b11245
common : use fs::path for cache dirs (#29595) Avoid useless string conversions on Windows. No need for BSD or emscripten special cases. Signed-off-by: Adrien Gallouët angt@huggingface.co Website: http
Knowledge catalogue
common : use fs::path for cache dirs (#29595) Avoid useless string conversions on Windows. No need for BSD or emscripten special cases. Signed-off-by: Adrien Gallouët angt@huggingface.co Website: http
chat : fix Muse Glimmer ignoring response_format json_schema with --jinja (#29615) chat : fix Muse Glimmer ignoring response_format json_schema with --jinja Fixes #29613 chat : accept json fences and
vulkan : reuse descriptor sets when bindings are constant (#29280) vulkan : reuse descriptor sets when bindings are constant vulkan : bump buffer_destroy_count before destroying the buffer Website: ht
llama : fix init in several tools/examples (#29632) Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/51066139 macOS/iOS: macOS Apple Silicon (arm64) macOS Ap
ggml : collect all input tensors into graph_inputs (#29634) graph_inputs was populated while splitting the graph, so it only contained the inputs that are used as srcs of some node. With pipeline para
common : add fs_write_atomic() (#29642) Check for buffered write errors when closing downloaded files. Use UTF-8 paths when writing ETag files on Windows. Write in binary mode on Windows. Signed-off-b
tests : adjust server string regex to also match m2 utlra results (#29648) Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/51122498 macOS/iOS: macOS Apple S
common : use fs::path for config dir (#29649) Signed-off-by: Adrien Gallouët angt@huggingface.co Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/51176378 ma
server : remove the built-in UI's service worker when the UI is not served (#29565) With --path or --no-ui, /sw.js returned 404, and a 404 does not remove a service worker, so browsers kept showing th
common : stop accepting draft tokens at EOG (#29638) common : stop accepting draft tokens at EOG cont : remove the test Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/a
hexagon: add FP32 GELU_ERF and GEGLU_ERF support (#29631) hexagon: add FP32 GELU_ERF and GEGLU_ERF support hex-erf: reduce register pressure in kernels Website: https://llama.app Attestations: https:/
ggml : require input tensors to be GGML_OP_NONE (#29647) Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/51266318 macOS/iOS: macOS Apple Silicon (arm64) mac
ggml : accumulate f16 dot products in f32 on AVX512-FP16 (#29545) Supersedes #29530 Signed-off-by: Adrien Gallouët angt@huggingface.co Website: https://llama.app Attestations: https://github.com/ggml-
vocab : keep NORMAL in PLaMo-2 and PLaMo-3 (#29580) vocab : keep NORMAL in PLaMo-2 and PLaMo-3 The PLaMo-2 and PLaMo-3 vocabularies mark as NORMAL. Current EOG token heuristic matched it by text and a
ggml : fix c++ odr by properly using GGML_COMMON_DECL_CPP (#29504) fix c++ odr by properly using GGML_COMMON_DECL_CPP using actual field rather than macro Co-authored-by: Georgi Gerganov ggerganov@gma
vulkan: MOE aware mat_mul_id tile selection (#29182) mut_mul_id selected its matmul tile with total token count. For MoE dispatch grid the true N per workgroup is per-expert rows. At pp128 on Sarvam 3
vulkan : Load F32 A matrix 2 at a time when its 2-aligned (#29254) It turns out Intel doesn't particularly like loading F32s one at a time and we already have the _2aliagned load logic in mul_mat_vec,
arXiv:2609.35044v1 Announce Type: new Abstract: Preference-based alignment methods such as Direct Preference Optimization (DPO) use pairwise preferences labeled by human annotators to fine-tune langua
arXiv:2609.33210v1 Announce Type: new Abstract: Repetition is closely associated with memorization in generative models, but how other training images affect the retention and copying of targets remai
arXiv:2605.05386v2 Announce Type: replace-cross Abstract: Large language models increasingly operate in interactive settings where solving a task requires multiple rounds of information exchange with
arXiv:2609.32269v1 Announce Type: cross Abstract: Agents that answer questions from compressed or retrieved memory must recognize when the evidence a query needs is no longer in memory. Benchmarks for
arXiv:2609.35615v1 Announce Type: new Abstract: Behavioral Foundation Models (BFMs) are an emerging paradigm in reinforcement learning, playing a role analogous to large language models in natural lan
arXiv:2609.32743v1 Announce Type: new Abstract: Electroencephalography (EEG) foundation models (FMs) promise transferable neural representations, yet their advantages over strong supervised baselines
arXiv:2609.33254v1 Announce Type: new Abstract: Understanding how drugs interact with protein targets is fundamental to drug discovery, drug repurposing and the early identification of promising thera
arXiv:2609.34143v1 Announce Type: new Abstract: Existing 3D industrial anomaly detection mainly targets local geometric deviations. In contrast, many industrial anomalies violate object-level design o
arXiv:2609.35060v1 Announce Type: new Abstract: Inferring dynamics from snapshots of evolving distributions is fundamentally underdetermined: the Fokker-Planck equation constrains the drift F only thr
arXiv:2609.31753v1 Announce Type: cross Abstract: Video anomaly detection for patrol robots and surveillance systems must recognize abnormal interactions among familiar entities. Existing pixel-recons
arXiv:2609.35232v1 Announce Type: new Abstract: Visual-token compression is effective for improving the efficiency of vision-language models, but under extreme compression budgets, token pruning can b
arXiv:2609.32961v1 Announce Type: new Abstract: As LLM agents tackle longer tasks, they increasingly compress growing histories of reasoning, actions, and tool outputs. Compression can reduce token us
arXiv:2512.22500v4 Announce Type: cross Abstract: Modern nuclear data evaluation increasingly requires not only accurate scattering calculations, but also efficient methods for uncertainty quantificat
