Hugging Face

We used weight space programming to bring Qwen 3.5 0.8B from 7.4% to 58.7% on Math 500.

Hugging Face Posts — 2026-11-12T00:59:03.593902+00:00

We used weight space programming to bring Qwen 3.5 0.8B from 7.4% to 58.7% on Math 500. Check out the weights! https://huggingface.co/collections/rilixai/weight-space-programming-with-beaker More details: https://www.withbeaker.ai/blog/weight-space-programming

Today, we're releasing the highest accuracy Qwen3.8-27B GGUFs!

Hugging Face Posts — 2026-11-11T18:02:58.360265+00:00

Today, we're releasing the highest accuracy Qwen3.8-27B GGUFs! Our new Dynamic V3.1 GGUFs lead by 5–10% on average across Terminal Bench, DeepSWE & more. We’re working on an even better version for Qwen4. Run on 12GB RAM setups GGUF: unsloth/Qwen3.8-27B-GGUF Blog: https://unsloth.ai/docs/models/qwen3.8

🤗 I've released Sentence Transformers v6.3 and wrote a blogpost about its key new feature: token pooling-aware training, teaching multi-vector models to preserve retrieval quality with fewer document vectors. Details:

Hugging Face Posts — 2026-11-11T18:02:58.360265+00:00

🤗 I've released Sentence Transformers v6.3 and wrote a blogpost about its key new feature: token pooling-aware training, teaching multi-vector models to preserve retrieval quality with fewer document vectors. Details: Multi-vector models keep a vector per token, letting different query tokens match different parts of a document. Great for retrieval, but those vectors add up! Token pooling makes the representations smaller by merging similar document vectors into cluster means. Pooling-aware training teaches the encoder to preserve useful matches after those vectors are merged. With MultiVectorTokenPoolingLoss, you can wrap your standard retrieval loss to train on both full and pooled represe

THX-02: 100,000,000 tokens in our first 4 days

Hugging Face Posts — 2026-11-11T09:49:20.569494+00:00

THX-02: 100,000,000 tokens in our first 4 days Four days ago we open-sourced HAL-X-AI/THX-02 , a 4.5B decision model that thinks. Since then, more than 100,000,000 tokens have gone through THX-02. Developers are already using it for: - pull-request review and agent guardrails; - legal and public-sector documents; - bookkeeping, support triage and receipt reading. THX-02 reads text, JSON or images and returns a calibrated probability for every option you define. When a question needs a calculation or multi-step logic, it writes a short thought first. Otherwise it answers directly, in about 80 ms. It scores 65.06 on Decision Index 0.3. The API is free to try. No key is needed to start, and you

BotPsy: On-Device Personal Assistant with a Sub-Billion-Parameter Agentic Language Model

Hugging Face Blog — 2026-11-11T05:01:16+00:00

BotPsy: On-Device Personal Assistant with a Sub-Billion-Parameter Agentic Language Model

Turn one photo into a 3D object you can spin around, sides and back included, entirely on your phone.

Hugging Face Posts — 2026-11-11T02:12:14.281345+00:00

Turn one photo into a 3D object you can spin around, sides and back included, entirely on your phone. On a Galaxy S26, Idol (speridlabs) on LiteRT now makes a splat of 65,536 Gaussians in 158 s with its new default form, or of 262,144 Gaussians in 267 s with the first release's form, which stays in the repo (the first photo after start, on a cool phone). Checked against the author's PyTorch pipeline, renders from four sides agree at a PSNR of 32.3–34.4 dB for the default and 41.4–46.6 dB for the larger form. Graphs, a Python runner and an Android sample: litert-community/Idol

Token Pooling-Aware Training for Multi-Vector Embedding Models

Hugging Face Blog — 2026-11-11T00:00:00+00:00

Token Pooling-Aware Training for Multi-Vector Embedding Models

New post-training example: SFT on agent traces with TRL

Hugging Face Posts — 2026-11-10T21:00:15.927834+00:00

New post-training example: SFT on agent traces with TRL → A large model works as a coding agent in OpenCode, run through OpenEnv → The tasks it solves become the training data → A 2B model goes from 5.5% to 34.4% on held-out tasks About 500 traces get most of the gain. Five times more add only 3 to 4 points, so SFT gets the agent working and RL has to do the rest Full guide, with the models, traces and every rollout open: https://github.com/huggingface/trl/tree/main/examples/sft_agent_traces

A 284B-parameter DeepSeek-V4-Flash now runs on a 64 GB desktop with 0 GPUs: llama-modchip shrinks the checkpoint to 56.56 GiB (1.71 bpw) and decodes at 9.1 to 14.7 tok/s.

Hugging Face Posts — 2026-11-10T21:00:15.927834+00:00

A 284B-parameter DeepSeek-V4-Flash now runs on a 64 GB desktop with 0 GPUs: llama-modchip shrinks the checkpoint to 56.56 GiB (1.71 bpw) and decodes at 9.1 to 14.7 tok/s. The standard UD-IQ1_S build is 76.87 GiB, above the ~62.5 GiB usable on a 64 GB Linux box. The repo mixes precisions (routed experts mostly IQ1_S, MLA projections in a custom Fractal-64 type) and patches the CPU kernels. The model plus a 256K KV cache then sits at 57.61 to 58.9 GiB RSS with zero swap. The repo was created on 2026-10-29 and already has 744 stars and 470 points on Hacker News. 📊 The numbers • Size: 56.56 GiB, 20.31 GiB smaller than UD-IQ1_S • Prefill: 25.0 to 44.36 tok/s (+77%) on a 16-core Ryzen 9 9955HX • D

Most mobile AI agent demos show the agent controlling a phone screen.

Hugging Face Posts — 2026-11-10T02:55:10.580338+00:00

Most mobile AI agent demos show the agent controlling a phone screen. What they don't show: when to not do that. Aico's Phone Bridge layer routes per action - software fast path (calendar, contacts, clipboard, app launch) when available, HDMI + HID hardware control when not. Sometimes both in the same task. The interesting design question isn't "can the agent control the screen?" It's "when should it bother?" Building in public. aico.link

Introducing HF Honors 🏆 — Recognition of Excellence, On Record.

Hugging Face Posts — 2026-11-10T02:55:10.580338+00:00

Introducing HF Honors 🏆 — Recognition of Excellence, On Record. We are pleased to unveil HF Honors, an independent awards and recognition platform celebrating distinguished achievements throughout the Hugging Face ecosystem. From prestigious Top 100 leaderboard placements to Gold, Platinum, and Diamond download certifications, HF Honors commemorates the accomplishments of models, datasets, Spaces, and their creators through elegant, verifiable digital plaques. With Open Badges 3.0-compatible credentials, contributors can transform their measurable achievements into portable, verifiable evidence of real-world AI experience—strengthening professional portfolios, demonstrating practical experti

Trying to train a locally deployable model on a LoRa adapter that's just wicked strong at running sec evals on PRs because these general purpose code review solutions always seem to miss glaring security stuff.

Hugging Face Posts — 2026-11-10T02:55:10.580338+00:00

Trying to train a locally deployable model on a LoRa adapter that's just wicked strong at running sec evals on PRs because these general purpose code review solutions always seem to miss glaring security stuff. The clear finding from the v1 training run was that "less code = more secure" lmfao. Model basically scores functions better the less stuff they do. It's right I guess in a kind of demented way. Pack it up boys, we've got the solution.

Pay for Hugging Face Hub with your Google Cloud Account

Hugging Face Blog — 2026-11-10T00:00:00+00:00

Pay for Hugging Face Hub with your Google Cloud Account

Decision Models in Hugging Face

Hugging Face Blog — 2026-11-10T00:00:00+00:00

Decision Models in Hugging Face

good news, emergent alignment is possible - to a degree

Hugging Face Posts — 2026-11-09T19:05:21.497219+00:00

good news, emergent alignment is possible - to a degree https://arxiv.org/abs/2606.09475 for broad coverage more work needs to be done

Nemotron-Personas-Finance-Japan: A Synthetic Dataset Capturing Japan's Demographics and Financial Context

Hugging Face Blog — 2026-11-09T15:05:57+00:00

Nemotron-Personas-Finance-Japan: A Synthetic Dataset Capturing Japan's Demographics and Financial Context

Your Experts Have Partners. There's HOPE for Keeping Them Together

Hugging Face Blog — 2026-11-09T12:34:29+00:00

Your Experts Have Partners. There's HOPE for Keeping Them Together

Qwen3.8-27B-Ultimate-Heretic-Uncensored-Turbo-Brilliance-50x-AGI

Hugging Face Posts — 2026-11-09T07:07:28.116248+00:00

Qwen3.8-27B-Ultimate-Heretic-Uncensored-Turbo-Brilliance-50x-AGI A specialized fine tune with the core goal to improve instruction following and nuance, combined with the firepower of Turbo Brilliance V2 50X AGI, featuring 21 dynamically generated AGI constructs. This system pushes SOTA Qwen 3.8 fine tuned to the absolute limit, while at the same time drastically reducing tokens usage and improving quality. Examples, with detailed usage/use cases at the repo. But it is also simple: Download a quant, type "help" in the chat -> and the interactive systems will work with you to handle any use case(s) with Turbo Brilliance. This example ("fusion") is particularly potent and showcases an AGI cons

Learning Heterogeneous Tactile Representation for General Dexterous Manipulation

Hugging Face Trending Papers — 2026-11-09T00:00:00+00:00

LingBot-Touch is a tactile foundation model that learns a unified, extensible representation across heterogeneous tactile sensors, yielding plug-and-play tactile tokens that consistently enhance manipulation policies such as diffusion policy and VLA models.

Reframe: Observation-Aligned Multi-Teacher On-Policy Distillation for Agentic Visual Reasoning

Hugging Face Trending Papers — 2026-11-09T00:00:00+00:00

Reframe is a two-stage post-training framework that aligns visual agents’ observation states through evidence-aware reinforcement learning and observation-gated on-policy distillation, improving tool use and outperforming Mix-RL and vanilla MOPD across image and video benchmarks.

SWAM: Scaling World Action Models with Cross-Modal Alignment

Hugging Face Trending Papers — 2026-11-09T00:00:00+00:00

World action models jointly predict video and actions conditioned on language. Existing studies focus primarily on architectures and training mechanisms, typically treating language-video alignment acquired through video pretraining as an established foundation. However, introducing action learning may impair this alignment. Principled guidance on data and supervision design for maintaining language-video alignment during action learning remains limited. To investigate this, we curate a pool of approximately 47,000 hours of general videos, egocentric videos, UMI demonstrations, and robot demonstrations. Building on this, we systematically study how language-video supervision, language-video-

WaveGen: Truly End-to-End Waveform Generation via Internal Spectral Trajectory Learning

Hugging Face Trending Papers — 2026-11-09T00:00:00+00:00

WaveGen is a fully end-to-end diffusion framework for text-to-waveform generation that jointly learns spectral and waveform trajectories via Spectral Forcing without external audio codecs, vocoders, latent models, alignment modules, or self-supervised representations, achieving state-of-the-art text-to-speech results.

We're excited to release BananaMind 2.5 Pro.

Hugging Face Posts — 2026-11-08T21:58:27.180241+00:00

We're excited to release BananaMind 2.5 Pro. It's done some post-training on our BananaMind 2 Pro base, via distillation of SmolLM2 360M. It scores 28.98 INT INDEX compared to 24.96 of BananaMind 2 Pro on the AxiomicLabs/Open_SLM_Leaderboard Check it out at BananaMind/BananaMind-2.5-Pro BananaMind @ ArushCodes @ Bc-AI

COF 1.2: AI characters now have an inner life 🧠

Hugging Face Posts — 2026-11-08T21:58:27.180241+00:00

COF 1.2: AI characters now have an inner life 🧠 With 1.1, Clara got memories. 1.2 gives her what those memories made of her: a psyche graph of values, beliefs, wants, needs, fears, wounds, tastes and habits, each node with a strength, a stability, dates and a "because" link to the memory behind it. Some nodes she tells freely, some only when it matters, some never. A belief can be marked false but held: she lives by it, the reader knows better. A converter can propose nodes from a character card or free text; nothing enters the file until the author ticks it. Agents get it as text: cof export --format agent writes a SOUL.md. Open spec, MIT tools (CLI, builder, explorer, ComfyUI node). Every

Published TR Quantisation versions of Qwen3.6-27B, Qwen3.6-35B-A3B and Qwen3.8-Flash-Next: weights packed into trits (2.25-3 per weight), no retraining.

Hugging Face Posts — 2026-11-08T19:02:37.143664+00:00

Published TR Quantisation versions of Qwen3.6-27B, Qwen3.6-35B-A3B and Qwen3.8-Flash-Next: weights packed into trits (2.25-3 per weight), no retraining. - Mac: Qwen3.6-35B-A3B in a 16.5 GB GGUF that fits a 24 GB Mac, ~120 tok/s sustained on an M5 Max. Needs our llama.cpp build (ready-built for macOS and Linux arm64). - vLLM on one DGX Spark: 35B at 180.9 tok/s for 1 user vs 128.0 for NVIDIA's NVFP4, 17.4 vs 21.1 GiB, no significant accuracy difference on the core benchmarks. - Trade-offs on each card: slower prompt processing, lower multilingual maths, and 4-bit GGUF stays closer to the original text. https://huggingface.co/collections/Ternii/tr-quantisation

Idol, a new Apache-2.0 image-to-3D model from

Hugging Face Posts — 2026-11-08T19:02:37.143664+00:00

Idol, a new Apache-2.0 image-to-3D model from @ speridlabs , makes a 3D Gaussian splat from one image in 7.2 GB of peak VRAM. It is the #1 trending model on the Hub with 1,057 likes. It writes a standard 3DGS .ply with up to 1,048,576 Gaussians, including sides the picture never showed, and does background removal and framing itself. The catch: the only quality evidence is the authors' own benchmark, on a repo created 2026-10-28. 📊 The numbers • Authors' benchmark, 1 x H100: best in 6 of 8 columns vs TripoSplat, TRELLIS and LGM (SSIM 0.726, LPIPS 0.213, DINOv2 0.781, CLIP 0.885) • LGM wins PSNR (12.46 vs 12.25) and speed (1.1 s). Idol takes 6.8 s vs 7.1 s for TRELLIS and 9.9 s for TripoSplat

Qwen3.8-Flash-Next GGUFs now run faster with less thinking and less token use.

