Loading…
Hugging Face
AI platform and community providing machine learning models, datasets, libraries, and tools for building AI applications.
- 316 posts summarised
- latest
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
From their blog
Latest articles
Build a Domain-Specific Embedding Model in Under a Day
The tutorial presents a pipeline for adapting a general-purpose embedding model when it misses fine-grained distinctions in domain documents and labeled query–document data is unavailable. It generates synthetic question–answer pairs with NeMo Data Designer, filters them for quality, mines hard negatives with a 95% positive-score margin ceiling, and unrolls multi-hop queries for contrastive training. The recipe fine-tunes Llama-Nemotron-Embed-1B-v2, evaluates base and adapted models with BEIR metrics, and exports the result to ONNX or TensorRT. On the reported datasets, it improved Recall@10 and NDCG@10 by more than 10%; Atlassian increased Recall@60 from 0.751 to 0.951 on JIRA data. The complete workflow can run in roughly 2–3 hours for about 500 documents, using a single 80GB Ampere-or-newer GPU.
State of Open Source on Hugging Face: Spring 2026
The post examines how open source AI on Hugging Face changed over the past year across competition, geography, technical trends, and emerging communities. It combines Hugging Face and community-contributed metrics to assess users, models, datasets, downloads, repositories, and specialized sub-ecosystems. In 2025, the platform reached 13 million users, more than 2 million public models, and over 500,000 public datasets, while the top 200 models represented 49.6% of downloads and China surpassed the U.S. in monthly and overall downloads. The analysis also describes rising participation by independent developers, increased Chinese releases, and active communities in robotics and science, framing open source AI as overlapping ecosystems rather than a uniform market.
Holotron-12B - High Throughput Computer Use Agent
Holotron-12B is a multimodal computer-use model released by H Company and available on Hugging Face. Post-trained from NVIDIA’s Nemotron-Nano-12B-v2-VL-BF16 on proprietary data, it targets agents that must perceive, decide, and act in interactive environments while handling long contexts with multiple images. Its hybrid State-Space Model and attention architecture stores a constant state per layer per generated sequence instead of attention’s per-token key and value activations, reducing memory demands for long-context inference. In a WebVoyager evaluation on one H100 with vLLM, it reached 8.9k tokens/s at concurrency 100 versus 5.1k for Holo2-8B, and its benchmark score rose from 35.1% to 80.5% after training on localization and navigation data.
Introducing Storage Buckets on the Hugging Face Hub
Hugging Face introduces Storage Buckets as a mutable, S3-like object storage layer on the Hub for intermediate machine-learning artifacts such as checkpoints, optimizer states, processed shards, logs, and traces. Unlike versioned Models and Datasets repositories, Buckets support frequent writes, overwrites, directory synchronization, and cleanup while retaining standard Hub permissions and public or private visibility. Built on Xet, they chunk and deduplicate content across files, reducing repeated bytes, bandwidth, transfer time, and, for Enterprise customers, billed storage. The CLI and Python APIs support creation, syncing, dry runs, plans, inspection, individual file operations, and programmatic workflows; HfFileSystem and fsspec also allow compatible libraries to read and write through hf:// paths. Pre-warming can place hot data near AWS or GCP compute, while roadmap work aims to connect Buckets directly with versioned repos.
Keep the Tokens Flowing: Lessons from 16 Open-Source RL Libraries
At-scale reinforcement-learning training is constrained by long, variable data-generation workloads that leave training GPUs idle, especially for reasoning and agentic models. The surveyed architecture separates inference and training onto distinct GPU pools, connects them through a rollout buffer, and synchronizes weights asynchronously so both stages can proceed concurrently. Sixteen open-source libraries are compared across orchestration, buffering, weight synchronization, staleness, partial rollouts, LoRA, and distributed backends; Ray appears in 8 of 16 orchestration implementations, while NCCL broadcast is the default weight-transfer method. The proposed TRL async trainer favors a lightweight bounded queue with per-token model_version metadata, packed NCCL transfers, and experimental support for prefix-resume or abort-and-retry handling of rollouts interrupted by weight updates.