arXiv:2609.33800v1 Announce Type: new Abstract: Multimodal learning beyond two modalities commonly leverages a specific modality (e.g., text) to bind other modalities. However, how to establish a more
arXiv:2609.32467v1 Announce Type: new Abstract: Predicting transcriptional responses to unseen compounds is limited by fragmented chemical coverage and heterogeneous experimental platforms and gene pa
arXiv:2609.33303v1 Announce Type: new Abstract: Despite recent progress in irregular time series forecasting, the field still lacks a unified benchmark for fair and comprehensive evaluation. Existing
arXiv:2609.34622v1 Announce Type: new Abstract: Poisson denoising of scientific data requires methods that account for signal-dependent noise while accommodating different data dimensionalities and pr
arXiv:2607.10427v2 Announce Type: replace Abstract: Local motion blur detection requires pixel-level localization of blurred regions. Existing benchmarks let models rely on gradient shortcuts that fai
arXiv:2609.33722v1 Announce Type: new Abstract: Language models are increasingly used as proposal models for black-box search, from program optimization to molecular design. Existing approaches typica
arXiv:2506.07325v2 Announce Type: replace Abstract: Model Predictive Path Integral (MPPI) control provides a sampling-based framework for optimal control, while Control Barrier Functions (CBFs) provid
BREAKING: Grok is helping power America .gov, the U.S. government’s new AI website. 🇺🇸 U.S. Chief Design Officer Joe Gebbia says Grok & Gemini power the site. Americans can ask questions and get answe
arXiv:2609.34272v1 Announce Type: new Abstract: BF16 is now standard in large-scale pretraining, including in fused attention kernels such as FlashAttention, and these kernels are widely trusted. When
arXiv:2609.32570v1 Announce Type: new Abstract: Clustering algorithms that operate on nearest-neighbor graphs, such as FINCH (First Integer Neighbor Clustering Hierarchy), depend heavily on the qualit
arXiv:2609.32462v1 Announce Type: new Abstract: Motion-language models are typically scored by motion-language evaluators, but how well these evaluators ground language structure remains unclear. Here
arXiv:2609.33114v1 Announce Type: cross Abstract: For decades, statistical inference has largely been developed one problem at a time. Given a scientific target, such as a treatment effect or a regres
arXiv:2609.34387v1 Announce Type: new Abstract: Existing adaptive reasoning methods for driving Vision-Language-Action (VLA) models primarily focus on whether to reason, overlooking how reasoning shou
arXiv:2609.35195v1 Announce Type: new Abstract: Accurate segmentation of post-treatment brain metastases is essential for treatment planning, longitudinal disease monitoring, and quantitative assessme
arXiv:2609.34475v1 Announce Type: new Abstract: LLMs cannot forget the way we delete a file. Strangely, we are asked to remove something that was never put anywhere in particular. What the model took
arXiv:2609.32157v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models for autonomous driving produce natural-language reasoning alongside predicted trajectories, but whether this reasoni
arXiv:2609.33147v1 Announce Type: new Abstract: Federated low-rank adaptation (LoRA) enables collaborative fine-tuning of large language models without centralizing private client data. Its factorized
arXiv:2609.32814v1 Announce Type: new Abstract: Fast matrix multiplication algorithms keep the product fixed and search for a cheaper way to evaluate it. We instead ask whether a Transformer's learned
On September 29, 2026, Twitter user Severin Field (@sevdeawesome) posted that an AI‑generated video from Claude 5.5 Opus responding to the phrase “AI is a normal technology” was the best AI‑created vi
In September 2026, Anthropic unveiled a suite of new AI products—Claude Tag, Sonnet 5.5, Fable/Mythos 5.1—and infrastructure releases such as Opus 5.5 and a plugins portal, following a record 47 billi
arXiv:2609.35462v1 Announce Type: new Abstract: Patients often have several co-occurring clinical conditions, and the findings needed to identify and disambiguate them emerge over the course of a cons
arXiv:2609.33628v1 Announce Type: new Abstract: Prompt injection is a leading security risk for LLMs and LLM-based applications such as agents. State-of-the-art red-teaming methods for prompt injectio
arXiv:2609.33807v1 Announce Type: new Abstract: How well can general-purpose multimodal models turn visual understanding and reasoning into embodied manipulation via executable code? We introduce Code
Codex Security Cloud is getting a major upgrade, with access to cyber-capable models through Daybreak Blue included by default. It scans entire GitHub repos, continuously reviews new commits, investig
arXiv:2609.34749v1 Announce Type: new Abstract: Real-world driving is inherently multi-agent, yet most existing driving world models generate observations from a single ego vehicle. Independently exte
arXiv:2609.34782v1 Announce Type: cross Abstract: Many physical tasks in human environments require collaboration, from assisting a partner to jointly manipulating an object. Yet, existing humanoid be
arXiv:2609.35315v1 Announce Type: new Abstract: Large Language Model (LLM)-based agents promise to automate scientific discovery, yet exploring the vast hypothesis space remains costly. Existing princ
arXiv:2609.34899v1 Announce Type: cross Abstract: Multi-vector retrievers built on vision-language models lead visual document retrieval (VDR), but they run a multi-billion-parameter query encoder on
arXiv:2609.32935v1 Announce Type: new Abstract: Multi-human multi-robot (MH-MR) teams combine robotic autonomy with human expertise, but effective task allocation requires coordinating scarce, dynamic
arXiv:2609.34348v1 Announce Type: new Abstract: Gradient updates on different data generally do not commute: training a language model on two data sources in opposite orders gives different weights, e