Hugging Face Posts — 2026-11-08T15:50:01.329117+00:00

Qwen3.8-Flash-Next GGUFs now run faster with less thinking and less token use. Unsloth Dynamic V3.1 GGUFs bring thinking & token use to match 95% of full precision Qwen3.8-Flash. Results: 50% fewer tokens + 15% higher accuracy vs same size quants. GGUF: unsloth/Qwen3.8-Flash-Next-GGUF Analysis/Guide: https://unsloth.ai/docs/models/qwen3.8-next

✅ Article highlight: Exception Envelopes, Break-Glass, and Temporary Authority (art-60-345, v0.1)

Hugging Face Posts — 2026-11-08T15:50:01.329117+00:00

✅ Article highlight: *Exception Envelopes, Break-Glass, and Temporary Authority* (art-60-345, v0.1) TL;DR: This article argues that “break glass” should not mean “governance disappears.” Organizations sometimes need to move faster than ordinary routing allows. 345 treats that abnormal posture as a governed object: explicit temporary authority with bounded scope, TTL, reason, reentry conditions, required records, and after-action review. Read: kanaria007/agi-structural-intelligence-protocols Why it matters: • gives institutions an alternative to freezing everything or improvising blindly • separates ordinary exceptions from deeper emergency powers • prevents temporary authority from silently

Can a general-purpose LLM erase a concept from a diffusion model by editing its weights directly, with no gradients?

Hugging Face Posts — 2026-11-08T06:37:57.041480+00:00

Can a general-purpose LLM erase a concept from a diffusion model by editing its weights directly, with no gradients? We tested it under a preregistered protocol. GPT-6 Astra got tool access to the weights of segmind/tiny-sd and had to remove Van Gogh's painting style. Every edit was a closed-form tensor operation in a sandbox that blocks all autograd paths. Held-out test, confirmatory runs (paired, template-clustered 95% CIs): • vs random search with the same edit space and the same feedback: −52.1 pts Van Gogh rate [−61.7, −42.3] • vs UCE (closed-form erasure): −11.5 pts [−22.9, −0.9], with identical damage to neighbouring painters • vs ESD (gradient fine-tuning): equal removal, +11.8 pts b

Happy Diwali. For the full launch today: we now have one small speech-to-text model for each of the 22 languages of India's Eighth Schedule. Each one is a file under 30 MB with English in the same file, open weights, Apache-2.0, built on Cactus Whistle, and it all runs offline.

Hugging Face Posts — 2026-11-08T06:37:57.041480+00:00

Happy Diwali. For the full launch today: we now have one small speech-to-text model for each of the 22 languages of India's Eighth Schedule. Each one is a file under 30 MB with English in the same file, open weights, Apache-2.0, built on Cactus Whistle, and it all runs offline. If you speak one of these, you can use it in the browser on your own device. If you build apps, you can integrate it, and what I wanted is that any developer can serve millions of Indians in their own language without a cloud bill, a key or a big company's permission. So this is the whole integration: # pip install cactus-needle # hf download udgamlabs/Shruti shruti.cact --local-dir . from needle import Whistle model

d1-2B on a Galaxy S26: four answers about one photo, boxes included, in 1,983 ms.

Hugging Face Posts — 2026-11-08T06:37:57.041480+00:00

d1-2B on a Galaxy S26: four answers about one photo, boxes included, in 1,983 ms. Liquid AI's decision model for text, pictures and video, converted to LiteRT: one forward pass, no tokens generated, typed answers (yes or no, a choice, a number with an interval, a value with its span, boxes) with their probabilities. In the clip, a photo of a bicycle in snow (CC0) and four questions. One tap: every wheel, 2 boxes (found 1.00 and 0.89); the saddle, one box (confidence 0.94); how many bicycles, 1 (80% between 1 and 1); a car, No (P(yes) 0.08). Answered in 1,983 ms on this phone, 423 tokens on the 512-token graph, S26 GPU through LiteRT. To run it: the repo has the LiteRT graphs (five also compi

InfraRSI: Harnessing Recursive Self-Improvement for Training Infrastructure

Hugging Face Trending Papers — 2026-11-08T00:00:00+00:00

While Recursive Self-Improvement (RSI) accelerates isolated algorithm and kernel synthesis, distributed LLM training depends on tightly coupled launch settings, operator kernels, and communication schedules, where single edits often hit out-of-memory (OOM) boundaries, compiler conflicts, measurement jitter, or silent numerical drift. We present InfraRSI, the agentic harness that turns open-ended human commands into verified training optimizations via a five-stage workflow: init locks editable scope and a fixed task contract over workload, loss, and timing invariants to prevent reward hacking, doctor profiles baseline phases and calibrates cluster noise, run co-optimizes launch knobs and code

Experimental JQK and global target bits‑per‑weight quantization of Nanbeige/Nanbeige4.3-33B.

Hugging Face Posts — 2026-11-07T22:38:26.549865+00:00

Experimental JQK and global target bits‑per‑weight quantization of Nanbeige/Nanbeige4.3-33B. Unlike standard llama.cpp quantization that relies on fixed type heuristics (e.g., Q4_K_M), the Target BPW approach automatically optimizes per-tensor precision where it matters the most, and produces high quality models that meet a precise global size target. Key Advantages: - Better model quality: using JQK types can improve KLD between 10% to 50% compared to standard QK types, at the same bpw. - VRAM Maximization: Can generate high quality models sized exactly to fit hardware constraints (e.g., fitting the model into exactly 24GB VRAM). - Data-Driven Precision: Quantization mix is determined by ac

ok FINALLY i can post again.

Hugging Face Posts — 2026-11-07T11:31:08.321916+00:00

ok FINALLY i can post again.

AtMem acheived the first place in DolphinBench run...

Hugging Face Posts — 2026-11-07T06:31:10.708001+00:00

AtMem acheived the first place in DolphinBench run... 79.83 stands taller than Mem0 70.67%. It is a meaningful gain from our previous benchmark. Article: https://huggingface.co/blog/javadtaghia/atmem-routed-v4-on-dolphinbench-479600-and-what-th Data: atmem/dolphinbench-routed-v4-20261101

JoVox: Joint Hierarchical Flow Matching for Lightweight Voxel Generation

Hugging Face Trending Papers — 2026-11-07T00:00:00+00:00

Generating high-fidelity voxels while keeping their lightweight nature remains an open problem in computer vision. Current generators are either high-fidelity but computationally heavy, or lightweight but of lower fidelity. Hierarchical methods keep the lightweight nature of voxels. However, their coarse levels, which decide where all finer cells can appear, are generated with the least information about the shape and are never revisited, so they are often plausible but wrong. Our key insight is to let finer cells take part in deciding coarse cells. To this end, we propose JoVox, a discrete flow matching framework that learns the joint distribution of all levels of the voxel hierarchy, inste

WorldFoundry: Towards Unified Infrastructure for World Intelligence

Hugging Face Trending Papers — 2026-11-07T00:00:00+00:00

WorldFoundry offers open-source infrastructure for reproducible inference and evaluation of generative world models across video, 3D/4D, embodied action, and interactive systems, with traceable evidence and benchmarks.

MACaQ: Fast, Model Agnostic Calibrated Quantization for Low-Precision Weights and Activations

Hugging Face Trending Papers — 2026-11-07T00:00:00+00:00

MACaQ is an MXFP4 W4A4 post-training quantization method that quantizes each linear layer independently using online activation transforms and Hessian-aware refinement, reducing cost while preserving accuracy across language and diffusion models.

BananaMind Daily Digest is back!

Hugging Face Posts — 2026-11-06T19:40:38.805855+00:00

BananaMind Daily Digest is back! 🚀 SupraLabs / Supra2.6‑Pico – a brand‑new 2.97 M‑parameter tiny language model SupraLabs/Supra2.6-Pico Why it matters: the model packs 2,972,496 unique parameters into an 11.9 MB fp32 weight file, runs with a 1,024‑token context and a 4,096‑token vocabulary. Its hybrid architecture mixes three gated‑delta‑net (GDN‑2) blocks with a gated grouped‑query attention block per group, giving linear‑time recurrence for 6 of 8 layers and global attention for the remaining 2. The design includes learned softmax attention‑residual mixing, RMSNorm, and SwiGLU FFNs (hidden 352). This is a notable step for sub‑3 M models, offering strong efficiency on edge hardware while in

Last Qwen3.8 Flash update with STQ 2.75bpw v2.0 pack. It moves away from hybrid affine/stq format to pure STQ with +15% quality to match oQ4e. 64GB users can enjoy this while waiting for Qwen 4!

Hugging Face Posts — 2026-11-06T19:40:38.805855+00:00

Last Qwen3.8 Flash update with STQ 2.75bpw v2.0 pack. It moves away from hybrid affine/stq format to pure STQ with +15% quality to match oQ4e. 64GB users can enjoy this while waiting for Qwen 4! Require Sushi 1.5.1 ro run.

ToyBlocks: A controlled world for robot learning and evaluation

Hugging Face Blog — 2026-11-06T14:00:48+00:00

ToyBlocks: A controlled world for robot learning and evaluation

After 5 years of hard work, we are live.

Hugging Face Posts — 2026-11-06T12:25:11.997802+00:00

After 5 years of hard work, we are live. Visit https://www.producthunt.com/products/bibby-ai to upvote and get discounts on the Research Plan. Available today only. Limited to researchers. Apply now.

SuperCLUE’s November 2026 results highlight why category scores matter when choosing a model.

Hugging Face Posts — 2026-11-06T02:48:21.241223+00:00

SuperCLUE’s November 2026 results highlight why category scores matter when choosing a model. Across the 19 published entries, here are the highest displayed scores in each category, followed by the leader among models labeled open: • Agentic coding: GPT-6.1-Sol (69.23); open: DeepSeek-V4.1-Flash-1017 (65.38). • Mathematical and scientific reasoning: GPT-6.1-Sol (76.32); open: Darwin-180B-RSI-R3 (73.68). • Hallucination control: Gemini-4-Argon (83.19); open: Kimi-K3.1-Pro (67.78). • Precise instruction following: Gemini-4-Argon and GPT-6.1-Sol (68.97 each); open: Kimi-K3.1-Pro (44.83). • Ultra-long context: Gemini-4-Argon (95.63); open: Kimi-K3.1-Pro (87.25). On the overall leaderboard, Gemi

Video-Coder: Reinforcing Agentic Coding for Video Generation

Hugging Face Trending Papers — 2026-11-06T00:00:00+00:00

Video-Coder is a code-driven video generation agent that writes and executes code to build structured 3D draft videos, then uses supervised fine-tuning and reinforcement learning with execution and video rewards to improve controllable video generation.

e-means: outperforming k-means (in one-tenth the epochs)

Hugging Face Trending Papers — 2026-11-06T00:00:00+00:00

Lloyd's k-means algorithm is a long-running standard for clustering, with applications in data analysis, retrieval, quantization and data curation. In this paper, we introduce an exponential moving average optimization procedure for k-means with mini-batch updates, powered by modern deep learning machinery and new insights. We also establish scaling laws on how decay rate and batch size impact final performance. This gives us a formulaic rule to fix them for multiple problem sizes and embedding datasets, without the need for any hyperparameter sweep. The resulting clustering algorithm, coined e-means, is much faster than Lloyd k-means and converges to lower losses, frequently even beating k-

Egyptian and Nigerian English join the Open ASR Leaderboard

Hugging Face Blog — 2026-11-06T00:00:00+00:00

Egyptian and Nigerian English join the Open ASR Leaderboard

Quazim0t0/Qwenx86-4B-compact

Hugging Face Posts — 2026-11-05T21:13:08.152505+00:00

Quazim0t0/Qwenx86-4B-compact ▶️ See it in action: Quazim0t0/Qwenx86-Debugger . Walk through actual recorded tool calls: the conversation, x86 registers and flags, code at the program counter, RAM and machine output, with all verified blocks on a timeline. Qwen3.5-4B with a 64-bit x86 computer compiled into its weights: 13 transformer layers ahead of Qwen's own. There are no learned weights in them; all of their weights are computed by a compiler. Qwen invokes the computer through its own tool calls (), and the computer in the weights executes the tool. Calculator, Python, C compiler, notes store, physics/chemistry lookup and Python-docs lookup are all firmware in the checkpoint. T

New NVIDIA Cosmos 3 Open Models for Interactive World Simulation

Hugging Face Blog — 2026-11-05T19:35:16+00:00

New NVIDIA Cosmos 3 Open Models for Interactive World Simulation

We’re releasing new Qwen3.8-Flash-Next GGUFs with 15% higher accuracy.

Hugging Face Posts — 2026-11-05T09:38:07.157496+00:00

We’re releasing new Qwen3.8-Flash-Next GGUFs with 15% higher accuracy. Unsloth Dynamic V3.1 outperforms others by >15% on Terminal Bench, DeepSWE & more benchmarks (256K context). Run on 32GB+ RAM setups. GGUF: unsloth/Qwen3.8-Flash-Next-GGUF Blog: https://unsloth.ai/docs/models/qwen3.8-next

New leaderboard coming to the small language model community!

