---
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

### [Fine-tune Any LLM from the Hugging Face Hub with Together AI](https://yomu.fyi/post/fine-tune-any-llm-from-the-hugging-face-hub-with-together-ai.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Zain Hasan, Artem Chumachenko, Egor Timofeev, Max Ryabinin
- Published: Sep 10, 2025
- Type: Announcement

Together AI and Hugging Face have introduced an integration enabling developers to fine-tune compatible Hugging Face Hub models directly on Together AI's managed infrastructure. To launch a fine-tuning job via the Python SDK, users provide a base model from Together's catalog as a configuration template alongside the target Hugging Face repository identifier. This base model template dictates GPU allocation, memory configuration, training pipelines, and inference setup for custom models with matching architectures and sizes. The workflow operates bidirectionally, pulling from public or token-authenticated private repositories and optionally pushing completed checkpoints back to the Hub upon completion. Geared toward CausalLM models under 100 billion parameters, the capability enables faster iteration cycles and domain adaptation without custom DevOps infrastructure.


### [Jupyter Agents: training LLMs to reason with notebooks](https://yomu.fyi/post/jupyter-agents-training-llms-to-reason-with-notebooks.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Baptiste Colle, Hanna Yukhymenko, Leandro von Werra
- Published: Sep 10, 2025
- Type: Problem & solution

Small language models often struggle to compete with large frontier models on complex, agentic data science tasks. To improve notebook-based reasoning, researchers simplified agent scaffolding down to roughly two hundred lines of code with dedicated execution and final answer tools, boosting baseline easy accuracy on the DABStep benchmark from 44.4 percent to 59.7 percent. They constructed a curated training dataset by deduplicating two terabytes of Kaggle notebooks, automatically fetching five terabytes of linked datasets, and scoring educational value and relevance using Qwen3-32B. Question-answer pairs grounded in verified execution traces were generated to fine-tune compact Qwen3-4B thinking and instruct models. The team released the trained models alongside the dataset and execution sandboxes, establishing a foundation for reinforcement learning and distillation on notebook workflows.


### [mmBERT: ModernBERT goes Multilingual](https://yomu.fyi/post/mmbert-modernbert-goes-multilingual.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Marc Marone, Orion Weller, William Fleshman, Eugene Yang, Dawn Lawrie, Ben Van Durme
- Published: Sep 9, 2025
- Type: Announcement

mmBERT is a massively multilingual encoder model trained on more than 3T tokens across over 1,800 languages to improve upon existing multilingual architectures like XLM-R. Built upon ModernBERT with a Gemma 2 tokenizer, mmBERT employs a three-phase training curriculum consisting of pre-training on 60 languages, mid-training on 110 languages, and a final decay phase covering 1,833 languages. The training pipeline integrates an inverse mask ratio schedule, dynamic language temperature annealing, and TIES merging across three decay variants. Benchmark evaluations demonstrate strong natural language understanding on English GLUE and multilingual XTREME, as well as competitive retrieval performance on MTEB v2 and CoIR. The release includes standard base and small models alongside open training data and checkpoints.


### [Welcome EmbeddingGemma, Google's new efficient embedding model](https://yomu.fyi/post/welcome-embeddinggemma-google-s-new-efficient-embedding-model.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Tom Aarsen, Joshua, Alvaro Bartolome, Aritra Roy Gosthipaty, Pedro Cuenca, Sergio Paniego
- Published: Sep 4, 2025
- Type: Announcement

Google DeepMind released EmbeddingGemma, a multilingual embedding model with 308 million parameters and a 2048-token context window designed for on-device applications. Based on the Gemma3 transformer backbone, the architecture replaces causal attention with bidirectional attention to function as an encoder, followed by mean pooling and two dense layers producing 768-dimensional vectors. The model incorporates Matryoshka Representation Learning, allowing outputs to be truncated down to 512, 256, or 128 dimensions for reduced memory and storage footprints. Trained on approximately 320 billion multilingual tokens across more than 100 languages, the quantized model operates under 200 MB of RAM. In domain-specific evaluations on the MIRIAD dataset, fine-tuning increased NDCG@10 from 0.8340 to 0.8862, outperforming larger baselines.


### [SAIR: Accelerating Pharma R&D with AI-Powered Structural Intelligence](https://yomu.fyi/post/sair-accelerating-pharma-r-d-with-ai-powered-structural-intelligence.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Arman Zaribafiyan, Georgia Channing, Rudi Plesch, Zane Beckwith
- Published: Sep 2, 2025
- Type: Announcement

SandboxAQ released the Structurally Augmented IC50 Repository (SAIR), an open-source dataset containing 5.24 million computationally co-folded 3D protein-ligand structures paired with empirical IC50 binding potency data. The repository addresses training data scarcity in structure-based drug discovery, where over 40 percent of the included target proteins lack experimental structures in the Protein Data Bank. To construct the dataset, engineers executed over 130,000 GPU hours of the Boltz1 co-folding model on 760 NVIDIA H100 processors hosted via NVIDIA DGX Cloud on Google Cloud Platform. Infrastructure optimizations maintained over 95 percent GPU compute utilization, compressing the generation timeline from three months to three weeks. Quality validation using PoseBusters confirmed that 97 percent of the predicted complexes met physical plausibility and chemical sanity standards.


