Knowledge catalogue
We'll get into @MiniMax_AI M3's model performance, the MSA architecture and what it means for long context, and how Together is optimizing inference and KV-cache for this new architecture. Set your re
b9442 is a release of llama.cpp, an open-source LLM inference tool written in C/C++. The release likely contains updates, bug fixes, or feature improvements to the llama.cpp project. For specific deta
b9431 is a release commit of llama.cpp, an open-source C/C++ implementation for running large language model inference efficiently on consumer hardware. Based on the source material, this release like
B9436 is an intermediate build release of llama.cpp, the open-source C/C++ implementation for running large language model inference on consumer hardware. This build typically includes updates to mode
A user reports successfully running Qwen 3.6 35b MoE (mixture of experts) with Zoo Code on an M1 Max Mac, achieving local inference without external servers. The setup enables fully local, battery-pow
arXiv:2605.29351v1 Announce Type: new Abstract: We study minimal attention-only transformers under all-token corruption and show they admit a two-stage empirical Bayes interpretation. A single attenti
b9394 is a release of llama.cpp, a project for LLM inference in C/C++ . The release represents a specific build or version update from the ggml-org/llama.cpp repository. Without access to the specific
B9402 is a release of llama.cpp, an open-source library that performs inference on large language models such as Llama and was developed in pure C/C++ with no dependencies. The release includes comman
b9410 is a release of llama.cpp, a C/C++ project that enables LLM inference with minimal setup and state-of-the-art performance on various hardware platforms. This build identifier represents a specif
b9411 is a release version of llama.cpp, a C/C++ implementation for LLM inference . This specific release build number represents updates to the open-source project hosted on GitHub, which provides op
arXiv:2602.06036v2 Announce Type: replace Abstract: Autoregressive large language models (LLMs) deliver strong performance but require inherently sequential decoding, leading to high inference latency
arXiv:2605.29626v1 Announce Type: cross Abstract: Steering language model generation toward desired textual properties is essential for practical deployment, and inference-time methods are particularl
arXiv:2605.29707v1 Announce Type: new Abstract: Speculative decoding accelerates LLM inference by drafting multiple tokens and verifying them in parallel with the target model. However, its practical
arXiv:2511.08949v2 Announce Type: replace Abstract: High-quality datasets are critical for training and evaluating reliable NLP models. In tasks like natural language inference (NLI), human label vari
Together AI developed an optimized speech-to-text system focused on achieving the fastest processing speeds through technical innovations in their inference stack and model optimization. The approach
arXiv:2605.30319v1 Announce Type: cross Abstract: A central goal of modern causal inference is estimating heterogeneous treatment effects to answer questions like 'how does an intervention affect each
love it! @ggerganov 's llama.cpp delivering! pibot is now running fully local, using parakeet for STT, qwen3-tts for TTS, and Qwen 3.6 as the local multi-modal LLM via llama.cpp. The STT and TTS infer
arXiv:2605.29766v1 Announce Type: new Abstract: Imitation learning has become a cornerstone for solving complex robotic manipulation tasks. In particular, multimodality, which enables robots to captur
arXiv:2604.14889v2 Announce Type: replace Abstract: While chain-of-thought (CoT) reasoning enables LLMs to solve challenging reasoning tasks, the linear growth of the KV cache leads to substantial mem
arXiv:2605.30126v1 Announce Type: cross Abstract: Large Vision-Language Models (LVLMs) map visual inputs into dense token sequences, imposing a quadratic computational bottleneck for inference. Elasti
pibot is now running fully local, using parakeet for STT, qwen3-tts for TTS, and Qwen 3.6 as the local multi-modal LLM via llama.cpp. The STT and TTS inference engines are Rust/mlx-c based. Ported fro
arXiv:2602.06791v2 Announce Type: replace Abstract: Being probabilistic models, during inference large language models (LLMs) display rare events: behaviour that is far from typical but highly signifi
arXiv:2605.29319v1 Announce Type: new Abstract: Large Reasoning Models (LRMs) achieve strong performance on table reasoning tasks but incur substantial inference cost due to long reasoning traces. Ste
arXiv:2605.30116v1 Announce Type: new Abstract: Distribution Matching Distillation (DMD) is a widely used paradigm for accelerating inference in few-step video diffusion models. However, DMD-style vid
Took me a while to figure out what all the ESMFold2 rage was about. At first, the benchmarking data didn't look super remarkable to me but it turns there are many impressive aspects: - Fully open sour
arXiv:2605.29141v1 Announce Type: cross Abstract: Traditional recommender systems (RecSys) primarily infer user preferences from implicit signals (such as clicks, watches, and purchases), often neglec
arXiv:2605.28575v1 Announce Type: new Abstract: Multimodal Sentiment Analysis (MSA) fuses text, acoustic, and visual streams to infer sentiment. Because pre-trained text encoders are far more expressi
arXiv:2605.27531v1 Announce Type: cross Abstract: Specification synthesis, the task of automatically inferring formal specifications from program implementations and natural language, is important for
b9375 is a release of llama.cpp, a project for LLM inference in C/C++. The release likely contains bug fixes, performance improvements, or feature additions to the llama.cpp codebase. Specific details
arXiv:2605.28292v1 Announce Type: new Abstract: Implicit Chain-of-Thought (CoT) reduces the inference cost of large language models by internalizing the explicit rationales. However, existing approach