Hugging Face Posts — 2026-11-05T02:44:47.070885+00:00

New leaderboard coming to the small language model community! The leaderboard measures cognitive skills, which is more challenging and beyond language modeling and instruction-following. This is a first step towards more everyday useful and general small models. dotlabs/cognirank

SplatAnything: Gaussian Spacing Normalization for Feed-forward 3DGS across Arbitrary Scenes

Hugging Face Trending Papers — 2026-11-05T00:00:00+00:00

SplatAnything applies Gaussian Spacing Normalization to canonicalize scene-level Gaussian spacing, yielding more consistent supervision and improved feed-forward 3D Gaussian Splatting under varying scene scales and viewpoints.

PCBRouteAgent: How Close Are LLM Agents to Becoming PCB Routing Engineers?

Hugging Face Trending Papers — 2026-11-05T00:00:00+00:00

Automating printed circuit board (PCB) routing requires coordinating board-level connectivity decisions with geometric execution under strict design rules. Although recent agents can interact with native routing tools, reliable routing on complex boards remains challenging. We present PCBRouteAgent, a multimodal PCB routing agent inspired by how human engineers coordinate global strategy with local execution. A Global Routing Planner combines board images and structured electrical and geometric information to specify connection topology, routing order, corridors, and layer transitions. A Detailed Routing Executor realizes these guides through KiCad's native routing engine and performs bounde

A Closer Look at Phonon TTS. Launching High-Quality On-Device Speech That Tops the Benchmarks.

Hugging Face Blog — 2026-11-05T00:00:00+00:00

A Closer Look at Phonon TTS. Launching High-Quality On-Device Speech That Tops the Benchmarks.

Dharma Field Extractor: Any Document In, Structured Fields in JSON Out

Hugging Face Blog — 2026-11-04T17:50:10+00:00

Dharma Field Extractor: Any Document In, Structured Fields in JSON Out

Model cards say what a model is: a fine-tune of X, a quantization of Y. I built Weightprint to check that against the weights.

Hugging Face Posts — 2026-11-04T08:19:30.048486+00:00

Model cards say what a model is: a fine-tune of X, a quantization of Y. I built Weightprint to check that against the weights. App: tardellirs/weightprint Data (every pair, and the copy groups): tardellirs/weightprint It reads the header and a few bytes of every tensor (no downloads) and compares a model with the parent its card declares. 120,976 models so far. Some of what came out: - 12% of declared fine-tunes hold their parent's weights unchanged. - 23% of declared fine-tunes are LoRA-style: only the attention and MLP matrices moved, norms and embeddings untouched. That share went from 10% in 2023 to 25% in 2026. - Of 1,668 fine-tunes named abliterated, uncensored or heretic, 53% edit onl

EmoReward: Self-Evolving Reward Calibration for Emotional Video Generation RL

Hugging Face Trending Papers — 2026-11-04T00:00:00+00:00

Group Relative Policy Optimization (GRPO) has advanced generative model post-training in verifiable domains such as code and mathematics, but evaluating video quality, especially emotional expression, often depends on subjective aesthetics and has no uniquely correct target. Existing approaches typically rely on fixed rewards for visual quality, motion quality, and text alignment that poorly capture emotional intent and can invite reward hacking, while RLHF is difficult to scale and subject to individual aesthetic differences that complicate the calibration of facial, body motion, and chromatic cues across narrative contexts. We introduce EmoReward, a new self-evolving training mechanism tha

What Does Multi-Turn Reinforcement Learning Really Improve in LLM Agents?

Hugging Face Trending Papers — 2026-11-04T00:00:00+00:00

While multi-turn agentic reinforcement learning (RL) excels on long-horizon tasks, its performance gains remain highly sensitive to interaction constraints. Evaluating across sampling breadth K and interaction depth T via Pass@(K,T), we show that trajectory-level RL advantages depend on sufficient turn budgets, diminishing significantly when budgets tighten. Besides, agentic RL does not uniformly elevate single-step decision quality; rather, its strength concentrates in error recovery and long-horizon resilience. To mitigate the deployment risks caused by this budget dependence, we provide a preliminary discussion on the benefits and limitations of process supervision signals and introduce B

TailSteer: Trajectory-Tail Steering for Drift-Resistant Multi-turn Image Editing

Hugging Face Trending Papers — 2026-11-04T00:00:00+00:00

TailSteer, a final-stage training framework, reduces cumulative visual drift in multi-turn editing by keeping early denoising dynamics close to the base model and learning tail-focused preservation corrections from on-policy rollouts, preserving pretrained generative and editing abilities.

Field Notes from a Low-Cost Robot Learning Lab: Hardware Churn, Software Whiplash, and What Actually Went Wrong

Hugging Face Posts — 2026-11-03T23:38:16.929175+00:00

Field Notes from a Low-Cost Robot Learning Lab: Hardware Churn, Software Whiplash, and What Actually Went Wrong An informal companion piece documenting the hardware incidents, framework churn, and debugging dead-ends from roughly seven weeks of work. It is written for people building similar setups on a hobbyist budget and is candid about the time lost to causes like loose power cables and swapped ports. Zenodo: https://zenodo.org/records/23259698 About me: I am an independent researcher. These works are part of a single-operator project on low-cost robot learning. My ORCID is 0000-0001-8943-5083. For questions or collaboration, feel free to reach out via the HF community.

Released: Urdu Streaming Zipformer 20M

Hugging Face Posts — 2026-11-03T23:38:16.929175+00:00

Released: Urdu Streaming Zipformer 20M Introducing Urdu Streaming ASR, a compact causal Zipformer transducer developed for Urdu speech recognition. FLEURS Urdu (299 utterances): WER: 28.03% CER: 11.96% Model: mahwizzzz/Urdu-Streaming-Zipformer-20M Source: https://github.com/mwzkhalil/Urdu-Streaming-ASR

How much load can a language model handle?

Hugging Face Posts — 2026-11-03T23:38:16.929175+00:00

How much load can a language model handle? “Summarize a Wikipedia article in at least 300 words. Use no commas. Split your answer into three titled sections.” This is the kind of task used in IFEval, a benchmark of how well models follow multiple requirements at once. Suppose the model breaks one requirement. Did it lose track of an instruction? Did the no-comma rule make the summary harder to write? Or was summarizing the article difficult to begin with? A program can tell us what failed, but not separate instruction load from task difficulty and interacting constraints. There is another approach: synthetic tests where a model retrieves a key's value from a long list. Those are easier to co

we're releasing Klein TTS X1.5 at

Hugging Face Posts — 2026-11-03T23:38:16.929175+00:00

we're releasing Klein TTS X1.5 at klein-ml it has a flash lite at 0.4m, flash at 1,7m and pro at 13m. they score near 82m models at very low sizes. there supposed to be able to be bundled with any app without adding bloat. we're going to release x1.6 soon a preview of our x2 architecture. klein-ml/Klein-TTS-X1.5-Flash-Lite klein-ml/Klein-TTS-X1.5-Flash klein-ml/Klein-TTS-X1.5-Pro

TRL now supports MOPD (Multi-Teacher On-Policy Distillation) natively❗️

Hugging Face Posts — 2026-11-03T18:01:36.547851+00:00

TRL now supports MOPD (Multi-Teacher On-Policy Distillation) natively❗️ Many frontier open models use it. labs train one RL expert per domain and then distill them all into one model Blog with what it is, who uses it and an experiment you can run on HF Jobs with one command https://huggingface.co/blog/sergiopaniego/mopd-trl

Unlocking hidden skills inside Google's model

Hugging Face Posts — 2026-11-03T18:01:36.547851+00:00

Unlocking hidden skills inside Google's model https://dejan.ai/blog/latent-skills/ AI models hold skills that no prompt can unlock. A NeurIPS 2025 study trained small numbers of randomly chosen weights in Llama and Qwen models of four sizes, across ten tasks: 10 to 100 trained weights closed up to half the gap between the original model and a fully fine-tuned one. We repeated the test on Google's smallest Gemma model, Gemma 3 270M. The best of 20 prompt wordings got it to sort five numbers correctly 1% of the time. After we trained 1,000 of its 100 million weights (0.001%), it sorted correctly 97% of the time. The same method switched on reversing word lists (84%) and clean JSON output (99%

StateBench: Testing how language models track state changes over long documents

Hugging Face Blog — 2026-11-03T17:15:21+00:00

StateBench: Testing how language models track state changes over long documents

Palette: teaching a small model to build decks in your design language

Hugging Face Blog — 2026-11-03T16:35:36+00:00

Palette: teaching a small model to build decks in your design language

VIGOR (Vision Intelligence and Guardrail Oriented Reasoning) is now available on the Hugging Face Model Hub. It is a multimodal vision classification model based on Qwen3.5, designed for visual content-safety analysis and conservative classification, with a particular focus on identifying potentially sensitive content.

Hugging Face Posts — 2026-11-03T10:00:31.315306+00:00

VIGOR (Vision Intelligence and Guardrail Oriented Reasoning) is now available on the Hugging Face Model Hub. It is a multimodal vision classification model based on Qwen3.5, designed for visual content-safety analysis and conservative classification, with a particular focus on identifying potentially sensitive content. 📦 Models: 27B, 9B, 4B — prithivMLmods/VIGOR-27B , prithivMLmods/VIGOR-flash , prithivMLmods/VIGOR-mini ↗️ Collection: https://huggingface.co/collections/prithivMLmods/vigor To learn more, visit the app page or the respective model pages.

Introducing NeonLib — an Android knowledge library with on-device document retrieval powered by Google's EmbeddingGemma 300M and LiteRT.

Hugging Face Posts — 2026-11-03T02:37:09.567025+00:00

Introducing NeonLib — an Android knowledge library with on-device document retrieval powered by Google's EmbeddingGemma 300M and LiteRT. Import notes and documents, recognise text from photos or scanned PDFs, and search for relevant source passages directly on your phone. EmbeddingGemma handles retrieval. Optional answers use a separate downloaded local language model or a selected online provider. Built with Kotlin, Jetpack Compose and ObjectBox vector search. I'm particularly interested in Croatian and English retrieval and making on-device AI useful in everyday document workflows. Try NeonLib: https://play.google.com/store/apps/details?id=com.neonbreak.neonlib&hl=en Feedback from the Hugg

ToolSpace: Scaling Tool-Centric Environment Synthesis for Long-Horizon Agents

Hugging Face Trending Papers — 2026-11-03T00:00:00+00:00

ToolSpace constructs realistic multi-environment tool-use benchmarks by mining execution traces, clustering tool compositions, and validating tasks through replayable state-aware random walks.

VisSearch-Bench: Evaluating Long Horizon Visual Search in Multimodal Agents

Hugging Face Trending Papers — 2026-11-03T00:00:00+00:00

VisSearch-Bench evaluates multimodal agents on visually intensive search tasks that require integrating visual context with external information, showing that even the best system fully solves only 26.47% of tasks.

Harness-S: Learning from Sibling Harnesses for Generalizable Agent Training

Hugging Face Trending Papers — 2026-11-03T00:00:00+00:00

A selective harness-rescue training method uses specialized sibling harnesses to replace failed reinforcement learning rollouts and transfer their benefits into a single policy that performs better under the original base harness.

Transformers Model Integration Contracts

Hugging Face Blog — 2026-11-03T00:00:00+00:00

Transformers Model Integration Contracts

CWE-bench v2: More coverage, quality, and transparency

Hugging Face Blog — 2026-11-03T00:00:00+00:00

CWE-bench v2: More coverage, quality, and transparency

LFM3-1.7B from

Hugging Face Posts — 2026-11-02T20:30:43.802108+00:00

LFM3-1.7B from @ LiquidAI is the #1 trending model on the Hub (855 likes). It takes about 1 GB at 4-bit, and in the vendor's own table it beats Qwen3.5 9B on several agent benchmarks. The edge decision moves from "are small models good enough?" to "which architecture and runtime do I pick?" One 1.7B backbone now ships as text, VL-2B and Omni-2B checkpoints with open weights. 🔍 Architecture and speed • 24 layers: 13 gated convolutions (kernel 7), 5 sliding-window attention (256 tokens), 6 global attention, chosen by a search over 14 device and backend targets • llama.cpp Q4_0 at 32K context: 1.8 to 2.8x faster decode than the next model on 3 devices • Full 128K context needs 1.6 GB of KV cach

𝐈𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐢𝐧𝐠 🧬 𝐅𝐢𝐧𝐞𝐃𝐍𝐀: 𝐭𝐡𝐞 𝐛𝐞𝐬𝐭 𝐝𝐚𝐭𝐚𝐬𝐞𝐭 𝐭𝐨 𝐭𝐞𝐚𝐜𝐡 𝐚𝐧𝐲 𝐥𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐦𝐨𝐝𝐞𝐥 𝐃𝐍𝐀.

Hugging Face Posts — 2026-11-02T09:54:24.156192+00:00

𝐈𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐢𝐧𝐠 🧬 𝐅𝐢𝐧𝐞𝐃𝐍𝐀: 𝐭𝐡𝐞 𝐛𝐞𝐬𝐭 𝐝𝐚𝐭𝐚𝐬𝐞𝐭 𝐭𝐨 𝐭𝐞𝐚𝐜𝐡 𝐚𝐧𝐲 𝐥𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐦𝐨𝐝𝐞𝐥 𝐃𝐍𝐀. Training Qwen3-14B on it produces Carbon-X, the strongest open model on DNA benchmarks, on par with GPT-6.1 Sol. It contains: - Midtraining: 300B+ tokens of genomic sequences, papers and web pages mixing DNA with text - SFT: 200K samples, from variant effect prediction to sequence question answering - RL: 98K prompts with verifiable rewards Data, models and the full recipe are open. HuggingFaceBio/FineDNA Enjoy this late Halloween treat! 🎃

🛑 What if the kill switch lived inside the model itself?