### [Policy Gradient with PyTorch](https://yomu.fyi/post/policy-gradient-with-pytorch.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Thomas Simonini
- Published: Jun 30, 2022
- Type: Tutorial

Policy gradient algorithms optimize reinforcement learning policies directly without learning intermediate action-value functions. Instead of assigning discrete Q-values to actions, these methods parameterize a stochastic policy and apply gradient ascent on an objective score function. This setup eliminates manual exploration tuning, resolves perceptual aliasing in identical states, and naturally accommodates continuous or high-dimensional action spaces. The Reinforce Monte Carlo policy gradient algorithm operationalizes this approach by collecting full episodic trajectories and adjusting parameters along the gradient of the log-action probabilities scaled by return. Practitioners implement this algorithm using PyTorch to evaluate agent robustness across environments like CartPole-v1, PixelCopter, and Pong.


### [Liftoff! How to get started with your first ML project 🚀](https://yomu.fyi/post/liftoff-how-to-get-started-with-your-first-ml-project.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Nima Boscarino
- Published: Jun 29, 2022
- Type: Tutorial

Beginners in machine learning frequently face difficulties when selecting a framework and defining the scope of their initial hands-on project. Sentence Transformers provides an accessible starting point by computing dense vector representations for sentences, paragraphs, and images, enabling semantic search through cosine similarity. A structured four-step strategy guides practitioners through listing library capabilities, identifying compelling datasets, selecting a familiar secondary tool, and brainstorming project concepts. Applying this framework, a song lyrics dataset was paired with Gradio Blocks to build a prompt-based playlist generator. Developing this application required evaluating pre-trained models, hosting generated embeddings on Hugging Face Spaces, and utilizing multi-processor support to accelerate embedding generation.


### [Accelerate Large Model Training using DeepSpeed](https://yomu.fyi/post/accelerate-large-model-training-using-deepspeed.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Sourab Mangrulkar, Sylvain Gugger
- Published: Jun 28, 2022
- Type: Tutorial

Training large models on hardware with limited GPU memory frequently causes out-of-memory errors when using standard Distributed Data Parallel. To solve this bottleneck, Hugging Face Accelerate integrates DeepSpeed ZeRO data parallelism to shard optimizer states, gradients, and model parameters across workers. For a 900-million-parameter DeBERTa model, ZeRO Stage 2 increased the maximum per-device batch size from eight to forty while achieving a 3.5-fold training speedup over DDP without degrading accuracy or F1 score. Advanced configurations allow sequence-to-sequence chatbot finetuning and full ZeRO Stage 3 CPU offloading for models like the 1.5-billion-parameter GPT-XL, enabling batch size 16 training where DDP fails. Accelerate enables these memory optimizations through simple configuration files and the accelerate launch command with minimal or zero code modifications.


### [Announcing Evaluation on the Hub](https://yomu.fyi/post/announcing-evaluation-on-the-hub.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Lewis Tunstall, Abhishek, Tristan Thrush, Sasha Luccioni, Leandro von Werra, Nazneen Rajani, Aleksandra Piktus, Omar Sanseviero, Douwe Kiela
- Published: Jun 28, 2022
- Type: Announcement

Hugging Face introduced Evaluation on the Hub, a no-code tool designed to streamline the evaluation of machine learning models across diverse datasets. Traditional evaluation approaches often suffer from reproducibility issues, implementation inconsistencies, and cumbersome workflows across varied metrics and datasets. Powered by AutoTrain and Hugging Face Spaces, the service allows users to configure tasks, map dataset columns, select evaluation metrics, and run tests directly from dataset pages. Completed evaluations automatically open pull requests on the respective model cards to encode standardized verification metadata and update public dataset leaderboards. This system establishes consistent benchmarking pipelines across tasks like image classification, text summarization, and named entity recognition without requiring local code execution.


