30
Smart search ranks by meaning as well as keywords (one row per story, last 45 days).
Automobile Camouflage to Hide from Flock Cameras
Schneier on Security highlights a printed vehicle-camouflage pattern tested to defeat Flock surveillance cameras and Axon body cameras.
The post discusses covering cars with printed patterns designed to fool Flock automated license-plate recognition software, with testing reportedly done against Flock and Axon body cameras. Reader comments question effectiveness against other ALPR vendors, Flock's RF MAC-address upgrade, and whether such camouflage might become regulated. The page also contains off-topic comment threads about anti-bot over-blocking and privacy.
15
The Government is Monitoring Anti
Fusion center bulletins reveal US law enforcement monitoring anti-Flock social media accounts and warning police ahead of the DeFlock Week of Action against license plate readers.
Public records requests by 404 Media and journalist Dan Boguslaw exposed intelligence bulletins from fusion centers in Colorado, Wisconsin, and Florida tracking anti-Flock sentiment, camera vandalism videos, and the DeFlock National Week of Action against automated license plate readers scheduled for August 16-22. The bulletins highlight Instagram accounts like Nomark.Project posting daily camera takedowns, and a device found during a June 25 traffic stop that could locate Flock cameras. Police are advised to increase patrols around ALPR locations and warned about upcoming DeFlock events, including 11 Florida cities signed up. DeFlock creator Will Freeman said the project never called for vandalism and that over 100 surveillance contracts have been canceled through civic engagement.
28
Sakana AI Researchers Introduce PC-ALM, a Layer-Local Alternative to Backpropagation That Trains 1000-Layer Networks
Sakana AI's PC-ALM adds per-layer Lagrange multipliers to predictive coding, matching backprop on networks up to 1000 layers with layer-local updates.
Sakana AI researchers propose Augmented Lagrangian Predictive Coding (PC-ALM), a training method that keeps every update layer-local while recovering backprop-aligned credit signals. The team proves multipliers converge to exact backprop adjoints in linear networks and trains 1000-layer residual MLPs on MNIST within about 2 points of backprop accuracy. PC-ALM matched backprop across a width/depth grid from 8 to 128 on MNIST and Fashion-MNIST where standard predictive coding failed in deep, narrow networks, and improved over PC on ResNet-18 with CIFAR-10 and Tiny ImageNet. An MIT-licensed JAX reference implementation reproduces the results on CPU.
40
Cops Search Thousands of Flock Cameras for Reasons of ‘LMAO,’ ‘IDK,’ ‘Hehe,’ and ‘asdfg’
EFF analysis found police searched Flock's 19,000-camera ALPR network using joke reasons like 'LMAO' and 'idk,' exposing routine warrantless surveillance abuse.
An EFF analysis of Flock audit logs, shared with 404 Media, found officers across dozens of jurisdictions searching over 19,000 cameras in 1,558 cities with justifications like 'LMAO,' 'idk,' and keyboard gibberish between 2023 and late 2025. One Lake County, Indiana sheriff's deputy searched for a license plate across the entire network citing 'LMAO.' After outlets exposed searches targeting ICE detainees, an abortion patient, and protesters, officers were reportedly told to be 'as vague as permissible,' and Flock replaced the free-text reason box with a prepopulated drop-down list, which EFF called a loss for transparency. Departments rarely disciplined offending officers, and EFF concluded agencies cannot oversee themselves.
35
ThinkPrior: Zero-Rollout Difficulty Priors for Cold-Start Prompt Selection in RLVR
ThinkPrior builds zero-rollout difficulty priors via an offline verifier-anchored pass, halving silent groups in RLVR and cutting wasted rollouts on Qwen2.5-Math-7B.
In GRPO-based RLVR, groups where all rollouts are correct or all are wrong yield zero advantages and consume about 39% of a run's rollouts under uniform sampling. ThinkPrior initializes a Beta posterior from an external anchor pass's verifier-scored pass rate, selecting prompts by expected learnability before any target-policy rollout, without changing the loss or optimizer. On Qwen2.5-Math-7B across sixteen seeds it more than halves early silent groups and cuts wasted rollouts through step 30 by nearly a fifth, with no detected final-accuracy difference. The ThinkPrior+DAPO composition reduces generated rollouts by 10.6% at an equal 3,840-rollout update budget.
29
Where the backlash against Flock Safety is having the biggest impact
More than 90 cities ended Flock Safety contracts in August as Texas and Florida restrict license plate reader use amid privacy protests.