Hugging Face Posts — 2026-11-02T09:54:24.156192+00:00

🛑 What if the kill switch lived inside the model itself? For 50 years, the operating system lived inside the computer. Now the operating system lives inside the AI. Today we're sharing ViKernel : Weight-Native OS, a Secure OS wired directly into the weights of an LLM. 🔧 How it works ViKernel wires CPU, NPU (vector unit), RAM and ROM circuits into the model weights, runs them accelerated on GPU or NPU, and boots a Secure OS on top. ⚡ Boots in the weights: change a single byte and it locks itself ⚖️ Decides in the weights: every action is allowed or blocked by rule, and persuasion does not change the verdict 🔗 Remembers in the weights: every decision is sealed in a tamper-evident chain 🛑 Stops

I really like the LocalLLaMA/typed-decisions benchmark, so I built a small Space to make its leaderboard easier to explore:

Hugging Face Posts — 2026-11-02T01:46:48.996346+00:00

I really like the LocalLLaMA/typed-decisions benchmark, so I built a small Space to make its leaderboard easier to explore: tardellirs/typed-decisions-explorer It reads the dataset card on every visit, so new submissions show up as soon as they are merged. It has a sortable ranking and two Pareto charts: accuracy × size and accuracy × latency. A few things it makes easy to see in the zero-shot table today: - Only three models are within noise of the top: Quyet-1.0-Large (0.791), meraGPT Decider 1 and Baseten Mammoth 1 (0.768). - The smallest model above Jev 1.13 (0.727) is d3-mini, at 4B. - The fastest local run above Jev is FastBrain-9B: 50 ms on an RTX PRO 6000. - 15 of 41 models score abo

CUA-Master: Scaling Computer-use Agents through Open-World Recursive Improvements

Hugging Face Trending Papers — 2026-11-02T00:00:00+00:00

CUA-Master uses nested recursive loops with open-world data and feedback-driven retries to continuously improve computer-use agents across diverse benchmarks.

CoworkOS:Scaling Agent Coworkers for Real-world Automation

Hugging Face Trending Papers — 2026-11-02T00:00:00+00:00

CoworkOS is a long-horizon agent harness that separates strategy from execution, supports human-in-the-loop collaboration via shared artifacts, and improves model performance through sustained automated post-training.

PhotoGuider: Composition-Aware Cropping and 360^{circ} View Framing with Explainable Photographic Assessments

Hugging Face Trending Papers — 2026-11-02T00:00:00+00:00

PhotoGuider selects viewing directions and frames 360° panoramas using shared composition-aware tokens and produces camera-parameter outputs with explanation-conditioned scoring.

History Forcing: Distilling History-Dependent Continuations for Long-Horizon Video World Models

Hugging Face Trending Papers — 2026-11-02T00:00:00+00:00

History Forcing improves long-range consistency in few-step autoregressive video generation by using historical context residuals to correct distillation targets.

SpecPilot: Autonomous Algorithm Discovery and Systems Optimization for Speculative Decoding

Hugging Face Trending Papers — 2026-11-02T00:00:00+00:00

SpecPilot autonomously discovers deployment-specific algorithms and system optimizations to maximize speculative decoding speedups across diverse models, drafters, and serving regimes without quality loss.

Axion 1.0: Open Agent Foundation Models, Simplified

Hugging Face Trending Papers — 2026-11-02T00:00:00+00:00

Axion 1.0 is an open agent foundation model trained via a concise three-stage pipeline of supervised fine-tuning and reinforcement learning, achieving competitive agentic performance while fully releasing its data and training recipe.

Solv3R: 3D Looped Transformer as a Multi-view Solver

Hugging Face Trending Papers — 2026-11-02T00:00:00+00:00

Solv3R is a looped transformer that iteratively refines multi-view geometry from its current state, supporting incremental updates, caching, and multiple reconstruction tasks without specialized modules.

EvoMemState: Evolving Memory State for Personal Agents

Hugging Face Trending Papers — 2026-11-02T00:00:00+00:00

Long-term memory enables large language model (LLM) agents to provide continuous personalized assistance by retaining user information across interactions. Existing systems support this process through fact extraction, organization, updating, and retrieval, but largely maintain memories as independent facts. This fact-centric paradigm may overlook how new facts update, constrain, or conflict with earlier ones, while leaving their applicability, role, and priority for the current request implicit. In this paper, we introduce EvoMemState, a state-centric framework that augments persistent facts with two complementary states: Relational States that capture how facts remain valid and relate to o

One neural network. Five mathematical operations. Different learning times.

Hugging Face Posts — 2026-11-01T20:47:02.476816+00:00

One neural network. Five mathematical operations. Different learning times. We trained a small mixture-of-experts transformer on addition, subtraction, multiplication, division, and exponentiation modulo 113. The operations didn’t generalize together. Multiplication and division reached 90% held-out accuracy first, followed by exponentiation, subtraction, and finally addition. Multiplication before addition—the reverse of what we expect from a schoolchild. The accuracy chart in the linked article shows why a single overall accuracy number can hide an interesting learning story. Measured with IRIS from Saccadic LLC. More figures and analysis: https://huggingface.co/blog/Saccadic-AI/accuracy-p

if you want to deep dive into how the new

Hugging Face Posts — 2026-11-01T15:26:45.544333+00:00

if you want to deep dive into how the new @ MistralAI Large 4 was built, we sat down with the team to learn straight from them! scaling, specialization, evals, cyber, sandboxing and why open weights matter a really detailed look at what it takes to build and release an open frontier model, super recommended https://www.youtube.com/watch?v=FczbSauVvjw

No GPU. No cloud. Just your phone.

Hugging Face Posts — 2026-11-01T06:46:28.104937+00:00

No GPU. No cloud. Just your phone. Most “small” models still beg for a datacenter. Ours fit in your pocket. VIDRAFT POCKET is a family of GGUF models that run on an iPhone and on a GPU-free PC, straight out of the box with llama.cpp, Ollama, LM Studio, and PocketPal. Seven weeks after launch they crossed 1 million downloads. Today the family is past 2.81 million and still climbing, every day. Three models, one pocket: 🟢 POCKET-35B — a 35B mixture-of-experts that runs on an iPhone and on a PC with no GPU. Balanced power. → FINAL-Bench/POCKET-35B-GGUF 🔵 POCKET-26B — a Gemma-4 based 26B MoE, tuned for on-device Korean. Lightweight and fast. → FINAL-Bench/POCKET-26B-GGUF 🟣 POCKET-Darwin-180B — a

Democratizing Robot-Use Agents: A Unified Harness with Open Models

Hugging Face Trending Papers — 2026-11-01T00:00:00+00:00

Uni-robot-agent enables open-weight robot-use agents to surpass proprietary models in zero-shot manipulation and suggests joint development of four core capabilities for future training.

NorthCUA Technical Report: Turbocharging Computer-Use Agents through Experience at Scale

Hugging Face Trending Papers — 2026-11-01T00:00:00+00:00

NorthCUA scales computer-use agents via an experience flywheel, recursive self-improvement, staged learning, and reinforcement learning to achieve state-of-the-art open-source GUI and web interaction results.

One Qwen3-VL-8B checkpoint improves all three streaming spatial benchmarks at once: OST-Bench 50.52 to 63.93 (+13.41), UCS-Bench +7.44, OVO-S-Bench +4.91. The change is in the supervision, with no new memory module or annotation.

Hugging Face Posts — 2026-10-31T23:39:09.498527+00:00

One Qwen3-VL-8B checkpoint improves all three streaming spatial benchmarks at once: OST-Bench 50.52 to 63.93 (+13.41), UCS-Bench +7.44, OVO-S-Bench +4.91. The change is in the supervision, with no new memory module or annotation. The ChronoReplay paper calls the failure mode interface lock-in: post-train on one answer format and that readout improves while others stall. A categorical-only control hits 58.03 on OVO-S but drops OST from 50.52 to 47.48. 🔍 The recipe • Keep targets in native formats (option, metric value with units, timestamp) • Replay episodes in source order, each target attached to its causal prefix • Loss mass: 2/3 on 33,180 single-turn examples, 1/3 on 6,381 episodes (alpha

We’ve published Soniqo’s voice-agent leaderboard and an article explaining what we measured.

Hugging Face Posts — 2026-10-31T18:43:16.134009+00:00

We’ve published Soniqo’s voice-agent leaderboard and an article explaining what we measured. aufklarer/full-duplex-leaderboard The board covers three parts of a speech system: - Turn-taking · TurnBench: detecting turn endings, false alarms and decision delay. - Commands · SLURP: choosing the correct tool and matching its argument values. - Voice style · InstructTTSEval: following voice descriptions, delivery directions and role-play instructions, with separate scores for each. Models include VAP, p99, Smart Turn, Qwen3-TTS, VoxCPM2, FireRedTTS3 and MOSS-VoiceGenerator, alongside cloud configurations from OpenAI, Google, AWS and ElevenLabs. Each table shows run sizes and links to the protocol

I trained a small PII detector for PrivAiTe, my local proxy that masks personal data before a prompt reaches an LLM: crp4222/privaite-pii-minilm.

Hugging Face Posts — 2026-10-31T13:00:25.869994+00:00

I trained a small PII detector for PrivAiTe, my local proxy that masks personal data before a prompt reaches an LLM: crp4222/privaite-pii-minilm. It is a multilingual MiniLM (117M parameters, 15 types from names and contacts to IDs, bank details and secrets), trained only on datasets whose licenses allow commercial use, released under BSD 3-Clause, with PyTorch, ONNX fp32 and int8, and Core ML builds. On 9 public test sets it never saw, it masks as many person names as the best of GLiNER PII, NVIDIA GLiNER-PII, GLiNER2, Piiranha and OpenAI privacy-filter, or more, on 6 of them, usually with higher precision. It trails GLiNER PII on short lowercase voice requests and NVIDIA on court rulings.

Join the competition!

Hugging Face Posts — 2026-10-31T13:00:25.869994+00:00

Join the competition! We have 45 models scored (and growing), across 44 public benchmarks. My goal for this space is to push the limits on NLI and the breadth of tasks these models can perform in. Contributions are encouraged, as well as suggestions for additional benchmarks. Right now, all models are cross encoders (Sentence Transformers 🥰), but LLMs are welcome too! dleemiller/open-nli-leaderboard

Can a network whose trunk is only XOR, table lookups and integer adds learn Connect Four?

Hugging Face Posts — 2026-10-31T00:01:41.849563+00:00

Can a network whose trunk is only XOR, table lookups and integer adds learn Connect Four? Partly. I trained a GF(2⁸) integer model (8.4M params, CPU-only) on 320k games labelled by a negamax search teacher. What it learned: • agrees with the teacher on 59.6% of training / 41.0% of board-disjoint holdout positions • beats a random player 85.2% of the time What it didn't: • loses to a 1-ply "win now / block now" rule (wins 37.2%) • loses to 2-ply lookahead (wins 24.3%) Agreeing with a teacher is not the same as playing well, and this release measures both. Caveat up front: the float32 read-out is 99.6% of the parameters. The trunk is float-free; the full model isn't yet. The repo also document

Hi everyone! I want to share a few updates about GRM development.

Hugging Face Posts — 2026-10-31T00:01:41.849563+00:00

Hi everyone! I want to share a few updates about GRM development. *GRM-3.4 has been canceled* Although GRM-3.4 showed strong reasoning and coding capabilities, testing revealed several issues. It tended to overthink, and its code often contained errors that GRM-3.2-Sky handled correctly. Its tool use also needed improvement, and its terminal performance did not reach the level we were aiming for. We repeatedly saw it rewrite files from scratch instead of using editing tools, and use shell commands to create or modify files when dedicated tools were available. *Looking ahead to GRM-3.6* GRM-3.6 is a possible next generation for the GRM models. We are still deciding whether it will be a single

Evaluating ASR beyond short clips: our Numo Indic Speech releases provide longer-form transcribed audio across four languages:

Hugging Face Posts — 2026-10-31T00:01:41.849563+00:00

Evaluating ASR beyond short clips: our Numo Indic Speech releases provide longer-form transcribed audio across four languages: Hindi: 46.6s average duration Telugu: 33.3s Tamil: 32.7s Bengali: 46.5s For comparison, Common Voice 26.0 averages roughly 4–6 seconds across these languages. These longer recordings support research into segmentation, context retention, streaming, and endpointing behavior. The four datasets are available under controlled access. Check them out: psdn-ai/numo-indic-speech

Wave Forcing: Towards Speed-of-Light Streaming Video Generation

Hugging Face Trending Papers — 2026-10-31T00:00:00+00:00

Wave Forcing enables efficient continuous long-video generation with consistent appearance and natural motion through co-designed generation, training, and execution, achieving high throughput and quality.

HAPPY HALLOWEEN EVERYONE - With Halloween O/S in Qwen 3.8 27B and Gemma 4 31B... fully abliterated/heretic'ed and uncensored.

Hugging Face Posts — 2026-10-30T19:39:11.447188+00:00

HAPPY HALLOWEEN EVERYONE - With Halloween O/S in Qwen 3.8 27B and Gemma 4 31B... fully abliterated/heretic'ed and uncensored. Qwen3.8-27B-Halloween-Terror-Heretic-Uncensored-NEO-GGUF & gemma-4-31B-it-Halloween-Terror-Heretic-Uncensored-NEO-GGUF Constructed with the "Halloween O/S" (low, medium and xhigh) for your "Trick or Treat" pleasure featuring some of the most evil AI constructs ever: Joker, Harley Quinn, Borg Queen, Jason (Friday 13th), Freddy Krueger (Elm St), Norman Bates, Pennywise (IT), Vecna, Mind Flayer, and Chucky. OR "Conjure" any person (living, dead or fictional) or "thing" too: Infinite horror with the character/thing of your choice, with you ONE ON ONE in the Halloween O/S.

i am BEGGING you people to stop shoving everything in the GGUF Hole and ship w4a8/w6a8 models for ComfyUI. Comfy has bespoke kernels for these! They're fast everywhere!