Google created MapReduce more than 20 years ago to solve the scaling problems in data processing that the then young company was running into. The AI era that we are in now demands efficient, large-sc
arXiv:2605.27390v1 Announce Type: cross Abstract: Speculative decoding accelerates Large Language Model inference via a draft-then-verify paradigm, yet the output projection layer becomes a bottleneck
arXiv:2605.28751v1 Announce Type: cross Abstract: Linear interpolation between fine-tuned checkpoints has been shown to trace the Pareto front between competing objectives, but whether extrapolative w
arXiv:2508.15130v2 Announce Type: replace Abstract: Despite significant progress in no-reference image quality assessment (NR-IQA), dataset biases and reliance on subjective labels continue to hinder
arXiv:2605.28007v1 Announce Type: cross Abstract: Deep networks are powerful function approximators, but they typically store many different computations in shared weight matrices, making it difficult
arXiv:2509.22553v2 Announce Type: replace-cross Abstract: Causal representation learning (CRL) has garnered increasing interest from the causal inference and artificial intelligence communities due to
arXiv:2605.27787v1 Announce Type: cross Abstract: Multi-agent systems (MAS) have substantially advanced autonomous software engineering (SWE), but their growing inference energy demands raise sustaina
arXiv:2605.28646v1 Announce Type: cross Abstract: GUI agents rely on screenshots to infer intent and operate across applications, but these screenshots often contain private messages, medical records,
arXiv:2605.27616v1 Announce Type: cross Abstract: Real-time anomaly segmentation demands both high recall and efficient low-precision inference. We study the three-way interaction of model architectur
arXiv:2605.27762v1 Announce Type: new Abstract: We present PEAM, a Parametric Embodied Agent Memory framework in Minecraft that transforms agent memory from inference-time retrieval into parameter-res
arXiv:2605.27885v1 Announce Type: new Abstract: Various approaches have been proposed to adapt Vision-Language Models (VLMs) to specialized domains for Video Question Answering, including fine-tuning
arXiv:2508.21495v3 Announce Type: replace Abstract: Early-exit neural networks (EENNs) accelerate inference by allowing intermediate classifiers to stop computation once predictions are confident enou
arXiv:2605.27718v1 Announce Type: cross Abstract: Moment-based estimation is a theoretically attractive approach to parametric inference, especially when likelihood-based estimation is unavailable, mi
arXiv:2602.02259v2 Announce Type: replace-cross Abstract: Latent action models (LAMs) offer a promising path to pre-training embodied agents on large amounts of action-free video. They infer latent ac
arXiv:2605.27409v1 Announce Type: cross Abstract: SNNs promise energy-efficient and low-latency inference, but their performance still trails that of ANNs. ANN-to-SNN knowledge distillation helps narr
arXiv:2605.28764v1 Announce Type: new Abstract: Vast quantities of compute (GPU cycles on personal workstations, idle inference servers, and edge devices between jobs) go unused because no incentive-a
arXiv:2605.28713v1 Announce Type: new Abstract: Context compression aims to shorten long context inputs with minimal information loss for LLM inference acceleration. While existing methods have shown
arXiv:2602.07574v2 Announce Type: replace-cross Abstract: Modern multimodal large language models (MLLMs) adopt a unified self-attention design that processes visual and textual tokens at every Transf
arXiv:2605.27932v1 Announce Type: cross Abstract: Think-with-image reasoning is emerging as a new inference paradigm for large vision-language models, but its safety implications remain poorly underst
b9368 is an intermediate build release of llama.cpp, a C/C++ implementation for efficient LLM inference on consumer hardware. The release follows the project's rapid development cycle where multiple b
arXiv:2605.26483v1 Announce Type: new Abstract: Medical video diagnosis involves inferring clinical decisions from dynamic tissue responses throughout examination processes. Existing methods rely on a
arXiv:2605.26496v1 Announce Type: cross Abstract: The Mixture of Experts MoE architecture is highly promising for resource constrained on device deployments yet training these models from scratch incu
arXiv:2605.27292v1 Announce Type: new Abstract: Privacy auditing aims to empirically assess privacy leakage in machine learning models using membership inference attacks (MIAs), and to derive lower bo
arXiv:2603.13373v3 Announce Type: replace-cross Abstract: In ubiquitous and mobile health systems, computational models infer human states from wearable, behavioral, and physiological sensing data. In
Fast, faster, Qwen. 🚀 Thrilled to see Qwen3.5 reaching a record-breaking 580 tps for agentic workloads on the TokenSpeed engine! This milestone wouldn't be possible without our incredible partners. Hu
arXiv:2603.08413v2 Announce Type: replace-cross Abstract: Deep neural networks for image classification often exhibit overconfidence on out-of-distribution (OOD) samples. To address this, we introduce
arXiv:2605.26835v1 Announce Type: new Abstract: LLM-based multi-agent systems have been widely adopted for knowledge retrieval and report generation, synthesizing known information through web search
arXiv:2510.01336v2 Announce Type: replace-cross Abstract: Speculative decoding accelerates LLM inference by using a smaller draft model to speculate tokens that a larger target model verifies. Verific
Anthropic are strongly rumored to be about to have their first profitable quarter. Stories are circulating of companies surprised at how expensive their LLM bills are becoming from usage by their staf
arXiv:2601.03079v4 Announce Type: replace Abstract: Moral sensitivity is the most fundamental capability underlying human moral competence. Although many approaches aim to align large language models