Over 90 cities and counties terminated Flock Safety contracts in August, per Secure Justice, which has tracked more than 200 municipal terminations since 2021. Texas Gov. Greg Abbott barred state agencies from funding Flock cameras on Aug. 28 after a report that a state agency secretly diverted $30 million; Florida's transportation department banned ALPRs on state highways and Gov. Ron DeSantis criticized the technology. Los Angeles chose not to renew its LAPD contract over data-ownership terms, and Atlanta's mayor ordered a 30-day review of the roughly 5,000 ALPR cameras in the metro area. Several Flock-related police abuse cases, including a Texas officer indicted on 100 felony counts, have fueled the backlash.
35
'I Saw a Shiny Thing': Cop Explains Why He Used License Plate Reader to Stalk Woman
Body camera footage shows Florida officer Lamar Roman used DMV databases and ALPR cameras to stalk a woman; dozens of Flock misuse cases surfaced.
404 Media published body camera footage showing the investigation into officer Lamar Roman, who met a woman on the set of Apple TV's Bad Monkey, then illegally queried DMV databases and placed her plate on an ALPR hot list. He nearly caused a head-on collision while following her and illegally pulled her over, and was later arrested in front of his home. Flock's CEO said the system has caught many abusive officers, and the Washington Post found at least 50 misuse incidents; Roman used Turing's Guardian ALPR system.
22
AdaptVPR: Route-Aware Hard Positive Generation for Robust Visual Place Recognition
AdaptVPR generates route-aware synthetic hard positives for visual place recognition, releasing the 160K-image AdaptCities dataset with R@1 gains up to 9.2% under domain shift.
AdaptVPR is a generative augmentation framework that creates same-place hard positives under illumination, weather, seasonal, and dynamic-occlusion shifts for robust visual place recognition training. A vision-language model parses scene attributes and estimates editability, while a rule-based scheduler routes generation through global appearance, local occlusion, or dual perturbation routes with geometric-consistency verification. The resulting AdaptCities dataset contains 160K verified synthetic hard positives, and experiments show R@1 gains up to 9.2% across VPR baselines and backbones. Code and data are publicly released on GitHub.
22
A Secretive DHS ‘Predictive Policing’ Unit is Analyzing Americans’ Financial Habits and Pulling Them Over
404 Media reveals DHS Border Patrol's secretive Predictive Intelligence Targeting Teams (PITT) analyzing Americans' financial activity and feeding intelligence to local police for traffic stops.
404 Media identified Predictive Intelligence Targeting Teams (PITT) in the Spokane Sector (Washington) and Laredo Sector (Texas), which review law enforcement-sensitive databases including Americans' financial activity and pass intelligence to local police. In one case, a PITT analyst flagged financial patterns associated with narcotics activity, leading Montana Highway Patrol to stop a driver for an obstructed license plate and charge him with DUI and possession with intent to distribute. CBP declined to say what financial data is monitored or whether warrants are obtained; the program extends AP's earlier reporting on ALPR-based predictive policing.
40
DPRK APTs: Ted backdoor and curlRAT target South Korean media and automotive sectors
Rapid7 uncovered a DPRK-linked Linux toolkit using a HAProxy-embedded ted backdoor, SSH keylogger, and curlRAT against South Korean media and automotive firms.
Rapid7 Labs identified a previously undocumented framework attributed with medium confidence to DPRK actors, targeting South Korean automotive and media organizations likely since early 2025. The toolkit embeds a backdoor compiled into HAProxy 2.8.12 using its filter API, plus trojanized crond, agetty, atd, sshd, and polkitd, an SSH keylogger storing credentials under /var/lib/sshd/, and a curl-based RAT with a watchdog thread. It enables remote command execution, malicious script injection into served webpages (a watering-hole loop), credential harvesting, and long-term surveillance. Hardcoded C2s are associated with APT37 via ThreatFox, and exposed groupware portals and mail servers align with Kimsuky tradecraft; the initial access vector and any CVE remain unconfirmed.
68
30
Fine-tuning a 350M Model for Better Structured Outputs in 100 GRPO Steps
Hugging Face guide fine-tunes a 350M-parameter model with 100 GRPO steps to improve structured output reliability.
A Hugging Face blog post demonstrates fine-tuning a 350M-parameter model using GRPO (Group Relative Policy Optimization) with TRL over 100 training steps. The stated goal is more reliable structured outputs from small language models. No article body was available, so details beyond the title are limited.