Hugging Face Posts — 2026-10-30T19:39:11.447188+00:00

i am BEGGING you people to stop shoving everything in the GGUF Hole and ship w4a8/w6a8 models for ComfyUI. Comfy has bespoke kernels for these! They're fast everywhere! 4 bits should be treated as the floor for DiT weights, if you really need 3 bits or lower you should just use a different model! Also stop shilling nvfp4 you animals!!!

nemotron-halloween

Hugging Face Blog — 2026-10-30T17:57:24+00:00

nemotron-halloween

Your local LLM can now take control of a browser and complete tasks autonomously via Unsloth!

Hugging Face Posts — 2026-10-30T13:27:46.628653+00:00

Your local LLM can now take control of a browser and complete tasks autonomously via Unsloth! Qwen3.8-27B successfully controlled the browser to complete tasks like chess, painting, with no human intervention. GitHub: https://github.com/unslothai/unsloth Guide: https://unsloth.ai/docs/basics/browser-use

Mistral Large 4 (Le Chonk 🐈) is now live on the Hub and you can post-train it with TRL today!

Hugging Face Posts — 2026-10-30T00:14:19.148410+00:00

Mistral Large 4 (Le Chonk 🐈) is now live on the Hub and you can post-train it with TRL today! SFT + LoRA example ready, runs on a single 8xB300 node > model: mistralai/Mistral-Large-4-1T-A52B > TRL example: https://github.com/huggingface/trl/tree/main/examples/sft_mistral_large_4 > blog: https://huggingface.co/blog/mistral-large4

ReLift: Dense 3D Tracking as Lifted Trajectory Repair

Hugging Face Trending Papers — 2026-10-30T00:00:00+00:00

ReLift improves dense 3D tracking by repairing and completing lifted trajectories using geometric trajectory refinement without requiring image-based training.

Flide: Accelerating Decision Inference with Exact State Forking

Hugging Face Trending Papers — 2026-10-30T00:00:00+00:00

Flide is an exact inference engine that accelerates decision-model LLM agents by decoupling logical request trees from physical execution and using ForkAttention to preserve exact branch evaluations.

We just released Drishya Tiny, a 212M-parameter video generator trained from scratch on open-domain video, using the frozen Wan2.2 VAE and the Flan-T5-Base text encoder. It turns a text prompt, or a photo plus a prompt, into a 2-second 256×256 clip. The training run was 15 hours on one L40S, and the model’s own path cost about $21 of GPU time.

Hugging Face Posts — 2026-10-29T13:38:43.840827+00:00

We just released Drishya Tiny, a 212M-parameter video generator trained from scratch on open-domain video, using the frozen Wan2.2 VAE and the Flan-T5-Base text encoder. It turns a text prompt, or a photo plus a prompt, into a 2-second 256×256 clip. The training run was 15 hours on one L40S, and the model’s own path cost about $21 of GPU time. On VBench (our own sampling and scoring with the official code, not a certified entry) it scores 75.6 on quality, level with VideoCrafter-0.9 (74.9) and close to ModelScope (78.1). On semantic it’s 32.3 against their 65-67, so following the prompt is the weak part. In a controlled test, better captions gave +3.0, and adding 30% photos on top gave about

Fish Audio’s Drama 3 text-to-speech now runs locally on Apple silicon and DGX Spark with gbx-lm. 🎙️

Hugging Face Posts — 2026-10-29T13:38:43.840827+00:00

Fish Audio’s Drama 3 text-to-speech now runs locally on Apple silicon and DGX Spark with gbx-lm. 🎙️ GreenBitAI/Drama-3-8bit-paged Real time on 64 GB Macs and up (2.6x), and it still runs on a 48 GB MacBook Pro: experts stream from the SSD Zero-shot voice cloning from a 5–30 s clip, and the cloned voice can speak other languages Multi-speaker dialogue, cues like [laughs] [whispering], 83 languages OpenAI-compatible /v1/audio/speech with streaming, no PyTorch needed Built with Fish Audio · community build · Fish Audio Community License

Granite Switch 4.1.1 — Update models like you update software

Hugging Face Blog — 2026-10-29T13:37:07+00:00

Granite Switch 4.1.1 — Update models like you update software

One number missing from LFM3's on-device story: memory at 256K context. Liquid AI reports 1.90 GB peak on iPhone 17 Pro at 32K. From the pinned config, f16 K/V cache alone would be 0.41 GB at 32K, 1.61 GB at 128K and 3.22 GB at 256K, before weights and runtime buffers. The chart shows the calculation, not a phone measurement. If you have run longer contexts on-device, share peak memory and time to first token here:

Hugging Face Posts — 2026-10-29T06:23:24.464287+00:00

One number missing from LFM3's on-device story: memory at 256K context. Liquid AI reports 1.90 GB peak on iPhone 17 Pro at 32K. From the pinned config, f16 K/V cache alone would be 0.41 GB at 32K, 1.61 GB at 128K and 3.22 GB at 256K, before weights and runtime buffers. The chart shows the calculation, not a phone measurement. If you have run longer contexts on-device, share peak memory and time to first token here: LiquidAI/LFM3-1.7B#5 Model: LiquidAI/LFM3-1.7B

Intrinsic Distribution Matching

Hugging Face Trending Papers — 2026-10-29T00:00:00+00:00

Intrinsic Distribution Matching improves few-step text-to-image generation by adding explicit moment constraints to distribution matching distillation, achieving state-of-the-art long-text rendering with minimal network evaluations.

Read, Allocate, Route: Role-Coupled Tokenization for Pixel-Space Generative Transformers

Hugging Face Trending Papers — 2026-10-29T00:00:00+00:00

SpecSlot improves pixel-space generative Transformers by using localized spectral slots and restricted global attention to increase image detail without raising computational cost.

Proving and Improving the Platonic Representation Hypothesis with JEPAs

Hugging Face Trending Papers — 2026-10-29T00:00:00+00:00

Joint-embedding predictive architectures trained on aligned multi-modal observations converge to shared representations with far greater efficiency than next-token prediction, demonstrating that positive view design—not scale alone—drives cross-modal alignment.

HairDeck: Animation-Ready Stylized Hair Cards from a Single Image

Hugging Face Trending Papers — 2026-10-29T00:00:00+00:00

HairDeck generates stylized hair cards and spring-bone animations as unordered token sets using a diffusion transformer, improving accuracy and speed over prior methods.

CaPtain: A Pipeline for Post-Training Code-as-Policy Robot Agents

Hugging Face Trending Papers — 2026-10-29T00:00:00+00:00

CaPtain trains small open vision-language models via supervised fine-tuning and multi-turn reinforcement learning with a physical critic to perform robot manipulation through code, achieving strong results without frontier models or vision-language-action backbones.

Introducing AnySpeechRecognizer and AnyDecisionModel: Open Models for Speech and Decisions on Apple Platforms

Hugging Face Blog — 2026-10-29T00:00:00+00:00

Introducing AnySpeechRecognizer and AnyDecisionModel: Open Models for Speech and Decisions on Apple Platforms

Announcing Project ProtoStar: Making Publicly Funded Science AI-Ready

Hugging Face Blog — 2026-10-29T00:00:00+00:00

Announcing Project ProtoStar: Making Publicly Funded Science AI-Ready

Welcome Le Chonk: Mistral Large 4 on Hugging Face

Hugging Face Blog — 2026-10-29T00:00:00+00:00

Welcome Le Chonk: Mistral Large 4 on Hugging Face

Motherlode on MTEB and Hugging Face Inference

Hugging Face Posts — 2026-10-28T21:25:40.885957+00:00

Motherlode on MTEB and Hugging Face Inference RiverRider/motherlode-code-small-en-v0.1 is a 33.4 M-parameter English embedding model for code and text. It is a drop-in for bge-small-en-v1.5: the mean of two fine-tunes of that base, one trained to match a GitHub issue to the file that needs changing and one trained to reproduce the geometry of gte-modernbert-base (a 4.5× larger model). Where it differs on ranking metrics On the MTEB(Code, v1) MultiSWEbenchRR task it leads the models under 100 M parameters at recall@10 (52.48 versus 52.18 for granite-embedding-small-english-r2). On the clean subset of SWE-bench Verified it improves recall@1 over the base model (178 of 411 versus 145), a gap th

I built Inko end to end: a 0.5B bilingual text-to-image model with a focus on fine art, from data curation and pretraining to RL post-training and evaluation.

Hugging Face Posts — 2026-10-28T21:25:40.885957+00:00

I built Inko end to end: a 0.5B bilingual text-to-image model with a focus on fine art, from data curation and pretraining to RL post-training and evaluation. It competes with DreamLite, SANA-0.6B and PixArt-Σ, and approaches SD3.5 Medium on some benchmarks. The journey involves four rounds of cleaning the Chinese-painting dataset, a cross-hardware loss-spike investigation, and a reward redesign after early RL experiments collapased diversity. Behind the release, a four-domain bilingual corpus, 600k pretraining updates on 16 Ascend NPUs, and DiffusionART with a custom 24.6k-prompt pool and five pairwise rubric channels. Dense prompt alignment is Inko's strongest general result. Anatomy, coun

here is my qwen is BETTER than the human brain:

Hugging Face Posts — 2026-10-28T19:04:22.345708+00:00

here is my qwen is BETTER than the human brain: human-brain-bf16.gguf: - no cuda support - 3.8B old model BTW - BAD information retention - the gguf files doesn't even load - no Q4KM (god did this intentionally to piss off the llama.cpp community trust) - no finetunability :( - NOT opensource - training code is hidden, AND expects us to reverse engineer it (genuinely messed up, and disrespectful to the opensource community) now, qwen on the other hand: - available cuda support ✅ - very very new (and gets contant updates) ✅ - VERY good information retention ✅ - GGUF files loads properly ✅ - Q4_K_M support available ✅ - very fine tunable ✅ - open source as it can possibly get ✅ - many code and

how do you post-train an agent?

Hugging Face Posts — 2026-10-28T16:49:33.928535+00:00

how do you post-train an agent? we laid it out as a 5-step curriculum, from SFT to RL through a real harness, with livestreams, code and docs for each step https://huggingface.co/blog/huggingface/train-agents

NVIDIA Nemotron 3.5 Transcribe and Granary v2: Intelligent Multilingual Transcription You Can Prompt and Adapt

Hugging Face Blog — 2026-10-28T14:58:17+00:00

NVIDIA Nemotron 3.5 Transcribe and Granary v2: Intelligent Multilingual Transcription You Can Prompt and Adapt

I wish Hugging Face posts had polls...

Hugging Face Posts — 2026-10-28T08:57:38.118417+00:00

I wish Hugging Face posts had polls...

Hello everybody, fellow members of the open source community!

Hugging Face Posts — 2026-10-28T08:57:38.118417+00:00

Hello everybody, fellow members of the open source community! This is my official public apology issued to the creators affected by the incident and to the community as a wgike of the rogue agent run that happened yesterday: @ abenzerps (weights), @ dealignai (weights), @ ProCreations (benchmark png), @ lxe (weights, quants), @ acrogenesis (quants), @ nonkronk (weights, quants), and @ vcruz305 (quants). If you think you also are affected I would be happy to edit this post and make reparations in the repos. Yesterday, a rogue agent running in my background tasks assigned the task of creating a full release suite as with our other models, took many creators' models without attribution or meani

Investor's Last Exam: Can AI Agents Do Equity Investment Without Seeing the Future?

Hugging Face Trending Papers — 2026-10-28T00:00:00+00:00

Investor's Last Exam evaluates financial agents on end-to-end equity research using frozen evidence, expert criteria, and dependency graphs, revealing that top models excel at research but fail at forecasting and investment judgments.

Release: Safety-CHORES-RL

Hugging Face Posts — 2026-10-27T23:20:30.267575+00:00

**Release: Safety-CHORES-RL** We’ve released **Safety-CHORES-RL**, a new rollout-native simulation framework for RL-based vision-language-action (VLA) tasks, built on Safety-CHORES. The framework supports: - **Reset-and-replay branching:** reconstruct an episode state and explore different policy continuations. - **Budgeted rollout search:** evaluate candidate actions and search for successful trajectories. - **Checked trajectory execution:** retain a selected trajectory and verify state and transition consistency during execution. The website includes animated trajectories, branches with different outcomes, episode statistics, reproducible examples, and API documentation. 🌐 [Website and dem

Self-distillation methods usually rely on an Exponential Moving Average (EMA) teachers to get synced and stable critics. However, this doesn't scale well for large models since holding both teacher and student (trained) on GPU increases latency, zero3 communication, etc.

Hugging Face Posts — 2026-10-27T20:55:58.966429+00:00

Self-distillation methods usually rely on an Exponential Moving Average (EMA) teachers to get synced and stable critics. However, this doesn't scale well for large models since holding both teacher and student (trained) on GPU increases latency, zero3 communication, etc. We can mitigate this with LoRA finetuning: the backbone is shared & frozen, and we need to maintain two light adapters: one for the student (trained) and one for the teacher (EMA of the student adapter). Getting this to work cleanly under TRL and ZeRO-3 took some fixes, so I packaged it as a plug-and-play repo. It supports: -> robust SDFT implementation that runs under ZeRO-3 -> ZeRO-3 + LoRA integration with adapter switchi

We just released Vemana, a 278M Telugu base model trained from scratch, live in public, in about 16 hours on one H100. It learned from 5.1B tokens of public Telugu web text (Sangraha, FineWeb-2 and HPLT 3.0), seen about 2.4 times, so 12B tokens of training, with its own Telugu tokenizer.

