---
title: "Hugging Face"
description: "AI platform and community providing machine learning models, datasets, libraries, and tools for building AI applications."
---

# Hugging Face
> AI platform and community providing machine learning models, datasets, libraries, and tools for building AI applications.

## Articles

### [Welcome Gemma 4: Frontier multimodal intelligence on device](https://yomu.fyi/post/welcome-gemma-4-frontier-multimodal-intelligence-on-device.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: merve, Pedro Cuenca, Sergio Paniego, ben burtenshaw, Steven Zheng, Alvaro Bartolome, Nathan Habib
- Published: Apr 2, 2026
- Type: Announcement

Google DeepMind’s Gemma 4 family is available on Hugging Face as Apache 2-licensed, multimodal models intended for use across agents, inference engines, fine-tuning libraries, and on-device deployments. The family accepts image and text inputs across five sizes, while E2B, E4B, and 12B Unified also support audio; context windows range from 128k to 256k tokens. Its design combines alternating local and global attention, dual RoPE configurations, Per-Layer Embeddings, shared KV caches, and model-specific vision and audio handling, including an encoder-free unified 12B variant. The post documents integrations for Transformers, llama.cpp, MLX, WebGPU, Rust, and other tools, alongside fine-tuning examples and demos. Reported instruction-tuned results include a 1452 estimated text-only LMArena score for 31B and benchmark results spanning reasoning, coding, vision, audio, and long-context tasks.


### [Falcon Perception](https://yomu.fyi/post/falcon-perception.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: FalconPerception
- Published: Apr 1, 2026
- Type: Announcement

Falcon Perception is a 0.6B-parameter early-fusion Transformer for open-vocabulary grounding and segmentation from natural-language prompts. It processes image patches, text, and task tokens in one sequence, using hybrid attention so image tokens receive bidirectional context while text and task tokens decode causally. Its Chain-of-Perception interface predicts each instance as coordinate, size, and segmentation tokens, with specialized Fourier-feature and dot-product heads for dense outputs. The post also introduces PBench, which profiles attributes, OCR-guided identification, spatial understanding, relations, and crowded scenes. Falcon Perception reaches 68.0 Macro-F1 on SA-Co versus 62.3 for SAM 3, while presence calibration remains weaker at MCC 0.64 versus 0.82; the 0.3B Falcon OCR model scores 80.3 on olmOCR and 88.6 on OmniDocBench.


### [Any Custom Frontend with Gradio's Backend](https://yomu.fyi/post/any-custom-frontend-with-gradio-s-backend.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: yuvraj sharma, Abubakar Abid
- Published: Apr 1, 2026
- Type: Tutorial

gradio.Server enables applications built with React, Svelte, or plain HTML/JavaScript to use Gradio’s backend facilities instead of Gradio components. The post demonstrates this with Text Behind Image, whose browser UI layers a background, styled text, and a transparent foreground produced by BiRefNet, then exports a PNG client-side. A roughly 50-line Python backend extends FastAPI, exposes a standard HTML route, and defines an @app.api("remove\_background") endpoint; @spaces.GPU handles ZeroGPU allocation while Gradio supplies queuing, concurrency control, SSE streaming, and gradio\_client compatibility. The frontend calls the endpoint through the Gradio JavaScript client, so image-processing requests pass through the queue while rendering and compositing remain in the browser. The result is a custom frontend that retains Spaces hosting and Gradio infrastructure.


### [Granite 4.0 3B Vision: Compact Multimodal Intelligence for Enterprise Documents](https://yomu.fyi/post/granite-4-0-3b-vision-compact-multimodal-intelligence-for-enterprise-d.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Madison Lee, Rogerio Feris, Eli Schwartz, Dhiraj Joshi, Pengyuan Li, Isaac Sanchez
- Published: Mar 31, 2026
- Type: Announcement

Granite 4.0 3B Vision is a compact vision-language model designed to extract information from enterprise documents, including tables, charts, forms, and semantically meaningful key-value pairs. It combines a million-scale ChartNet dataset with a code-guided synthesis pipeline, DeepStack Injection for routing semantic and high-resolution visual features to different layers, and a LoRA adapter architecture built on Granite 4.0 Micro. The modular design supports standalone image processing, text-only fallback, and integration with Docling for multi-page document pipelines. Reported evaluations show 86.4% on Chart2Summary, leading the evaluated models, second place on Chart2CSV at 62.1%, leading scores on the listed table benchmarks, and 85.5% zero-shot exact-match accuracy on VAREX.