### [Getting Started With Embeddings](https://yomu.fyi/post/getting-started-with-embeddings.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Omar Espejel
- Published: Jun 23, 2022
- Type: Tutorial

Embeddings represent unstructured information such as text and images as numerical vectors in a shared semantic space. To demonstrate their utility, a simple semantic search engine is built over US Social Security Medicare frequently asked questions. The system generates 384-dimensional vector representations for thirteen FAQ entries by dispatching POST requests to the Hugging Face Inference API using the sentence-transformers/all-MiniLM-L6-v2 model. Incoming user queries are converted into matching vector representations and evaluated against stored dataset vectors using the util.semantic\_search function from the Sentence Transformers library. By calculating cosine similarity scores, the system retrieves and ranks the five most semantically relevant questions without requiring custom keyword rules or massive labeled training sets.


### [Convert Transformers to ONNX with Hugging Face Optimum](https://yomu.fyi/post/convert-transformers-to-onnx-with-hugging-face-optimum.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Philipp Schmid
- Published: Jun 22, 2022
- Type: Tutorial

Exporting Hugging Face Transformers models to the ONNX format can be accomplished through three different abstraction levels. The low-level approach utilizes torch.onnx.export, requiring manual specification of dummy inputs, input names, output names, opset versions, and dynamic axes configurations. At the intermediate level, the transformers.onnx package simplifies the conversion process by relying on FeaturesManager and prebuilt configuration objects to handle dynamic axis definitions automatically. The high-level method uses Hugging Face Optimum classes such as ORTModelForSequenceClassification by setting the from\_transformers flag to True inside from\_pretrained. This Optimum export leverages transformers.onnx internally and produces a model ready for immediate inference execution or integration into pipelines.


### [Intel and Hugging Face Partner to Democratize Machine Learning Hardware Acceleration](https://yomu.fyi/post/intel-and-hugging-face-partner-to-democratize-machine-learning-hardwar.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Julien Simon
- Published: Jun 15, 2022
- Type: Announcement

Intel has joined Hugging Face's Hardware Partner Program to accelerate Transformer training, fine-tuning, and inference on Intel platforms. Large Transformer models introduce latency bottlenecks in production workloads like search and chatbots, where hardware-level optimization typically requires tedious trial and error. To streamline model optimization, the collaboration introduces Optimum Intel, an open-source library integrating the Intel Neural Compressor for automated quantization, pruning, and distillation. A demonstration applies post-training dynamic quantization to a fine-tuned DistilBERT classification model using a CPU-only PyTorch setup. The quantized model converted 38 Linear and 2 Embedding operators to 8-bit integers, reducing evaluation duration by 1.34x while keeping the accuracy drop within a 5% threshold.


### [Director of Machine Learning Insights \[Part 3: Finance Edition\]](https://yomu.fyi/post/director-of-machine-learning-insights-part-3-finance-edition.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Britney Muller
- Published: Jun 14, 2022
- Type: Explainer

Machine learning leaders in financial institutions face significant operational hurdles when integrating automated models into production environments. Practitioners must navigate complex legacy architectures, strict regulatory oversight, and privacy mandates tied to personally identifiable financial records. Integrating models often fails when organizations lack clear communication buffers, deploy models using improper prediction windows, or treat systems as opaque black boxes without verifying underlying mechanics. Despite these challenges, financial applications increasingly rely on machine learning for anti-money laundering compliance, automated fraud screening, and loan underwriting. Success in this regulated sector demands comprehensive evaluation across representative input spaces, robust ongoing model monitoring, and explainable decision outputs.


### [The Annotated Diffusion Model](https://yomu.fyi/post/the-annotated-diffusion-model.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Niels Rogge, Kashif Rasul
- Published: Jun 7, 2022
- Type: Tutorial

Denoising Diffusion Probabilistic Models generate data samples by learning to reverse a discrete-time forward diffusion process that incrementally adds Gaussian noise to inputs. The forward process adds noise across a predefined schedule of variance parameters up to a fixed number of steps, transforming data into an isotropic Gaussian distribution. Because the true reverse conditional probability distribution is intractable, a neural network optimizes the variational lower bound to approximate the distribution mean at each step while keeping the variance fixed. Sampling subsequently proceeds by iteratively denoising pure Gaussian noise through successive network evaluations to reconstruct a clean data sample. Although diffusion models demonstrate high generation quality, their primary drawback remains the requirement of multiple forward passes during inference compared to models like generative adversarial networks.


### [Deep Q-Learning with Space Invaders](https://yomu.fyi/post/deep-q-learning-with-space-invaders.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Thomas Simonini
- Published: Jun 7, 2022
- Type: Explainer

Tabular Q-learning fails to scale to complex environments like Atari games due to massive observation spaces that make maintaining discrete state-action tables impractical. Deep Q-learning addresses this limitation by using a neural network to approximate Q-values for each possible action from input states. To process visual inputs effectively and capture temporal motion, raw game screens are grayscaled, resized to 84x84 pixels, and stacked in groups of four consecutive frames before passing through convolutional and fully connected layers. The training process stabilizes learning through experience replay buffers, fixed target networks that update periodically, and Double DQN architectures that decouple action selection from target evaluation. Agents can then be trained on environments like Space Invaders using frameworks such as RL-Zoo without tabular memory constraints.