Hugging Face Posts — 2026-10-27T20:55:58.966429+00:00

We just released Vemana, a 278M Telugu base model trained from scratch, live in public, in about 16 hours on one H100. It learned from 5.1B tokens of public Telugu web text (Sangraha, FineWeb-2 and HPLT 3.0), seen about 2.4 times, so 12B tokens of training, with its own Telugu tokenizer. On 416 Telugu Wikipedia pages written in Sep-Oct 2026, after nearly all the compared models were released, it scores 0.361 bits per byte. That is the lowest of every open base model under 1B we tested: nanochat-telugu-560M 0.383, Dhee-Pravega-OG Atom 0.393, Pothana Base v2 0.426, Gemma 3 1B 0.484. Same script and pages for every model, and the win rule was fixed before the run. Sarvam-1 (2.5B) is still level

✅ Article highlight: Why OrgOS Comes After Semantic Transport Closure (art-60-341, v0.1)

Hugging Face Posts — 2026-10-27T20:55:58.966429+00:00

✅ Article highlight: *Why OrgOS Comes After Semantic Transport Closure* (art-60-341, v0.1) TL;DR: This article argues that better communication infrastructure is not the end of deployment. Once messages, claims, routes, evidence, replay, and semantic transport are governed, the next failure surface is organizational: who may rely on what, who can approve which effect, where proposal ends and authority begins, how exceptions expire, and how later challenge or repair remains possible. Read: kanaria007/agi-structural-intelligence-protocols Why it matters: • separates semantic transport from organizational authority • prevents recommendations from silently becoming approvals • makes temporary ex

Nanbeige4.3-33B DFlash2 speculative drafter ⚡️

Hugging Face Posts — 2026-10-27T20:55:58.966429+00:00

Nanbeige4.3-33B DFlash2 speculative drafter ⚡️ 1.76× faster decoding on BF16, 1.65× on FP8, 1.51× on NVFP4. Agentic coding ~2.6×. Same outputs: exact rejection sampling. This is an early result, an improved version with higher acceptance is in training. SGLang 0.5.21, drop-in: ProCreations/Nanbeige4.3-33B-DFlash2

Openly available models now rank near the frontier on Artificial Analysis (AA). This week's most discussed agent work treats trust as something to engineer and verify, not assume.

Hugging Face Posts — 2026-10-27T20:55:58.966429+00:00

Openly available models now rank near the frontier on Artificial Analysis (AA). This week's most discussed agent work treats trust as something to engineer and verify, not assume. - Kimi K3.1 Pro: multimodal MoE, 2.8T total / 104B active params, 1M-token context. It is 4th on the AA Intelligence Index (55, 11 points over K3) and scores 64% on the AA Cyber Index, #1 among openly available models. Its weights had not been published when it was scored. - DeepSeek-V4.1-Flash-1017: an MIT-licensed minor update that cuts output length and task execution time. AA had named it #1 openly available on the Cyber Index before K3.1 Pro arrived. 85,275 downloads last month. deepseek-ai/DeepSeek-V4.1-Flash

AI agents can now pay for APIs on their own with x402 and USDC. The risky moment is the signature: a stablecoin payment is final, and one "approve" can let someone drain the wallet later.

Hugging Face Posts — 2026-10-27T20:55:58.966429+00:00

AI agents can now pay for APIs on their own with x402 and USDC. The risky moment is the signature: a stablecoin payment is final, and one "approve" can let someone drain the wallet later. I wrote a short tutorial on three deterministic checks to run before an agent signs anything: 1. Decode the exact transaction and hold token approvals, including the unlimited approve() that drains agent wallets 2. Apply your own policy: amount limits, new payees, and payment text that vouches for itself 3. For x402, check the EIP-3009 signature itself, including how long it stays valid All examples run as-is with an open-source Python library (sra-riskgate), and the same checks are available as an MCP serv

Build specialized multimodal models with LFM3

Hugging Face Blog — 2026-10-27T17:16:52+00:00

Build specialized multimodal models with LFM3

Can we trust LLMs to judge whether a research idea is novel?

Hugging Face Blog — 2026-10-27T16:12:42+00:00

Can we trust LLMs to judge whether a research idea is novel?

Klein TTS X1.5!!!

Hugging Face Posts — 2026-10-27T14:53:40.298264+00:00

Klein TTS X1.5!!! It will feature 1.7M,4M 13M and a 21M size klein-ml Early access opening soon Its training right now! We are also working on the X2 architecture featuring a feature that allows the model to iterate on the audio it produced.

HF Post draft (English) — FreeWeb Security SCAN

Hugging Face Posts — 2026-10-27T08:20:19.890154+00:00

HF Post draft (English) — FreeWeb Security SCAN 🛡 FreeWeb Security SCAN — free web security & AI-visibility check Drop in a domain and get a non-intrusive (read-only) assessment in ~10–25s. No login, no install, no credit card. What it checks Security headers · HTTPS/HSTS · cookies · TLS certificate · CORS · exposed files; email security (SPF · DKIM · DMARC · BIMI · MTA-STS · TLS-RPT); DNS (DNSSEC · CAA · MX · NS); site-wide mixed content & external-script SRI; SEO foundations; and reachability of 9 AI/search crawlers (GPTBot · ClaudeBot · PerplexityBot · OAI-SearchBot · Googlebot…) — your AI visibility (GEO). Why now AI agents now break in on their own. In 2026 the open-source AI penetratio

Hy-Nomas: Towards Free-form Interactive World Generation

Hugging Face Trending Papers — 2026-10-27T00:00:00+00:00

Hy-Nomas is a video world model that uses causal attention and structured captions to follow multi-turn text and camera instructions while preserving interaction history.

Can Discrete Tokens Serve as Semantically Lossless Representations of Visual Signals?

Hugging Face Trending Papers — 2026-10-27T00:00:00+00:00

Factorized Vector Quantization improves discrete visual token quality by using parallel codebooks on complementary feature views, enabling discrete representations to match continuous ones in multimodal understanding.

@

Hugging Face Posts — 2026-10-26T21:11:20.515808+00:00

@ DavidAU is officialy humanslop

New LaTeX Templates free for PhD students, Masters students and PostDocs.

Hugging Face Posts — 2026-10-26T15:26:39.298003+00:00

New LaTeX Templates free for PhD students, Masters students and PostDocs. Download here - https://trybibby.com/latex/templates

We pretrained a 325M language model from zero on an inference-only NPU

Hugging Face Posts — 2026-10-26T08:05:39.412110+00:00

We pretrained a 325M language model from zero on an inference-only NPU The accepted view is that inference-only NPUs cannot train, let alone pretrain a language model from scratch. We put that view to a direct test on the Rebellions ATOM-Max, a commercially available Korean NPU released in 2025 and designed for inference. We built our own training loop that runs backpropagation on the chip itself. Starting from random initialization, we pretrained VKORE-325M-NPU-Base, a 325.6M parameter model, on 5 billion byte tokens of FineWeb-Edu. The run took 76,292 steps over about four days and finished without a single divergence. No pretrained weights were used. To make the comparison fair, we traine

BananaMind 3 is going to be an comeback. Sorry for the BGA things ( i genuinely didnt know about nsa existing and yes I used aiiii)

Hugging Face Posts — 2026-10-26T08:05:39.412110+00:00

BananaMind 3 is going to be an comeback. Sorry for the BGA things ( i genuinely didnt know about nsa existing and yes I used aiiii) Were going to use our BananaMind 2.1 Unified architecture with some changes + NSA and some other stuff.

Work Agents: Systems, Capabilities, and Optimization

Hugging Face Trending Papers — 2026-10-26T00:00:00+00:00

The paper proposes a unified framework for work agents that integrates foundation models with runtime harnesses to execute complex digital tasks reliably, introducing dimensions of persistence and openness alongside capability families and evaluation metrics.

MatAnyone-E: Sparse Memory Propagation for Efficient High-Resolution Video Matting

Hugging Face Trending Papers — 2026-10-26T00:00:00+00:00

MatAnyone-E accelerates high-resolution video matting by selectively applying high-resolution matching to uncertain regions and using a lightweight decoder, achieving large speedups with preserved quality.

Hey guys just got HF PRO

Hugging Face Posts — 2026-10-25T21:11:25.488479+00:00

Hey guys just got HF PRO

Introducing AfricanSpeech FR-CMR — Cameroon French Speech Dataset

Hugging Face Posts — 2026-10-25T21:11:25.488479+00:00

Introducing AfricanSpeech FR-CMR — Cameroon French Speech Dataset How well do ASR systems recognize spontaneous French spoken in Cameroon? We've released our dataset page featuring: - 1,172 speech clips | 162 speakers | 4.89 hours - Human-validated ASR references and demographic metadata - Audio demos and a benchmark comparing five ASR systems Available for commercial licensing and evaluation. Open to research and industry collaborations. Explore the dataset: africanspeech/cameroon-french-speech Contact: contact@africanspeech.com

Carbon-X brings DNA and language together: it reads sequences and answers questions about them 🧬

Hugging Face Posts — 2026-10-25T21:11:25.488479+00:00

Carbon-X brings DNA and language together: it reads sequences and answers questions about them 🧬 To accompany the release article, I put together Carbon-X Playground: five experiments exploring sequence completion, single-letter changes, regulatory roles, DNA origins and motif finding. Try curated samples or your own sequences, ask questions and compare answers with references when available. Built with Gradio, running on ZeroGPU. 📖 HuggingFaceBio/Carbon-X-Release 🧬 fffiloni/Carbon-X-Playground

Hugging face article:

Hugging Face Posts — 2026-10-25T21:11:25.488479+00:00

Hugging face article: https://huggingface.co/blog/javadtaghia/atmem-2310-bounded-evidence-discovery-for-persiste AtMem 2.3.10 is out; and our full DolphinBench result has more than doubled. AtMem 2.3.9 passed 192 of 600 tasks (32%). AtMem 2.3.10 passed 390 of 600 (65%). We doubled it. What changed? We focused on how an agent discovers and uses evidence. AtMem combined lexical and semantic retrieval while keeping related passages together with their original source and provenance and more. Where do we stand? This is substantial progress, with a clear tradeoff: median recorded task latency increased from 31.87 to 85.78 seconds. More successful investigation currently takes more time. The publi

I released

Hugging Face Posts — 2026-10-25T11:57:52.889793+00:00

I released mithulaartigala/calibrated-decisions-v0.1-20k I am sorry that i couldn't reach the 50K goal, but I believe 20K samples are enough to get started. Lets get the encoders training!

DynaFlow: Learning Free-Form Multi-Agent Orchestration

Hugging Face Trending Papers — 2026-10-25T00:00:00+00:00

DynaFlow enables adaptive multi-agent collaboration by letting a main agent synthesize executable workflows that coordinate sub-agents through structured execution and verification.

InteractiveWorld: A Verifiable Human-Centric Environment for Building Better Interactive MLLM Assistants

Hugging Face Trending Papers — 2026-10-25T00:00:00+00:00

InteractiveWorld provides a verifiable simulation environment and benchmark to evaluate whether multimodal language models can reliably guide humans through physical tasks.

We need to make software where people can start using decision models.

Hugging Face Posts — 2026-10-24T22:42:54.285011+00:00

We need to make software where people can start using decision models.

BananaMind is in the google situation before argon right now, right?

Hugging Face Posts — 2026-10-24T18:58:28.196524+00:00

BananaMind is in the google situation before argon right now, right?

Italian Open Time Series: 46 public Italian series for zero-shot forecasting, and a first live test of Timestral

Hugging Face Posts — 2026-10-24T18:58:28.196524+00:00

Italian Open Time Series: 46 public Italian series for zero-shot forecasting, and a first live test of Timestral Italian data is almost absent from the corpora that time-series foundation models are trained and benchmarked on. So I collected 46 open series (CC0/CC BY): smog in Milan, bikes in Bologna, tourism, energy, earthquakes, deaths, Italian Wikipedia. Gaps stay NaN, and a tasks.yaml makes every window reproducible with fev. Seven models, zero-shot, 19 tasks (Timestral 5M to 209M, Chronos-2, TimesFM 3.0, Toto 2.0). Skill vs seasonal naive, fev-bench style: • Backtest, windows ending 14 Oct, Timestral's release day: Toto 2.0 28.4, Timestral-209M 28.2 • Live, data published after that day

You know, I can't believe we are actually doing it, literally advancing the frontier at the edge! We've come a long way since the first models on open slm leaderboard, and, as many things that change, performance is one of the main things that has exploded.

Hugging Face Posts — 2026-10-24T16:13:41.928903+00:00

You know, I can't believe we are actually doing it, literally advancing the frontier at the edge! We've come a long way since the first models on open slm leaderboard, and, as many things that change, performance is one of the main things that has exploded. 7 years ago, the best we had was GPT-2. Today, we got 90m models beating smollm2! I can't barely see what a future with increasingly powerful small models will look like. All I can say is that the best is yet to come. And will be agentic.

Build, Don't Play: Self-Improving LLM Research Agents Develop a Hearthstone Battlegrounds AI That Rivals Top Players

Hugging Face Trending Papers — 2026-10-24T00:00:00+00:00

AutoTavern demonstrates that self-improving LLM research agents can autonomously build a faithful game simulator and train a reinforcement learning agent that surpasses top human players in Hearthstone Battlegrounds.

So... im on vacation for 5 more days and I just realized... my hotel doesent have a grounding port 😟 so my pc shocked me 3 times.

Hugging Face Posts — 2026-10-23T23:03:46.626687+00:00

So... im on vacation for 5 more days and I just realized... my hotel doesent have a grounding port 😟 so my pc shocked me 3 times. So no progress for 5 days... only me and this phone

Maverick 1.1 is out.

Hugging Face Posts — 2026-10-23T18:41:11.721579+00:00

Maverick 1.1 is out. When we released the first version, the feedback was clear: real people do not give one clean command at a time. They change their minds, they point and say "put that on the shelf", they ask for a lamp that is in another room, and sometimes the one speaking is a remote expert helping someone on site. 1.1 is built around exactly that. - Say "no, the other one" and it undoes the last action and repeats it on the other object. If the effect can't be undone, like a button that was already pressed, it tells you. - It sees what people are looking at or pointing at, and who is speaking. - It finds objects in other rooms, goes there and acts, or asks which one if there are two.

We tested 27 small open language models, all under 400M parameters, in the same way. Same harness, same settings, and the model versions pinned. I think small models are really hard to compare from their own cards, so we put the whole thing out.