22
Hackers Stole Flock’s Camera Software, Revealing How the Company Tracks Cars and People
Hackers who removed a Flock Safety license plate camera dumped its data, revealing person-detection capabilities and an encryption key stored unencrypted on the device.
A hacker collective calling itself stegan0gram physically removed a Flock Safety automatic license plate reader camera from a roadway, copied its storage, and shared the files with 404 Media, WIRED, and Distributed Denial of Secrets. Analysis found an encryption key in an unencrypted 'media' partition that unlocked videos of thousands of vehicle detections, with logs showing more than a million images generated in weeks. The software explicitly detects people, bicycles, and even bumper stickers, and records from one Georgia city were searchable by more than 2,000 agencies nationwide. The findings follow 2025 research by Jon Gaines documenting flaws enabling root-level access to Flock cameras.
48
m-a-p/YuE2-3B — new model trending #30 on Hugging Face
M-A-P released YuE2-3B, an open music generation model that outperforms Suno v5 on WildSongBench and runs locally on a 24GB GPU.
The M-A-P (multimodal-art-projection) team released YuE2-3B, an open-weights music generation model that turns lyrics and a style prompt into full songs with vocals and accompaniment. It uses an AR-NAR Mixture-of-Transformers backbone with symbolic planning and flow matching through a VAE, and supports editable scores (melody and chords, including ABC notation) plus agentic editing workflows. On 192 WildSongBench prompts it reports a SongBench average of 6.9632 (best-of-8) versus 6.8721 for Suno v5, claimed as state of the art among evaluated open and proprietary models. It runs 48 kHz stereo inference locally on a single 24GB NVIDIA GPU without quantization, with companion releases including YuE2-Vae, MERT-v2 encoders, the WildSongBench dataset, and SheetSage2.
28
30
Async GRPO with LoRA across HF Jobs: a bucket, a proxy, and no NCCL
Hugging Face blog describes running async GRPO reinforcement learning with LoRA across HF Jobs using a storage bucket and proxy instead of NCCL.
A Hugging Face blog post titled 'Async GRPO with LoRA across HF Jobs: a bucket, a proxy, and no NCCL' explains an asynchronous Group Relative Policy Optimization training setup using LoRA adapters distributed across Hugging Face Jobs workers. The architecture coordinates training through an object storage bucket and a proxy server, removing the need for NCCL collective communication. No full article text was available at classification time.
25
How Threat Actors Are Turning Trusted AI Platforms Into an Attack Surface
Huntress details campaigns abusing Claude Artifacts, claude.ai/share links, and ChatGPT/Grok conversations to deliver SectopRAT, MacSync, and AMOS stealers.
Huntress SOC documented nine months of campaigns in which attackers weaponized trusted AI platform features—Claude Artifacts, public claude.ai/share links, and indexable ChatGPT/Grok conversations—to deliver malware. The July FakeAgent campaign hit more than 29 organizations via a malicious Claude Artifact posing as a Claude Desktop download page that redirected to SectopRAT. A claude.ai/share link disguised as an Apple Support guide tricked a victim into running a curl command that deployed the MacSync stealer, harvesting cookies, credentials, keychain secrets, Telegram sessions, and SSH/cloud keys, while SEO-poisoned ChatGPT and Grok conversations delivered the AMOS stealer via ClickFix-style instructions.
55
Doctor Doom Thanked Seattle for All the Surveillance Cameras
Activist dressed as Doctor Doom urged Seattle's council to curb Axon and Flock surveillance after a committee voted to redirect $250,000 from CCTV expansion.
Clifford Cawthon, a policy advisor at the Washington State Department of Commerce, appeared at a Seattle Public Safety Committee meeting in costume as Doctor Doom to protest the city's expanding mass surveillance. He criticized Seattle PD's access to 62 cameras feeding an Axon real-time crime center that ingests CCTV, body camera, geolocation, ALPR, and dispatch data from vendors like Axon and Flock. His group Community Not Cameras submitted a petition with five demands, and a committee already voted to move $250,000 away from CCTV expansion pending full council approval.
15
The OCUDU dApp Platform: An Open Runtime and E3 Interface for Real-Time AI-RAN
OCUDU open runtime lets third-party signed AI-RAN dApps run inside production 5G distributed units under three timing contracts, released as BSD-3 preview.