Ulysses Sequence Parallelism: Training with Million-Token Contexts
Ulysses Sequence Parallelism addresses the memory and compute pressure of training language models on sequences extending beyond tens of thousands of tokens, where attention remains quadratic in sequence length even with FlashAttention. It shards tokens across GPUs, projects local queries, keys, and values, then uses two all-to-all operations to exchange data so each GPU computes a subset of attention heads before restoring sequence-sharded outputs. The post describes integrations with Accelerate, the Transformers Trainer, and TRL's SFTTrainer, including variable-length handling and weighted loss aggregation for uneven valid-token counts. In reported four-GPU tests, SP=4 reduced per-GPU memory sufficiently for 96K-token training on H100 80GB GPUs, reached 13,396 tokens per second at 64K, and matched non-SP canonical token-normalized loss within 0.000005 maximum absolute difference.
LeRobot v0.5.0: Scaling Every Dimension
LeRobot v0.5.0 is a major release that expands supported hardware, policies, datasets, simulation, and its software foundation, following more than 200 merged PRs and more than 50 new contributors since v0.4.0. It adds full Unitree G1 humanoid support for locomotion, manipulation, teleoperation, and whole-body control, alongside OpenArm, Earth Rover, CAN-bus motor controllers, and other integrations. The policy additions include Pi0-FAST's autoregressive VLAs with FAST action tokenization, Real-Time Chunking for continuously blended predictions, Wall-X, X-VLA, SARM, and PEFT fine-tuning. For data and simulation workflows, streaming video encoding removes waits between recording episodes, while EnvHub loads environments from the Hugging Face Hub and IsaacLab-Arena supplies GPU-accelerated simulation. The codebase now requires Python 3.12, uses Transformers v5, supports installable third-party policy plugins, and adds remote Rerun visualization plus installation and documentation improvements.
Bringing Robotics AI to Embedded Platforms: Dataset Recording, VLA Fine‑Tuning, and On‑Device Optimizations
Deploying Vision–Language–Action policies on embedded robots is constrained by compute, memory, power, and real-time control requirements; synchronous inference can leave the arm idle and cause oscillatory corrections. NXP presents dataset-recording practices for a tea-bag placement task, including fixed cameras, controlled lighting, a gripper-mounted view, calibration backups, and diverse training and validation splits with recovery episodes. The example uses 120 episodes, three 640×480 cameras at 30 fps, batch size 8, and checkpoint selection based on training and validation success rather than loss alone. For deployment, the VLA is decomposed into vision, LLM backbone, and action-expert stages, with selective quantization and asynchronous scheduling constrained by inference time being shorter than action execution. On the i.MX 95, optimized ACT latency reached 0.32 seconds with 0.89 global accuracy, while ONNX FP32 ACT reached 0.96 global accuracy at 2.86 seconds.
Introducing Modular Diffusers - Composable Building Blocks for Diffusion Pipelines
Modular Diffusers introduces a composable alternative to Diffusers’ existing DiffusionPipeline for building diffusion workflows from reusable blocks. A ModularPipeline can load a workflow definition, load model components separately with settings such as bfloat16, and preserve the familiar inference API, while blocks for text encoding, image encoding, denoising, and decoding can be inspected, run independently, or rearranged. Custom Python blocks declare components, inputs, outputs, and computation logic; the example adds a Depth Anything V2 depth processor to a Qwen Image ControlNet workflow, with data flowing between compatible blocks. Blocks and complete pipelines can be published to the Hub, and Mellon can generate node interfaces from their consistent definitions. The post presents the system as an early, flexible direction and says Mellon is not ready for production use.
PRX Part 3 — Training a Text-to-Image Model in 24h!
PRX Part 3 reports a 24-hour speedrun for training a text-to-image diffusion model, combining techniques that performed well in earlier experiments under a budget of about $1,500 on 32 H200 GPUs. The recipe uses x-prediction in pixel space with 32-pixel patches and a 256-dimensional initial bottleneck, followed by LPIPS and DINOv2 perceptual losses, TREAD token routing, REPA alignment with DINOv3, and Muon for matrix parameters while Adam handles the rest. Training ran for 100,000 steps at 512px with batch size 1,024, then 20,000 steps at 1024px with batch size 512 and no REPA, across three synthetic datasets. The resulting model is described as usable, with strong prompt following and consistent aesthetics, though texture glitches, anatomy issues, and difficulty with hard prompts remain. The post attributes these limitations mainly to undertraining and limited data diversity and presents the recipe as a foundation for further scaling.