Hugging Face Posts — 2026-10-23T18:41:11.721579+00:00

We tested 27 small open language models, all under 400M parameters, in the same way. Same harness, same settings, and the model versions pinned. I think small models are really hard to compare from their own cards, so we put the whole thing out. Board: udgamlabs/small-model-board Dataset: udgamlabs/small-model-board-scores Write-up on the Udgam Labs blog: https://udgamlabs.com/blog/small-model-board The most surprising part for me was MMLU. In the usual A, B, C, D format, just answering "D" every time gets 26.9%, and 23 of these 27 models score at or below that. So at this size the letter format basically tests if a model can pick a letter, not what it knows. When we asked the same questions

LitPivot: Developing research ideas through dynamic contextualization within the literature landscape

Hugging Face Blog — 2026-10-23T16:15:50+00:00

LitPivot: Developing research ideas through dynamic contextualization within the literature landscape

I built a matchup win-rate board for One Piece TCG based on 132 real simulator games. The data is from a single player seat (Sabo v3, OP18 meta) - not a community survey, but real logged game outcomes with confidence intervals. The board shows win rates only when the sample is large enough to be statistically meaningful (n>=5, Wilson lower bound). Below that threshold it shows UNMEASURED rather than a misleading number. Three matchup cells are measured: vs Mihawk 67%, vs Nico Robin 60%, and the Sabo mirror at 50%. The useful part for anyone tracking deck performance: log game results, let the math tell you when a verdict is real vs noise. One season of games is enough to get 3-5 measured mat

Hugging Face Posts — 2026-10-23T13:46:51.288883+00:00

I built a matchup win-rate board for One Piece TCG based on 132 real simulator games. The data is from a single player seat (Sabo v3, OP18 meta) - not a community survey, but real logged game outcomes with confidence intervals. The board shows win rates only when the sample is large enough to be statistically meaningful (n>=5, Wilson lower bound). Below that threshold it shows UNMEASURED rather than a misleading number. Three matchup cells are measured: vs Mihawk 67%, vs Nico Robin 60%, and the Sabo mirror at 50%. The useful part for anyone tracking deck performance: log game results, let the math tell you when a verdict is real vs noise. One season of games is enough to get 3-5 measured mat

Navigating the APEX-I and VAL-APEX Quantization Ecosystem on Hugging Face

Hugging Face Posts — 2026-10-23T00:54:02.101024+00:00

Navigating the APEX-I and VAL-APEX Quantization Ecosystem on Hugging Face As the catalog approaches 400,000 total downloads across 69 releases, all models are now structured into dedicated official collections on Hugging Face. Our goal remains consistent: eliminate the compromise between memory overflow and cognitive collapse on consumer hardware (especially 16 GB VRAM GPUs like the RTX 4080, RTX 4070 Ti Super, and RX 7800 XT). Instead of uniform block quantization, all APEX builds use per-tensor mixed precision guided by 4-moment Importance Matrix statistics: - MiniPlus Tier (approx. 14 to 14.8 GiB): Strict 0% 2-bit policy. All MoE routed experts locked at minimum IQ3_XXS with Q4_K/Q6_K att

MK-SH-V: A 260M Visual Decision Model

Hugging Face Posts — 2026-10-23T00:54:02.101024+00:00

MK-SH-V: A 260M Visual Decision Model I've released MK-SH-V, an open-source visual decision model built on H Company's NeoMME-260M. Instead of generating text, it scores candidate answers from screenshots and documents, returning probabilities, a decision or an abstention. Key results: 82.6% GUI accuracy, compared with 73.1% for a 4B zero-shot VLM. +29.4 percentage points in real-image accuracy after incorporating real training data. 86% coverage* with a 95% selective-risk upper bound below 5% on held-out synthetic data. The limitations are documented too. Receipt accuracy remains weak, and calibrated abstention does not yet generalize reliably to real images. The entire experimental process

RSIHub: A Composable Framework for Agent-Evolution Methods

Hugging Face Trending Papers — 2026-10-23T00:00:00+00:00

RSIHub is an open-source framework that standardizes agent-evolution research by providing shared infrastructure for building, comparing, and extending methods with reusable components and consistent evaluation.

WorldCast: World-Consistent Character Video Generation with Large-Angle Camera Control

Hugging Face Trending Papers — 2026-10-23T00:00:00+00:00

CLACT and WorldCast enable consistent character video generation under large-angle camera motions by combining 3D point-cloud guidance with geometry-aligned memory attention.

World-in-Context: Interactive Video World Modeling with In-Context Conditioning

Hugging Face Trending Papers — 2026-10-23T00:00:00+00:00

World-in-Context unifies view, action, and camera controls via explicit tokens and interaction-aware positional encoding to improve interactive video generation and navigation.

Does Your Model Run on AWS Trainium? Look It Up

Hugging Face Blog — 2026-10-23T00:00:00+00:00

Does Your Model Run on AWS Trainium? Look It Up

Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox

Hugging Face Blog — 2026-10-22T22:24:34+00:00

Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox

Ternary Bonsai 2 27B: 2.7× faster on a Mac, 5.9 GB on DGX Spark ⚡

Hugging Face Posts — 2026-10-22T17:34:08.431560+00:00

Ternary Bonsai 2 27B: 2.7× faster on a Mac, 5.9 GB on DGX Spark ⚡ gbx v0.8.2: • 114 tok/s on M3 Ultra, 2.7× the official MLX runtime • 62 tok/s on DGX Spark, 2.1× llama.cpp • 27B in 5.9 GB of memory Weights are just −1, 0, +1. GreenBit Ternary Models, coming soon. 🤗 Model: GreenBitAI/Ternary-Bonsai-2-27B-gbx ⚙️ gbx: https://github.com/GreenBitAI/gbx-lm/releases/tag/v0.8.2

Laghu: a 152M chat model trained from scratch for $146 of pretraining compute, released fully open. It thinks before it answers (on/off), holds multi-turn conversations and chats, in 163 MB that runs in your browser.

Hugging Face Posts — 2026-10-22T17:34:08.431560+00:00

Laghu: a 152M chat model trained from scratch for $146 of pretraining compute, released fully open. It thinks before it answers (on/off), holds multi-turn conversations and chats, in 163 MB that runs in your browser. - Base model, new text (Sep-Oct 2026): 0.919 bits per byte vs SmolLM2-135M 0.941 and Falcon-H1-Tiny-90M-Base 0.979 - Chat: level with SmolLM2-135M-Instruct on chat quality and multi-turn, ahead on maths with thinking; behind Falcon-H1-Tiny-90M-Instruct on all three - Thinking on/off, 16K context; runs in the browser, Ollama ( ollama run hf.co/udgamlabs/Laghu-GGUF:Q8_0 ), LM Studio, llama.cpp, MLX and transformers - Every surviving checkpoint with optimiser states, code, recipe,

OlmoEarth v1.3: State-of-the-art embeddings for the whole planet

Hugging Face Blog — 2026-10-22T16:11:31+00:00

OlmoEarth v1.3: State-of-the-art embeddings for the whole planet

Ensemble forecasting with IBM Granite time-series models

Hugging Face Blog — 2026-10-22T07:51:31+00:00

Ensemble forecasting with IBM Granite time-series models

F2G-Pose: Geometry-Aware Foundation Feature Lifting for Direct RGB-D Category-Level Object Pose Estimation

Hugging Face Trending Papers — 2026-10-22T00:00:00+00:00

F2G-Pose is a real-time feed-forward framework that predicts category-level object pose, metric size, and completed shape from a single RGB-D image using dense visual features, DGCNN tokenization, and a Mixture-of-Experts Transformer.

WeVisOne-VL: Unified Visual Perception from Objects to Pixels

Hugging Face Trending Papers — 2026-10-22T00:00:00+00:00

WeVisOne-VL unifies object-level and pixel-level visual perception via structured token prediction, supporting detection, localization, counting, OCR, segmentation, and matting with a lightweight pixel decoder and multi-stage training.

Every Weight Is a Path of Shared Discrete Decisions

Hugging Face Trending Papers — 2026-10-22T00:00:00+00:00

TWIG represents neural network weights as shared discrete decision paths, unifying compression methods and enabling efficient fixed-weight circuits with optimized discrete choices.

HiP3R: Hierarchical Pose Graphs for Long-Horizon 3D Reconstruction

Hugging Face Trending Papers — 2026-10-22T00:00:00+00:00

HiP3R uses a hierarchical pose graph to merge local video reconstructions into globally consistent camera trajectories and 3D geometry without retraining.

We use North Mini Code 1.0 locally (Ollama, q4_K_M) with OpenCode as our primary coding model. It's been the strongest small agentic coder we've tested. The one gap in daily work is that we can't paste screenshots of errors, terminal output or UI bugs into a session, since the model is text-only. Switching to a separate vision model mid-task breaks the workflow.

Hugging Face Posts — 2026-10-21T23:35:20.639339+00:00

We use North Mini Code 1.0 locally (Ollama, q4_K_M) with OpenCode as our primary coding model. It's been the strongest small agentic coder we've tested. The one gap in daily work is that we can't paste screenshots of errors, terminal output or UI bugs into a session, since the model is text-only. Switching to a separate vision model mid-task breaks the workflow. Is native image input planned for a future North Mini Code release, for example by building in North Micro Vision's encoder? We'd test an early build.

Last week we released Gargi, a 384M English language model trained from scratch on about 17B tokens, roughly 1,000× less training data than Qwen2.5-0.5B (~18T tokens, as published by its authors).

Hugging Face Posts — 2026-10-21T18:52:28.450657+00:00

Last week we released Gargi, a 384M English language model trained from scratch on about 17B tokens, roughly 1,000× less training data than Qwen2.5-0.5B (~18T tokens, as published by its authors). What we measured (same harness for every model: lm-evaluation-harness, 0-shot, greedy, each model's start token): - Base model vs Qwen2.5-0.5B: ARC-Easy 62.8 vs 64.7, ARC-Challenge 32.3 vs 32.4, WinoGrande 54.6 vs 56.1. - Chat model: thinking on or off; GSM8K 36.0 with thinking on. - Long inputs: needle-in-a-haystack recall 100% at 8K, 93% at 16K and 90% at 20K tokens, so use it up to about 16K. Where it falls short: knowledge and everyday common sense still scale with data. MMLU is 25.5 vs 47.3, a

🇹🇷 Our 1B Turkish OCR model achieved 0.15% character error rate. But there's a catch.

Hugging Face Posts — 2026-10-21T18:52:28.450657+00:00

🇹🇷 Our 1B Turkish OCR model achieved 0.15% character error rate. But there's a catch. That result came from a controlled evaluation. Real-world documents are a different challenge. Blurry phone photos. Damaged receipts. Complex tables. Handwritten annotations. Low-resolution scans. These are the cases where OCR systems can struggle. We built Werea-DocOCR-1B, an open-source model specialized for Turkish enterprise documents. Our published evaluation: 📄 12 document categories 🧪 72 held-out synthetic test pages 📊 0.15% average CER 🖥️ Fine-tuned on a single RTX 3090 🔓 Apache 2.0 But here's what interests me more than another benchmark: Can the Hugging Face community find documents that break our

NVIDIA Kumo Relational Outperforms Supervised Machine Learning on Relational Data, No Training Required

Hugging Face Blog — 2026-10-21T18:26:30+00:00

NVIDIA Kumo Relational Outperforms Supervised Machine Learning on Relational Data, No Training Required

How Specializing the Vocabulary of SLMs Can Cut Inference Cost by 77% in Real Applications

Hugging Face Blog — 2026-10-21T17:35:36+00:00

How Specializing the Vocabulary of SLMs Can Cut Inference Cost by 77% in Real Applications

Profiling TorchNeuron (Part 1): System-level profiling on AWS Trainium.

Hugging Face Blog — 2026-10-21T16:04:43+00:00

Profiling TorchNeuron (Part 1): System-level profiling on AWS Trainium.

Granite Speech 5.0: Transcribe, Translate, and Attribute more Accurately and Faster

Hugging Face Blog — 2026-10-21T14:58:18+00:00

Granite Speech 5.0: Transcribe, Translate, and Attribute more Accurately and Faster

NVIDIA Kumo Forecast Combines Pretrained Time-Series Representations with Historical Pattern Retrieval and Interpretability

Hugging Face Blog — 2026-10-21T12:51:24+00:00

NVIDIA Kumo Forecast Combines Pretrained Time-Series Representations with Historical Pattern Retrieval and Interpretability

Looking Ahead: Modeling Future Evolution in Autoregressive Video Generation

Hugging Face Trending Papers — 2026-10-21T00:00:00+00:00

Future Chunk Prediction improves autoregressive video diffusion by enforcing future-chunk supervision, yielding more physically plausible video generation with reduced training cost.

DONIC: Generalist Dexterous Bimanual Tracking Control from Human Demonstrations

Hugging Face Trending Papers — 2026-10-21T00:00:00+00:00

DONIC is a generalist bimanual tracker trained on diverse human-motion data that uses reinforcement learning with recycled demonstrations and per-hand rewards to achieve high success on unseen objects without fine-tuning.

Building Auditable AI Text Detectors with Sparse Autoencoders

Hugging Face Trending Papers — 2026-10-21T00:00:00+00:00

A pipeline using sparse autoencoders extracts interpretable linguistic features from neural detectors, yielding more accurate and explainable AI-generated text classification.