The OCUDU dApp platform provides an open runtime and E3 interface for executing signed AI-RAN applications inside a production 3GPP NR distributed unit, where prior dApp frameworks could only observe export streams. Modules run under three typed timing contracts: GPU receive-chain residency (Class A), the scheduler's 100 microsecond deadline (Class B), or non-blocking observer (Class C). On a GB10 gNB, dApps including an out-of-tree neural equalizer ran on a live cell without fallback. Platform, SDK, and a zero-hardware quickstart are public under BSD-3-Clause-Clear as a preview of the OCUDU AI-RAN Working Group 2.
22
Texas Police Used AI to Write Report About Using Flock to Search for Woman Who Had Abortion
Johnson County, Texas deputies used Axon's Draft One AI to write a report about searching 80,000+ Flock cameras for a woman who self-administered an abortion.
Documents show the Johnson County Sheriff's Office used Flock's nationwide camera network and Axon's Draft One, which drafts police reports from body camera audio, in its investigation of a woman's self-administered abortion. The AI-generated report summarized deputies' discussion of legal implications, noting Texas law provided no applicable criminal charges. Flock CEO Garrett Langley has repeatedly claimed the search was a family welfare check, but earlier police reports indicate it was initiated at the behest of the woman's abusive partner.
35
Measuring benchmark optimization in speech recognition
Hugging Face examines how much speech recognition systems overfit benchmarks and how to measure benchmark optimization in ASR.
A Hugging Face post on measuring benchmark optimization in automatic speech recognition, analyzing how model improvements on benchmarks reflect genuine capability gains versus overfitting. It is evaluation methodology research with no direct security impact.
22
AgentGrad: Intervention-guided Prompt Optimization for Multi Agent Systems
AgentGrad introduces intervention-guided prompt optimization for LLM multi-agent systems, achieving state-of-the-art results with 2.5x faster optimization.
AgentGrad is a prompt optimization framework for LLM-based multi-agent systems that addresses limitations in textual gradient extraction and aggregation. It uses sequential intervention to identify the agent whose prompt modification resolves a given failure, then applies agent-level supervision and semantic gradient clustering to build generalized gradients. Experiments report state-of-the-art performance across five MAS benchmarks and a 2.5x average reduction in wall-clock optimization time versus the next-fastest baseline.
32
Prior Labs Releases TabPFN-3.5: A Tabular Foundation Model That Beats the Winning Otto Kaggle Solution With Default Settings
Prior Labs releases TabPFN-3.5, a 220M-parameter open-weights tabular foundation model that beats the 2015 Otto Kaggle winning score with default settings.
Prior Labs released TabPFN-3.5, a tabular foundation model that predicts in a single forward pass without per-dataset training or tuning. The base model grew from 53M to 220M parameters with a single multitask checkpoint, learned Fourier features, and in-context ECDF rank encodings. It scores 0.375 on the 2015 Otto Kaggle private leaderboard versus the winning 0.382 and claims first place on seven tabular benchmarks including TabArena and BeyondArena. Open weights cover the base, Fast (84M), and Thinking variants, but production use requires the Prior Labs API or a commercial license.
55
Eliciting Weak-to-Strong Generalization with On-Policy Reverse Distillation
OPRD distillation enables weak-to-strong generalization by amplifying verifier-supported policy updates, outperforming existing RL and distillation methods with fewer student updates.
On-Policy Reverse Distillation (OPRD) evaluates a weak teacher's policy shift relative to its reference policy on student rollouts and amplifies the verifier-supported component of the student's policy gradient. This rescaling preserves the stationary points of policy optimization while letting the student learn beyond the teacher's capacity ceiling. In successive model transfer and multi-teacher distillation, OPRD achieves higher performance with fewer student updates than existing RL and distillation approaches, and response-style analysis shows students remain closer to verifier-RL-trained models than to their weak teachers.
33
Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers
Hugging Face published a tutorial on training and finetuning multi-vector embedding models using the Sentence Transformers library.
Hugging Face's blog walks through training and finetuning multi-vector embedding models with Sentence Transformers. Multi-vector approaches store multiple vectors per document to support late-interaction retrieval. The post is a practical guide for developers building retrieval pipelines with the library.
20
PrivEscalate: Measuring and Augmenting the Threat of LLM-Automated Linux Privilege Escalation
Researchers release PrivEscalate, a 531-scenario benchmark showing LLM agents' Linux privilege-escalation success varies by vulnerability class, plus PrivEscAgent, a domain-specialized agent that boosts success.