### [Training mRNA Language Models Across 25 Species for $165](https://yomu.fyi/post/training-mrna-language-models-across-25-species-for-165.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Maziyar Panahi
- Published: Mar 31, 2026
- Type: Benchmark

OpenMed describes an end-to-end protein engineering pipeline that connects ESMFold structure prediction, ProteinMPNN sequence design, and custom mRNA codon optimization. The codon-language-model study compares CodonBERT, ModernBERT, and several CodonRoBERTa variants using 250,000 E. coli coding sequences, a 69-token codon vocabulary, masked-language modeling, and training on four 80GB A100 GPUs. CodonRoBERTa-large-v2 achieved perplexity 4.10, CAI Spearman correlation 0.404, and 7.7% synonymous recovery, while ModernBERT-base reached perplexity 26.24 and correlation 0.070. The work then trained a multi-species suite on 381,283 sequences spanning 25 species, producing four production models in 55 GPU-hours and documenting planned extensions such as larger training data and additional prediction heads.


### [TRL v1.0: Post-Training Library Built to Move with the Field](https://yomu.fyi/post/trl-v1-0-post-training-library-built-to-move-with-the-field.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Quentin Gallouédec, Steven Liu, Pedro Cuenca, Sergio Paniego
- Published: Mar 31, 2026
- Type: Announcement

TRL v1.0 marks a transition from research codebase to a library intended for production systems, with clearer stability expectations. It implements more than 75 post-training methods, but its design prioritizes trying, comparing, and using them over coverage alone. The library responds to a field in which PPO, preference optimization methods such as DPO, ORPO, and KTO, and RLVR-style approaches such as GRPO repeatedly changed which components of the training stack appear essential. TRL addresses that volatility with a stable core under semantic versioning, an experimental surface for faster-moving methods, and deliberately limited abstractions that favor explicit implementations over generic hierarchies. v1.0 also follows distributed breaking changes across the 0.x releases, so migration from the last 0.x version is described as minimal, while future work targets asynchronous GRPO, scaling, and actionable training diagnostics.


### [Liberate your OpenClaw](https://yomu.fyi/post/liberate-your-openclaw.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Clem 🤗, ben burtenshaw, Pedro Cuenca, Jeff Boudier, merve, Niels Rogge, Victor Mustar, Mishig Davaadorj
- Published: Mar 27, 2026
- Type: Tutorial

Anthropic is limiting access to Claude models in open agent platforms for Pro/Max subscribers, so the post describes ways to keep OpenClaw, Pi, or Open Code agents running with open models from Hugging Face. It presents Hugging Face Inference Providers as the quickest hosted route: users create a token, run openclaw onboard --auth-choice huggingface-api-key, and select a model such as GLM-5, or set a repo\_id in OpenClaw’s config. For fully local inference, it recommends installing llama.cpp, starting a server with Qwen3.5-35B-A3B-GGUF, and connecting OpenClaw to the local OpenAI-compatible endpoint with custom configuration; the example model is described as suitable for 32GB of RAM. The hosted option prioritizes speed and model availability, while local inference offers privacy, zero API costs, full control, and no rate limits.


### [A New Framework for Evaluating Voice Agents (EVA)](https://yomu.fyi/post/a-new-framework-for-evaluating-voice-agents-eva.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Tara Bogavelli, Gabrielle Gauthier Melancon, Katrina Stankiewicz, Nifemi Bamgbose, Hoang Nguyen, Raghav Mehndiratta, Hari Subramani, Fanny Riols
- Published: Mar 24, 2026
- Type: Benchmark

EVA is an end-to-end framework for evaluating conversational voice agents across task accuracy and spoken conversational experience. It simulates multi-turn live-audio conversations with a user simulator, the evaluated voice agent, deterministic tool execution, validation, and metrics derived from recordings, transcripts, and tool-call logs. The framework reports EVA-A for Accuracy and EVA-X for Experience, combines code-based and LLM- or LALM-as-Judge metrics, and uses pass@3 and pass^3 across three trials to capture peak performance and consistency. Its initial release contains 50 reproducible English-language airline scenarios and 15 tools covering rebooking, itinerary changes, cancellations, standby, and vouchers. Results for 20 cascade and audio-native systems show a consistent tradeoff: agents stronger at task completion tend to provide worse experiences, while a substantial gap between pass@3 and pass^3 indicates inconsistent success.