Mixture of Experts (MoEs) in Transformers
Mixture of Experts (MoEs) in Transformers let a Transformer retain its backbone while replacing selected feed-forward layers with routed expert sub-networks, allowing total model capacity to exceed the active parameters used per token. The post explains why sparse models can reduce inference cost and memory pressure, then focuses on making them first-class in Transformers. Its weight-loading refactor converts independently serialized expert tensors into packed runtime layouts through composable WeightConverter operations and lazy materialization. The Experts Backend supplies eager, batched_mm, and grouped_mm execution strategies, while expert parallelism shards experts across devices and aggregates partial outputs. Training support uses grouped GEMM and custom Triton plus LoRA kernels, with reported improvements of approximately 12× speed, over 35% VRAM reduction, and roughly 6× longer context.
GGML and llama.cpp join HF to ensure the long-term progress of Local AI
Hugging Face announces that GGML, the creators of llama.cpp, are joining the company to support the ggml and llama.cpp community as local AI advances. Georgi Gerganov and his team will continue dedicating 100% of their time to maintaining llama.cpp, retaining autonomy and leadership over its technical direction and community, while the project remains fully open-source and community driven. The technical focus is tighter integration between llama.cpp’s local-inference stack and Transformers’ model definitions, aiming to make shipping new models from Transformers’ “source of truth” nearly single-click, alongside better packaging and user experience for ggml-based software. The longer-term goal is an efficient inference stack that runs on devices and helps make open-source superintelligence accessible, with llama.cpp becoming ubiquitous and easier for casual users to deploy and access.
Train AI models with Unsloth and Hugging Face Jobs for FREE
The post presents a workflow for fine-tuning the LiquidAI/LFM2.5-1.2B-Instruct small language model with Unsloth on Hugging Face Jobs, including optional use of Claude Code, Codex, or another coding agent. It explains that Unsloth is described as providing about 2x faster training and 60% lower VRAM use than standard methods, while the model runs under 1GB of memory and is optimized for on-device deployment. The tutorial covers claiming free Hugging Face Jobs credits through the Unsloth Jobs Explorers organization, installing the hf CLI or a model-training skill, and submitting a job with a dataset, GPU flavor, timeout, and output repository. An agent can generate a UV training script using 4-bit loading and LoRA, submit the managed cloud-GPU job, report Trackio monitoring details, and push the trained model to the Hugging Face Hub.
IBM and UC Berkeley Diagnose Why Enterprise Agents Fail Using IT-Bench and MAST
IBM Research and UC Berkeley analyze how agentic LLM systems fail in enterprise IT automation, including incident triage, logs and metrics queries, and Kubernetes actions across long-horizon tool loops. To move beyond benchmark success rates, the analysis applies MAST (Multi-Agent System Failure Taxonomy) to execution traces, converting failures into structured signatures across system design, inter-agent misalignment, and task verification. It covers 310 ITBench SRE traces from Gemini-3-Flash, Kimi-K2, and GPT-OSS-120B, whose mean recalls were 75.5%, 28.6%, and 12.4%, respectively. Gemini averaged 2.6 failure modes per failed trace, compared with 4.7 for Kimi and 5.3 for GPT-OSS-120B, while FM-3.3, Incorrect Verification, was identified as the strongest predictor across models; the analysis recommends external verification, explicit termination controls, and clarification branches.
One-Shot Any Web App with Gradio's gr.HTML
The post presents Gradio 6’s gr.HTML as a way to build interactive web components in a single Python file, combining custom HTML templates, scoped CSS, and JavaScript interactivity without a separate build step. It describes applications including a Pomodoro timer, contribution heatmap, Kanban board, spinning wheel, and machine-learning viewers for detection, 3D camera control, and speech transcription. Within the component API, ${value} injects Python state into the template, props.value updates that state from JavaScript, and trigger('change') synchronizes it back to Python; CSS keyframes and native HTML5 drag events supply richer behavior. For reuse, a gr.HTML subclass packages the templates and script so the result can be used like built-in Gradio components, while reload mode and Spaces deployment support rapid iteration from generated code to a live app.