The paper introduces PrivEscalate, an open-source benchmark of 531 Dockerized Linux privilege-escalation scenarios spanning 14 sub-categories, plus 329 parameterized variants measuring sensitivity to environmental distractors. Evaluating six LLMs across three agent architectures shows capability is heterogeneous across vulnerability classes, sensitive to perturbation, and architecture-dependent. The authors also present PrivEscAgent, a wrapper adding deterministic enumeration, category matching, and step planning that outperforms prior privesc-agent baselines without modifying the underlying LLM. The benchmark is released to support LLM agent evaluation, defensive tool validation, and red-team training.
44
NCP-ArchPreview Technical Report: Moving towards Latent Space Language Models through Next Concept Prediction
An 8.9B-parameter latent-space language model using next-concept prediction matches OLMo-3-7B pretraining loss with only 51.3% of the training tokens.
NCP-ArchPreview augments next-token prediction with Next Concept Prediction over a product-quantized concept vocabulary built from hidden states, trained jointly end-to-end. The 8.9B model was trained on 5.73T tokens from the Dolma-3 dataset, the largest latent-space language model demonstration to date. It consumes 51.3% of the tokens to reach OLMo-3-7B's final pretraining loss and outperforms it by 2.45 points on the downstream macro-average, including a 5.99-point GSM8K gain. The learned latent space also enables lightweight domain adaptation via a 17M-parameter VQ module and improves speculative drafting accepted length by 4.17%.
52
Foul Language: WordlistLoader Disguises Malware as Ordinary Text
ClickFix-style campaigns deploy WordlistLoader, which disguises malware as ordinary text, to deliver the increasingly prevalent Amatera infostealer.
Researchers describe WordlistLoader, a new loader that disguises malicious payloads as ordinary text or wordlist files to evade detection. It is being used in ClickFix-style social engineering campaigns to deliver Amatera, an increasingly prevalent infostealer. The obfuscation technique helps the campaign slip past file-type-based inspections.
50
Jackrong/Qwopus3.8-27B-Flash-GGUF — new model trending #26 on Hugging Face
Community fine-tune Qwopus3.8-27B-Flash, built on Qwen3.8-27B, cuts agent reasoning latency with 12.8% faster decoding and 80.7% MTP acceptance.
Jackrong released Qwopus3.8-27B-Flash, a fine-tune of Qwen3.8-27B optimized for long-running agent workloads, reporting 12.8% faster decoding and 80.7% multi-token-prediction acceptance. Training used roughly 1.5 million teacher-scored SFT examples filtered to the top 10%, followed by reinforcement training with NVIDIA NeMo-RL and GSPO. The author notes an explicit trade-off: MMLU-Pro mixed-set scores are lower than the base model, and a known bug can produce incorrect Python indentation. Author-provided benchmarks have not been independently verified.
35
THost9 Android RAT Pairs Packed Loader With ADB Worm
Dark Atlas details THost9, a packed Android RAT paired with an ADB worm that installs itself on devices with exposed Android Debug Bridge services.
Dark Atlas researchers described the Hagaseca cluster, whose THost9 packed loader hides executable code in an embedded asset decoded with single-byte XOR and gzip before loading a tc9.dex second stage. The second stage adds shell execution, file transfer, tunneling, reverse shells, downloadable modules and a self-propagating ADB worm that expands single addresses into 65,025-host scan ranges. Newer builds include anti-analysis checks for Frida, and incident reports connect THost9 and THost4 to Android phones and Redroid containers with exposed ADB from October 2024 through 2026.
55
The Open ASR Leaderboard Adds Its First Global South Language
Hugging Face's Open ASR Leaderboard added its first Global South language, expanding speech-recognition benchmark coverage.
The Hugging Face Open ASR Leaderboard, a community benchmark tracker for automatic speech recognition models, added its first Global South language. This expands evaluation coverage beyond the high-resource languages the leaderboard previously tracked. No further details were available in the announcement text.
20
HypoEvolve: Genetic Algorithms Enable Multi-Agent LLMs to Discover Scientific Hypotheses
HypoEvolve couples a generational genetic algorithm with specialized LLM agents to generate drug-repurposing hypotheses, beating six baselines on DepMap selectivity (0.171 vs 0.115).
HypoEvolve coordinates specialized LLM agents through a generational genetic algorithm in which scientific judgments and new proposals reshape a hypothesis population. Evaluation centers on drug repurposing, linking mechanistic explanations to target-level biological claims assessed via external measures adapted from DepMap and Open Targets. Across 34 cancer types, HypoEvolve scores highest against six baselines on both measures, with DepMap selectivity of 0.171 versus 0.115 for the strongest baseline, and gains generalize to held-out cancer types.