### [Build a Domain-Specific Embedding Model in Under a Day](https://yomu.fyi/post/build-a-domain-specific-embedding-model-in-under-a-day.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Steve Han, Rucha Apte, Sean Sodha, Oliver Holworthy
- Published: Mar 20, 2026
- Type: Tutorial

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](https://yomu.fyi/post/state-of-open-source-on-hugging-face-spring-2026.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Avijit Ghosh, Lucie-Aimée Kaffee, Yacine Jernite, Irene Solaiman
- Published: Mar 17, 2026
- Type: Explainer

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](https://yomu.fyi/post/holotron-12b-high-throughput-computer-use-agent.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Pierre-Louis Cedoz, Hamza Benchekroun, Aurélien Lac, delfosse, Tony Wu, Antoine Bonnet, Kai Yuan, Aleix Cambray (H-AI), Alexandra, Axel Moyal
- Published: Mar 17, 2026
- Type: Announcement

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](https://yomu.fyi/post/introducing-storage-buckets-on-the-hugging-face-hub.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Lucain Pouget, Eliott Coyac, Adrien Carreira, Victor Mustar, Julien Chaumond, Quentin Lhoest, Pierric Cistac, Sylvestre Bcht, Hugo Larcher, Rajat Arya, Di Xiao, Assaf Vayner
- Published: Mar 10, 2026
- Type: Announcement

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](https://yomu.fyi/post/keep-the-tokens-flowing-lessons-from-16-open-source-rl-libraries.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Amine Dirhoussi, Quentin Gallouédec, Kashif Rasul, Lewis Tunstall, Edward Beeching, Albert Villanova del Moral, Nouamane Tazi, Leandro von Werra, Sergio Paniego
- Published: Mar 10, 2026
- Type: Benchmark

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](https://yomu.fyi/post/ulysses-sequence-parallelism-training-with-million-token-contexts.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Kashif Rasul, Stas Bekman
- Published: Mar 9, 2026
- Type: Explainer

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](https://yomu.fyi/post/lerobot-v0-5-0-scaling-every-dimension.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Steven Palma, Pepijn Kooijmans, Jade Choghari, Caroline Pascal, Khalil Meftah, Martino Russi, Nicolas Rabault, Michel Aractingi, Virgile BATTO, Thomas Wolf
- Published: Mar 9, 2026
- Type: Announcement

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](https://yomu.fyi/post/bringing-robotics-ai-to-embedded-platforms-dataset-recording-vla-fine.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Gaetan Bahl, Enzo Ruedas, Tess Boivin
- Published: Mar 5, 2026
- Type: Tutorial

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](https://yomu.fyi/post/introducing-modular-diffusers-composable-building-blocks-for-diffusion.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: YiYi Xu, Alvaro Somoza, Dhruv Nair, Sayak Paul
- Published: Mar 5, 2026
- Type: Announcement

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!](https://yomu.fyi/post/prx-part-3-training-a-text-to-image-model-in-24h.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: David Bertoin, Roman Frigg, Jon Almazán
- Published: Mar 3, 2026
- Type: Benchmark

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](https://yomu.fyi/post/mixture-of-experts-moes-in-transformers.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Aritra Roy Gosthipaty, Pedro Cuenca, merve, Ilyas Moutawwakil, Arthur Zucker, Sergio Paniego, Pablo Montalvo
- Published: Feb 26, 2026
- Type: Explainer

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](https://yomu.fyi/post/ggml-and-llama-cpp-join-hf-to-ensure-the-long-term-progress-of-local-a.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Georgi Gerganov, Xuan-Son Nguyen, Aleksander Grygier, Lysandre, Victor Mustar, Julien Chaumond
- Published: Feb 20, 2026
- Type: Announcement

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.


[Newer posts](https://yomu.fyi/company/hugging-face/page/6.md) · [Older posts](https://yomu.fyi/company/hugging-face/page/8.md)