Custom Kernels for All from Codex and Claude
Writing CUDA kernels that integrate with diffusers and transformers involves GPU-specific optimization, library integration, PyTorch registration, and environment compatibility. It presents a cuda-kernels agent skill that packages architecture guides, integration patterns, kernel templates, troubleshooting material, benchmarks, and Kernel Hub examples for coding agents. Claude and Codex used it to generate RMSNorm, RoPE 3D, GEGLU, and AdaLN kernels for LTX-Video in diffusers, and an RMSNorm kernel for Qwen3-8B in transformers, including CUDA source, PyTorch bindings, build configuration, and benchmark scripts. On H100 80GB HBM3, the LTX-Video RMSNorm tests averaged 1.88x speedup and the optimized pipeline with torch.compile reached 1.43x the listed baseline, while Qwen3-8B RMSNorm averaged 1.94x and reached 2.47x at 8192 tokens. The skill supports packaging kernels for precompiled distribution through the Hub, where get_kernel selects a matching binary across Python, PyTorch, and CUDA variants.
OpenEnv in Practice: Evaluating Tool-Using Agents in Real-World Environments
OpenEnv is an open-source framework from Meta and Hugging Face for evaluating AI agents against real systems rather than simulations, with stateful environments that connect to tools and APIs. Turing's production-grade Calendar Gym applies this model to calendar operations, combining access control lists, partial visibility, temporal reasoning, and multi-step workflows in isolated sessions. Agents often handled individual actions but became less reliable on longer or ambiguous tasks: success was close to 90% with explicit calendar identifiers and roughly 40% when requests used natural-language descriptions. More than half of failures involved malformed arguments or incorrect action ordering despite selecting the correct tool, while structured validation, permission, and datetime errors provided concrete recovery signals.
Transformers.js v4: Now Available on NPM!
Transformers.js v4 is now available on NPM, marking a major release after a year of development. It introduces a C++-rewritten WebGPU runtime tested across about 200 supported model architectures, enabling the same JavaScript code to run WebGPU-accelerated models in browsers, Node, Bun, Deno, and desktop applications. Specialized ONNX Runtime Contrib Operators support new export strategies for large language models, and the post reports a roughly fourfold speedup for BERT-based embedding models using MultiHeadAttention. The release also restructures the repository as a pnpm-workspaces monorepo, modularizes model definitions, moves examples to a separate repository, and switches from Webpack to esbuild, reducing build times from two seconds to 200 milliseconds and shrinking transformers.web.js by 53%. New APIs cover model asset inspection, caching, environment fetch controls, logging, and standalone tokenization, while added architectures include Mamba, MLA, and MoE models compatible with WebGPU.
Introducing SyGra Studio
SyGra 2.0.0 introduces Studio, an interactive environment for building, previewing, executing, and debugging synthetic data generation flows on a visual canvas. Users can configure connectors and models, preview source rows, reference automatically created state variables in prompts, define structured outputs, and inspect the generated YAML or JSON graph configuration. Studio streams node-level execution status, token usage, latency, cost, logs, and guardrail outcomes, while storing run history in .executions/runs/*.json and autosaving drafts. It also runs existing workflows, including a Glaive Code Assistant example with conditional revision loops, tunable dataset and batching inputs, and observable intermediate critiques; generated outputs can support training, evaluation, or annotation workflows.
Community Evals: Because we're done trusting black-box leaderboards over the community
Hugging Face is introducing Community Evals, a system for decentralized and transparent reporting of model evaluation results on the Hub. It responds to saturated benchmarks and inconsistent scores across model cards, papers, and evaluation platforms, while acknowledging that benchmark performance does not necessarily reflect real-world capabilities. Benchmark dataset repositories can register as benchmarks, define reproducible eval specifications in eval.yaml, aggregate scores from model repositories and community submissions, and display leaderboards in dataset cards. Model results are stored in .eval_results/*.yaml, while users can submit sourced scores through pull requests that remain visible as community results, with Git history and APIs supporting tracking and further aggregation. The beta feature aims to expose what was evaluated, how, when, and by whom, but it does not claim to solve benchmark saturation, test-set training, or the benchmark-reality gap.
Topics they write about
All topics →Weekly digest
The best engineering reads, once a week.