### [Graphcore and Hugging Face Launch New Lineup of IPU-Ready Transformers](https://yomu.fyi/post/graphcore-and-hugging-face-launch-new-lineup-of-ipu-ready-transformers.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Sally Doherty
- Published: May 26, 2022
- Type: Announcement

Graphcore and Hugging Face expanded the range of machine learning modalities and tasks available in Hugging Face Optimum. Developers can now access ten transformer models optimized for Graphcore IPUs across natural language processing, speech, and computer vision. The available architectures include BERT, ViT, GPT-2, RoBERTa, DeBERTa, BART, LXMERT, T5, HuBERT, and Wav2Vec2, complete with IPU configuration files and ready-to-use pre-trained or fine-tuned weights. The integration supports the Bow IPU processor, which uses 3D Wafer-on-Wafer stacking to achieve up to 350 teraFLOPS of AI compute. Optimum also integrates with the Poplar SDK 2.5, enabling compatibility with frameworks such as PyTorch, TensorFlow, Docker, and Kubernetes.


### [Introducing Pull Requests and Discussions 🥳](https://yomu.fyi/post/introducing-pull-requests-and-discussions.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Victor Mustar
- Published: May 25, 2022
- Type: Announcement

Hugging Face Hub released pull requests and discussions under a new community tab for all repository types, including models, datasets, and Spaces. This feature enables community members and repository owners to interact directly, propose model card updates, report biased generations, and contribute files. The pull request implementation departs from traditional Git hosting platforms by avoiding forks and standard branches. Instead, contributors push to custom ref branches stored directly on the source repository to bypass creating heavy forks for large machine learning artifacts. Furthermore, the Hub treats discussions and pull requests without hard distinctions, listing them together to streamline collaboration across machine learning assets.


### [Efficient Table Pre-training without Real Data: An Introduction to TAPEX](https://yomu.fyi/post/efficient-table-pre-training-without-real-data-an-introduction-to-tape.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Qian Liu
- Published: May 23, 2022
- Type: Problem & solution

Conventional language model pre-training relies on massive natural text corpora and general-purpose objectives like masked language modeling, leaving a substantial gap when applied to structured table tasks. To bridge this gap efficiently, TAPEX (Table Pre-training via Execution) pre-trains a language model, such as BART, by learning a neural SQL executor over a synthetic corpus. The system pairs web tables with systematically sampled executable SQL queries and gathers target execution outputs via an off-the-shelf SQL engine. During downstream fine-tuning, the model consumes natural language questions and flattened tables to predict answers. TAPEX achieved new state-of-the-art results across four benchmark datasets, including WikiTableQuestions and TabFact, while achieving a pre-training speedup of nearly 50 times over TaBERT.


### [An Introduction to Q-Learning Part 2/2](https://yomu.fyi/post/an-introduction-to-q-learning-part-2-2.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: Thomas Simonini
- Published: May 20, 2022
- Type: Tutorial

Q-Learning serves as a foundational off-policy reinforcement learning method for finding optimal action-value functions before advancing to deep reinforcement learning. The approach employs a tabular Q-function that updates state-action pairs at each step using temporal difference learning. Action selection balances exploration and exploitation via an epsilon-greedy strategy, where epsilon progressively decays as training proceeds. Updates compute a temporal difference target combining the immediate reward and the discounted maximum value of the subsequent state under a greedy policy. This discrepancy between the exploratory acting policy and the greedy updating policy classifies Q-Learning as an off-policy algorithm, enabling agents to learn optimal policies in environments like Frozen Lake and autonomous taxi navigation.


### [How Sempre Health is leveraging the Expert Acceleration Program to accelerate their ML roadmap](https://yomu.fyi/post/how-sempre-health-is-leveraging-the-expert-acceleration-program-to-acc.md)
- Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md)
- Author: system
- Published: May 19, 2022
- Type: Problem & solution

Sempre Health handles thousands of inbound patient text messages daily to manage medication adherence and prescription discounts. Although an existing rule-based system caught around 80% of inbound texts, the team required statistical machine learning to improve message parsing and reduce manual workload. Collaborating with Hugging Face through the Expert Acceleration Program, the engineers received guidance on labeling strategies, problem formulation, and model selection. They deployed a natural language processing pipeline into production to automatically classify and respond to incoming messages or initiate automated actions like refill requests. Since deployment, the new NLP system automatically handles nearly 20% of incoming messages that previously required operational support tickets.


[Newer posts](https://yomu.fyi/company/hugging-face/page/11.md) · [Older posts](https://yomu.fyi/company/hugging-face/page/13.md)