

GS-Codec: A Gaussian-Splatting Bottleneck for Neural Audio Coding

arXiv:2610.04651v1 Announce Type: cross Abstract: Neural audio codecs compress waveforms into compact discrete tokens that underpin speech language models, real-time communication, and large-scale aud

Hamiltonian Metric Learning and Energy-Based Training: A Dissipative Geometric Framework for Optimization

arXiv:2610.04969v1 Announce Type: new Abstract: Optimization in machine learning is usually expressed through iterative rules that update model parameters using information from the loss landscape. In

Harnessing Multimodal Large Language Models for Training-Free Human-Object Interaction Detection

arXiv:2610.06394v1 Announce Type: new Abstract: Human-object interaction (HOI) detection aims to localize human-object pairs and recognize their interactions. Traditional supervised methods perform st

Here's a benchmark that uses Lilypond https://x.com/aug5thmusic/status/2096030719156089029

model-releasessimon-willison--x

Here's a benchmark that uses Lilypond https://x.com/aug5thmusic/status/2096030719156089029 GPT-6 Astra has the best result yet on the Bach Benchmark. Its chorale contains no voice-leading errors, and

Homogeneous Semantic Alignment and Hierarchical Expert Routing for Radiology Report Generation

arXiv:2610.04499v1 Announce Type: new Abstract: Radiology report generation (RRG) aims to convert medical images into diagnostic texts to assist in clinical decision-making and alleviate the workload

How Execution Assumptions Change Short-Horizon Sharpe Rankings: Evidence from a Synthetic Trading Benchmark

arXiv:2610.05077v1 Announce Type: new Abstract: Backtests of LLM trading agents often assume that every order fills at the closing price. We ask whether this choice changes only reported returns or al

How RL Reshapes LLM Reasoning: Transferability, Coverage, and Scaling Laws

arXiv:2610.04158v1 Announce Type: new Abstract: Recent studies on reinforcement learning (RL) report seemingly conflicting evidence about large language model (LLM) reasoning. Training on mathematics

https://claude.ai/chat/834491a2-0f6c-44be-9cc9-53211c8b8405

A small thing I noticed: full-size screenshots use far more AI tokens than the model needs, so usage limits run out sooner. I built a free Firefox add-on, TokenSaver, that shrinks images before they a

model-releasesr-ollama

Hyperbolic Graph Representation Learning: Embed in One Metric, Optimize with Another

arXiv:2610.06745v1 Announce Type: new Abstract: Hierarchical graphs embed in hyperbolic space with lower distortion than in Euclidean space owing to its negative curvature. However, their gradient-bas

I asked Opus 5.5 to make an interactive website companion for the latest @AcquiredFM Home Depot episode. It came out pretty nice! Water colo…

model-releasesboris-cherny--x

I asked Opus 5.5 to make an interactive website companion for the latest @AcquiredFM Home Depot episode. It came out pretty nice! Water color illustrations all drawn by Claude, too: https://claude.ai/

I gave a 21M model a 6.4B-parameter lookup table. It matches a 114M dense model and runs with the table on an SSD (RX 9070)

I spent the last few weeks on a hobby research project and just made it public. The idea isn't new (product-key memory, Lample et al. 2019, and Meta's 'Memory Layers at Scale'): give a model a huge ta

I have a feeling this is a good open source AI week 🥰 - Monday: @reflection_ai Beam - Tuesday: @MistralAI Mistral Large 4 - Who is next? 👀…

model-releasesfireworks-ai--x

On October 6 2026 Sophia Yang announced an “open‑source AI week.” She highlighted Monday’s event featuring @reflection_ai’s Beam and Tuesday’s unveiling of @MistralAI’s Mistral Large 4, indicating ant

I have about $4000, what's the best setup to get?

Ideally I'd like to be able to run Qwen 3.8-Flash-Next with decent performance. I was thinking of just buying 2x Radeon AI R9700 (64GB VRAM), or maybe a DGX Spark but that was before the price hike. M

I tried using Gemma 4 (e2b) offline to identify birds. It confidently got them wrong, so I built a photo scavenger hunt instead

I wanted to build an offline bird/plant identifier that runs on a laptop with Ollama. Before building any UI, I tested it on 6 photos: Common Myna → gemma4:e2b said 'Jackdaw' (confidence: high). gemma

model-releasesr-ollama

i was told we can't fundraise and grow at the same time. it turned out we grew monthly usage 400% during fundraising. we have never run this…

model-releasesyohei-nakajima--x

i was told we can't fundraise and grow at the same time. it turned out we grew monthly usage 400% during fundraising. we have never run this fast. while taking 60 investor calls a week, we shipped 20