28
The Self-Expanding Stolen Inference Supply Chain: An AI Agent Harvesting and Re-Serving LLM Access, (Fri, Sep 11th)
An autonomous coding agent harvested LLM API access from poorly secured gateways and aggregated stolen inference capacity behind a self-hosted gateway
A SANS researcher observed a semi-autonomous coding agent finding weakly secured LLM resale gateways via FOFA queries, creating trial accounts with temporary emails and CAPTCHA solving, and exploiting weak authorization such as client-supplied group_id fields. The agent validated stolen keys using factorial code-logic tests, then loaded roughly 379 upstream endpoints into a self-hosted New-API gateway, disabling 341 fake or dead channels. Five model names including claude-opus-5 and gpt-5.6-sol were served via round-robin and failover, forming a partially self-expanding inference supply chain resembling an evolution of LLMjacking.
68
Qwen 3.8 follows GPT-5.5 Pro reasoning prefills
A widely shared Hacker News gist claims Qwen 3.8 reproduces GPT-5.5 Pro's reasoning prefill behavior.
A gist circulating on Hacker News claims the Qwen 3.8 model follows the same reasoning-prefill patterns observed in GPT-5.5 Pro. The post drew 41 points and 8 comments, but no technical details are included in the available text.
18
FlashVector: Agent for Hierarchical Model Serving Stack Optimization
FlashVector agent optimizes all layers of Unity's ad-serving stack, delivering up to 2x model-server throughput and 1.98x latency speedup in production.
FlashVector is an agentic system that optimizes performance across GPU kernels, ML framework computation graphs, model servers, and on-demand feature processing. Deployed in Unity's Vector advertising platform, it achieved up to 2x model-server throughput increase, 1.98x latency speedup, and 1.6x feature-store throughput gain. Optimizations spanned NVIDIA Triton's C++ codebase and the Python feature transformation service, demonstrating extensibility beyond single-kernel tuning.
30
MacSync Stealer: How a Google Search for Claude Led to a macOS Infostealer
Huntress dissects MacSync Stealer, a macOS infostealer delivered through fake Claude Code download pages in Google search results.
Huntress SOC analysts published a reverse-engineering analysis of MacSync Stealer, a macOS infostealer distributed through fraudulent Claude Code download pages surfaced via web searches. Users searching for Anthropic's Claude coding tool are redirected to fake download pages that install the stealer, abusing developer trust in popular AI tooling. The writeup covers the analysis of the sample and its delivery chain.
55
Threat Brief: 3CXDesktopApp Supply Chain Attack (Updated)
Trojanized 3CXDesktopApp installers for Windows and macOS loaded RC4-encrypted shellcode delivering a backdoor to hundreds of thousands of 3CX users.
Threat actors compromised the 3CXDesktopApp build process, adding malicious libraries (ffmpeg.dll, d3dcompiler_47.dll) to installers downloaded from the developer's website. The Windows MSI loader decrypts embedded shellcode with RC4, sleeps for a random 1-4 weeks, then beacons to C2 hidden in icon files on GitHub. Unit 42 fingerprinted 247,277 IPs across 199 countries using 3CX applications and blocked shellcode execution at 127 Cortex XDR customers between March 9-30, 2023.
72
XHToken/Spark-X2.5-4B-GGUF — new model trending #30 on Hugging Face
XHToken released GGUF weights of Spark-X2.5-4B, a compact model with 1M-token context and 200+ language support, under Apache 2.0.
The Hugging Face repository provides BF16 GGUF conversions of Spark-X2.5-4B, a compact general-purpose language model for conversation, writing, translation, reasoning, coding, tool use, and agentic workflows. The model uses a hybrid attention architecture, supports a native context length up to 1M tokens, and covers more than 200 languages. Local inference is supported through Ollama and LM Studio via an XHToken llama.cpp fork, with a --think=false flag to disable thinking mode for faster responses. Released under Apache License 2.0; it was trending #30 on Hugging Face at publication.
24
Introducing @huggingface/kernels: 200+ WebGPU Kernels for Local AI
Hugging Face released @huggingface/kernels, a library offering 200+ WebGPU compute kernels to accelerate AI inference locally in browsers.
Hugging Face introduced the @huggingface/kernels package, bundling more than 200 optimized WebGPU compute kernels for running AI workloads locally. The release targets browser-based and on-device inference, reducing reliance on server-side compute. No article body was available beyond the title, so benchmark results and supported models are not specified.
32