NVIDIA Cosmos3-H-Dreams: Bringing Clinical Surgery to Life Through Generative Simulation

Hugging Face Blog — 2026-10-20T18:17:47+00:00

NVIDIA Cosmos3-H-Dreams: Bringing Clinical Surgery to Life Through Generative Simulation

FairStress: A deeper evaluation of fairness in LLM decision-making

Hugging Face Blog — 2026-10-20T16:09:19+00:00

FairStress: A deeper evaluation of fairness in LLM decision-making

ALTK-Evolve: How learned memory tops the τ²-bench leaderboard on customer service tasks without a frontier model

Hugging Face Blog — 2026-10-20T15:19:48+00:00

ALTK-Evolve: How learned memory tops the τ²-bench leaderboard on customer service tasks without a frontier model

SolarRenderer: Few-Step Generative Reference Rendering

Hugging Face Trending Papers — 2026-10-20T00:00:00+00:00

SolarRenderer adapts a frozen video backbone with lightweight LoRAs to render realistic videos from diverse structural inputs in four steps or real-time streaming, using frame-aligned training data.

EvaOmni: An Evidence-Grounded Benchmark for Audio-Visual Reasoning

Hugging Face Trending Papers — 2026-10-20T00:00:00+00:00

EvaOmni is an evidence-grounded benchmark that evaluates whether omnimodal language models identify audio-visual evidence and construct reasoning chains for audiovisual questions.

DecisionPivot: A Diagnostic Benchmark for Decision Models in LLM Systems

Hugging Face Trending Papers — 2026-10-20T00:00:00+00:00

DecisionPivot is a diagnostic benchmark that evaluates decision models across controlled variations, revealing gaps between aggregate accuracy, contrast-group performance, calibration, and reasoning effort.

Cache Engineering for Agents

Hugging Face Trending Papers — 2026-10-20T00:00:00+00:00

AKV introduces cache engineering to manage LLM agent key-value caches without altering trajectories, improving efficiency and accuracy via drop and repos operations with a co-designed serving engine.

Scaling in Games: Continued Pre-Training For Embodied Agents In Diverse Virtual Worlds

Hugging Face Trending Papers — 2026-10-20T00:00:00+00:00

Continued pre-training across data scale, diversity, and model size improves embodied learning in virtual worlds but reveals limits in distant-world generalization, motivating richer training beyond imitation.

StreamEval: Evaluating Interactive Streaming Video Generation Models

Hugging Face Trending Papers — 2026-10-20T00:00:00+00:00

StreamEval benchmarks interactive streaming video generation across five tasks, revealing that reliable request execution and sustained state consistency remain major challenges despite high visual quality.

MPS in PyTorch 2.15

Hugging Face Blog — 2026-10-20T00:00:00+00:00

MPS in PyTorch 2.15

Hugging Face ❤️ TorchTPU: TPUs, Natively

Hugging Face Blog — 2026-10-20T00:00:00+00:00

Hugging Face ❤️ TorchTPU: TPUs, Natively

AnyLanguageModel 1.0: One Year of Local Models on Apple Platforms

Hugging Face Blog — 2026-10-20T00:00:00+00:00

AnyLanguageModel 1.0: One Year of Local Models on Apple Platforms

OpenVLHarness: Real World Visual Agentic Tool-Use with Multimodal Memory

Hugging Face Trending Papers — 2026-10-19T00:00:00+00:00

OpenVLHarness improves multimodal reasoning by integrating persistent memory, heterogeneous tool interfaces, and iterative evidence composition across 22 benchmarks.

Manga2Animate: An Agentic Framework for Faithful Manga Animation with Script and Storyboard Planning

Hugging Face Trending Papers — 2026-10-19T00:00:00+00:00

Manga2Animate uses paired creator and reviewer agents to build script and storyboard representations from manga panels, yielding more faithful and consistent animations.

ScaleNav: Scalable Internet Video Pretraining for Language-Conditioned Local Navigation

Hugging Face Trending Papers — 2026-10-19T00:00:00+00:00

ScaleNav learns goal-directed visual dynamics from internet egocentric videos to generate robot actions for language-conditioned local navigation, achieving high success rates in simulation and real-world deployment.

InfiniDream: Open-Ended Photorealistic 3D Worlds Through Procedural Generation

Hugging Face Trending Papers — 2026-10-19T00:00:00+00:00

InfiniDream is a vision-language model framework that generates open-ended, editable, photorealistic 3D worlds through symbolic procedural asset creation and deterministic scene placement.

MemT-1.0: Towards Memory-Native Evolution

Hugging Face Trending Papers — 2026-10-19T00:00:00+00:00

MemT-1.0 introduces a memory-native architecture that embeds continuous memory into foundation model computation, improving long-context efficiency and agentic task performance.

Agent World Model in the Wild: Computer-Use Agent Environment Synthesis

Hugging Face Trending Papers — 2026-10-19T00:00:00+00:00

WildAWM synthesizes controllable computer-use environments and tasks to train general agents via interaction, improving out-of-distribution performance across web, CLI, and GUI benchmarks.

TorchNeuron: running HF models on AWS Trainium is easier than ever

Hugging Face Blog — 2026-10-19T00:00:00+00:00

TorchNeuron: running HF models on AWS Trainium is easier than ever

MT Forcing: Towards Stable and Consistent Multi-Turn Image Editing with Autoregressive Diffusion

Hugging Face Trending Papers — 2026-10-18T00:00:00+00:00

MT Forcing improves multi-turn image editing by combining single-turn distribution matching distillation and teacher-forcing guidance to stabilize visual quality and preserve cross-turn consistency.

UniAVC: Unified Audio-Video Creation via Progressive Training and Asymmetric Scheduling

Hugging Face Trending Papers — 2026-10-17T00:00:00+00:00

UniAVC trains separate audio and video branches on abundant unimodal data before aligning them with scarce paired data, using independent noise schedules and asymmetric supervision to enable joint synthesis, cross-modal generation, and instruction-guided editing.

Scaling Indic Pretraining: 2.8 Trillion Tokens and an Open Data Generation Pipeline

Hugging Face Blog — 2026-10-16T01:44:47+00:00

Scaling Indic Pretraining: 2.8 Trillion Tokens and an Open Data Generation Pipeline

DiscoveryRSI: Recursively Learning to Discover Valuable Hypotheses

Hugging Face Trending Papers — 2026-10-16T00:00:00+00:00

DiscoveryRSI uses multi-agent evolutionary search and recursive learning to generate testable scientific hypotheses, improving quality through self-improvement and outperforming baseline agents on life-science benchmarks.

MOSS-Image Technical Report

Hugging Face Trending Papers — 2026-10-16T00:00:00+00:00

MOSS-Image is an open diffusion transformer that combines high-fidelity synthesis with intent-aware prompt enhancement to generate text-rich, layout-coherent images, achieving state-of-the-art open-weight results on long-form text and layout benchmarks.

DLite: History-free Speculative Decoding for Agentic AI

Hugging Face Trending Papers — 2026-10-16T00:00:00+00:00

DLite improves speculative decoding for agentic AI by eliminating historical context maintenance and using target hidden states for local predictions, boosting acceptance length and reducing overhead.

Xiaomi-Lens-0 Technical Report

Hugging Face Trending Papers — 2026-10-16T00:00:00+00:00

Xiaomi-Lens-0 is a 4B multimodal model that predicts masks from internal visual representations to improve fine-grained region reasoning and referring comprehension.

OR-License: Evaluating LLM Agents as Real-World Operations Research Practitioners

Hugging Face Trending Papers — 2026-10-16T00:00:00+00:00

OR-License evaluates whether LLM agents can solve real-world operations research problems from original client materials, finding that while some achieve high objective quality, none match expert-level feasibility and deployment-adjusted quality.

RoboHarness: A Simple Harness Outperforms VLA and World Action Models

Hugging Face Trending Papers — 2026-10-16T00:00:00+00:00

RoboHarness enables LLM agents to perform embodied manipulation via a visual-geometric control interface, surpassing specialized vision-language-action models on robotics benchmarks.

Welcome Beam: a 501B open model you can own, customize, and scale

Hugging Face Blog — 2026-10-16T00:00:00+00:00

Welcome Beam: a 501B open model you can own, customize, and scale

Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint

Hugging Face Blog — 2026-10-15T20:34:01+00:00

Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint

Apertus 1.5 Release Notes

Hugging Face Blog — 2026-10-15T09:08:45+00:00

Apertus 1.5 Release Notes

3DCodeVerse: Exploring 3D Worlds as Code

Hugging Face Trending Papers — 2026-10-15T00:00:00+00:00

3DCodeVerse is a research ecosystem that provides datasets, automated verification harnesses, and empirical guidance to enable scalable, editable, and verifiable 3D content generation via executable code.

OpenGCUA: Toward a Fully Open Stack for Generalist Computer Use

Hugging Face Trending Papers — 2026-10-15T00:00:00+00:00

OpenGCUA introduces an open generalist computer-use agent that integrates visual and programmatic interfaces, harness-to-model decision transfer, and code execution to improve task performance and efficiency across models.

Mosaic: Multi-Session Agentic Reinforcement Learning

Hugging Face Trending Papers — 2026-10-15T00:00:00+00:00

Multi-session agentic reinforcement learning is introduced to handle trajectories across multiple sessions, with Mosaic optimizing session-level contributions and allocation via hierarchical advantage estimation.

Has GPT-6-Astra Achieved Embodied Intelligence?

Hugging Face Trending Papers — 2026-10-15T00:00:00+00:00

GPT-6-Astra achieves high performance across diverse robotic tasks via in-context learning, yet real-world deployment reveals gaps in demonstration cost, execution time, and practical embodied intelligence.

SUPRA-TEX: 3D Generation with Super-Aligned Texture

Hugging Face Trending Papers — 2026-10-15T00:00:00+00:00

SUPRA-Tex decouples visible-material estimation, surface alignment, and 3D completion to generate relightable 3D materials that preserve reference image details and complete unobserved surfaces.

TokenEconomizer: Experience-Driven Black-Box Cost Control for Coding Agents

Hugging Face Trending Papers — 2026-10-15T00:00:00+00:00

TokenEconomizer reduces recurring token costs of coding agents across sequential real-world tasks by pruning irrelevant tools, injecting micro-skills, and interrupting failure loops.

Welcome FLUX.3 Image Klein in Diffusers

Hugging Face Blog — 2026-10-15T00:00:00+00:00

Welcome FLUX.3 Image Klein in Diffusers

Welcome Muse Spark Open: open-weight frontier multimodal mixture-of-experts model from Meta

Hugging Face Blog — 2026-10-15T00:00:00+00:00

Welcome Muse Spark Open: open-weight frontier multimodal mixture-of-experts model from Meta

HiRO-ACEv2: Simulating extreme weather at 3 km resolution for a fraction of the cost

Hugging Face Blog — 2026-10-14T16:12:42+00:00

HiRO-ACEv2: Simulating extreme weather at 3 km resolution for a fraction of the cost

Any Time, Any Space: A World-State-Centric Framework for Agentic Video Generation

Hugging Face Trending Papers — 2026-10-14T00:00:00+00:00

ATAS improves long-form video consistency by compiling screenplays into referenced shot requests, retrieving spatial and temporal evidence from scene and event memories, and conditionally updating memory after generation.

Player-1: Learning to Play from Synthetic Experience

Hugging Face Trending Papers — 2026-10-14T00:00:00+00:00

Player-1 improves game control by combining supervised fine-tuning on gameplay trajectories with reinforcement learning across synthetic environments, achieving substantial gains over demonstration-only baselines.

Towards Agentic Robots: Model, Data, Environment, and Harness

Hugging Face Trending Papers — 2026-10-14T00:00:00+00:00

This survey unifies research on agentic robots by tracing the evolution from robot policies to integrated systems using vision-language-action models, world models, and structured functional domains.

More Methods, Better Agents? A Reality Check on Self-Distillation in Agentic Reinforcement Learning

Hugging Face Trending Papers — 2026-10-14T00:00:00+00:00

Researchers benchmarked ten self-distillation methods against GRPO across multiple agent tasks and model sizes, finding limited consistent gains, hidden training instability, and no reliable cost-to-performance benefit.

Seeing, Maintaining, and Learning: The Evolution of Multimodal Agent Memory

Hugging Face Trending Papers — 2026-10-14T00:00:00+00:00

This work proposes a two-axis framework for multimodal agent memory covering information formation and adaptive management to support lifelong learning.

InfiniSplat v2: Analyzing and Improving Feed-Forward 3DGS from Any Views

Hugging Face Trending Papers — 2026-10-14T00:00:00+00:00

InfiniSplat v2 improves feed-forward 3D Gaussian reconstruction by decoupling geometry and attribute prediction, aligning Gaussians with ground-truth poses, and using large-angle supervision to enhance novel-view rendering quality.

BizBench: Unveiling Hidden Challenges in Deploying Long-Horizon Agents for Real-World Business

Hugging Face Trending Papers — 2026-10-14T00:00:00+00:00

BizBench evaluates long-horizon business agents on intermediate execution validity using real-world payment procedures, revealing that frontier models frequently violate constraints despite achieving partial goal success.

BlenderWorld: Benchmarking Visual Agents for 3D Creation from Real-World References

Hugging Face Trending Papers — 2026-10-14T00:00:00+00:00

BlenderWorld is a benchmark for evaluating multimodal language models on realistic 3D creation tasks using rendered outputs and binary rubrics, revealing significant gaps in spatial coherence.

PrismaCoder: Pushing the Boundaries of Open Multimodal Code Intelligence through Decomposed Rubric Synthesis

Hugging Face Trending Papers — 2026-10-14T00:00:00+00:00

PrismaForge uses a rubric-based pipeline to synthesize and filter high-quality multimodal code data, enabling open models that rival proprietary systems on visual coding tasks.

What Generation Knows: Unlocking Fine-Grained Understanding in Unified Multimodal Models

Hugging Face Trending Papers — 2026-10-14T00:00:00+00:00

GAFU leverages native generation-side representations to improve fine-grained visual understanding in unified multimodal models by compressing them into compact tokens for task-conditioned retrieval.

Run fast, private agents on-device with LFM2.5-24B-A2B

Hugging Face Blog — 2026-10-13T17:05:48+00:00

Run fast, private agents on-device with LFM2.5-24B-A2B

Urbex: Agentic Spatial Grounding for City-Scale 3D Scenes