Ideal Paths for Approximating Logistic Gradient Descent Trajectories at Large Initialization

arXiv:2610.04142v1 Announce Type: new Abstract: Modern training on a new task often starts from a previously trained model rather than from scratch, raising the question of how this initialization aff

If you are using Qwen3.8-27B + MTP, make sure to upgrade to DFlash for extra speed: llama serve -hf ggml-org/Qwen3.8-27B-GGUF --spec-type dr…

model-releasesgeorgi-gerganov--x

If you are using Qwen3.8-27B + MTP, make sure to upgrade to DFlash for extra speed: llama serve -hf ggml-org/Qwen3.8-27B-GGUF --spec-type draft-dflash --spec-draft-n-max 7 Requires the latest llama.cp

Image resolution enhancement for advanced semiconductor nodes

arXiv:2610.05809v1 Announce Type: new Abstract: Advanced semiconductor nodes are pushing the limits of feature sizes and require metrology with sub-nm resolution without compromising on the throughput

Inferring physical fields in coupled systems with unknown parameters from incomplete observations using physics-constrained attentive neural operators

arXiv:2610.05723v1 Announce Type: new Abstract: Given incomplete measurements of a single physical field in a coupled system with unknown parameters, can we infer its full physical state and identify

InstructTA: Instruction-Tuned Targeted Attack for Large Vision-Language Models

arXiv:2312.01886v4 Announce Type: replace Abstract: Large vision-language models (LVLMs) have demonstrated their incredible capability in visual question answering. However, this rich visual interacti

InteractionBench: A Real-Time Interaction Benchmark for Streaming Video Systems

arXiv:2610.05775v1 Announce Type: new Abstract: A video assistant must speak when its instruction warrants a response and stay silent otherwise. We introduce a benchmark that evaluates this decision f

Introducing EmbeddingGemma 2: A best-in-class open model for natively multimodal embeddings | Google

https://huggingface.co/google/embeddinggemma-2 EmbeddingGemma 2 is an open multimodal embedding model built by Google DeepMind which maps text (incl. code), images, video, and audio inputs—and combina

Introducing GeoGuess Bench! A benchmark to measure how good AI models are at playing GeoGuessr. We gave each model 210 photos from around th…

model-releasestogether-ai--x

Introducing GeoGuess Bench! A benchmark to measure how good AI models are at playing GeoGuessr. We gave each model 210 photos from around the world, asked it to predict where each photo was taken, and

Introducing Mistral Large 4: Le chonk

model-releasessimon-willison

Introducing Mistral Large 4: Le chonk Mistral are back in the game. Today they're releasing a preview of Mistral Large 4, a 1 trillion parameter, 49 billion active parameter model trained on their own

KITINet: KInetic Theory Inspired Inter-Channel Information Exchange for Residual Networks

arXiv:2505.17919v2 Announce Type: replace Abstract: Residual connections are foundational to modern neural networks, yet the standard additive update provides no explicit mechanism for structured inte

Knossos and Ariadne: Benchmarking and Learning Complete Diagram Topology Extraction with Vision-Language Models

arXiv:2610.04721v1 Announce Type: new Abstract: Structural diagrams are widely used to represent complex systems and relational information across scientific, engineering, procedural, and spatial doma

Latent Score-Based Bayesian Cramer-Rao Bound Estimation for High-Dimensional Imaging Systems

arXiv:2610.03956v1 Announce Type: cross Abstract: We propose a data-driven framework for estimating the Bayesian Cramer-Rao bound (CRB) in high-dimensional imaging systems with complex, analytically i

LawVu launches Lens to provide legal teams visibility into contract law

model-releasessiliconangle

New Zealand-based LawVu Ltd., a legal artificial intelligence startup, today announced the launch of Lens, an AI contract analysis tool that gives in-house legal teams visibility into risks and obliga

Learning in Continuous Games from Pairwise Preference Feedback

arXiv:2610.05428v1 Announce Type: cross Abstract: We study learning in continuous games when players receive only pairwise preference feedback, revealing which of two actions is preferred but neither

Learning the Context of Errors: Black-Box Online Adaptation of Time Series Foundation Models

arXiv:2606.14222v2 Announce Type: replace Abstract: The rapid evolution of Time Series Foundation Models (TSFMs) has advanced zero-shot forecasting across diverse domains. Inspired by the current form

Learning to Learn a Language

arXiv:2610.05879v1 Announce Type: cross Abstract: We present the Prior-Fitted Language Model (PFLM), a 300M-parameter byte-level transformer pretrained only on samples from a synthetic non-linguistic