Accelerate AI model deployment from cloud to edge

Arm on Hugging Face helps developers deploy Hugging Face models faster with optimized performance on Arm-based devices and platforms. Our guides, tools, and learning paths show how Arm integrates with major operating systems and frameworks, making it easier to build, optimize, and scale AI models across real-world use cases from cloud to edge, gaming to mobile.

Follow our curated Learning Paths to:

Explore Arm-optimized AI models available in our Hugging Face Model Collections
Use libraries and ML frameworks like PyTorch, ExecuTorch, llama.cpp, ONNX Runtime, and KleidiAI.
Streamline your journey - from discovery to deployment – across AI use cases like real-time chatbots, sentiment analysis, neural graphics, object detection and more.

What can I build with Arm on Hugging Face?

Explore curated learning paths using Hugging Face models, optimised to run on platforms like Raspberry Pi, smartphones, and Arm-based cloud servers.

Neural Graphics

Learning Path	Frameworks & Tools Used	Model(s) Featured	Market Application	Examples	Arm Learning Path
Neural Super Sampling in Unreal Engine	NSS Plugin for Unreal® Unreal® NNE Plugin for ML extensions for Vulkan Neural Graphics Model Gym	Neural Super Sampling (NSS)	Smartphone	Graphics upscaling Enchanted Castle Demo	Run NSS in Unreal →

Generative AI

Learning Path	Frameworks & Tools Used	Model(s) Featured	Market Application	Examples	Arm Learning Path
Build a RAG application	Zilliz Cloud, llama.cpp	All MiniLM L6 V2	Cloud & Datacenter	Document retrieval + Q&A pipelines	Build with Zilliz →
Accelerate NLP models for faster inference	PyTorch, KleidiAI	DistilBERT Base Uncased SST-2	Cloud & Datacenter	Sentiment analysis, text classification	Accelerate NLP →
Deploy an LLM chatbot with optimised performance	llama.cpp, KleidiAI	Dolphin 2.9.4, Llama 3.1 8B GGUF	Cloud & Datacenter	Real-time chatbots, enterprise assistants	Deploy with llama.cpp →
Run an LLM chatbot with PyTorch	PyTorch, Torchchat, Streamlit, KleidiAI	Llama 3.1 8B Instruct	Cloud & Datacenter	Inference pipelines with PyTorch	Run with PyTorch →
Deploy a RAG chatbot on Google Axion processors	llama-cpp-python, Faiss, KleidiAI,	Llama 3.1 8B GGUF	Cloud & Datacenter	RAG-based assistants at cloud scale	Deploy with Axion →
Build an Android chat app	ExecuTorch, XNNPACK, KleidiAI	Llama 3.2 1B Instruct	Smartphone	On-device chat apps	Build on Android →
Run Llama 3 on Raspberry Pi 5	ExecuTorch	Llama 3.1 8B	Raspberry Pi	Edge LLM deployment	Run Llama 3 on Pi 5 →

CV: Image Classification & Object Detection

Learning Path	Frameworks & Tools Used	Model(s) Featured	Market Application	Examples	Arm Learning Path
Profile AI/ML performance on mobile apps	ExecuTorch, Arm Performance Studio, Android Studio Profiler	MobileNet V2 1.0 224	Smartphone	App performance benchmarking	Profile mobile apps →
Run CV models on microcontrollers	Himax MCU, Arm toolchain	YOLOv8	IoT	Object detection on MCUs	Run on MCU →
Export PyTorch models for edge devices	PyTorch, ExecuTorch	DistilBERT Base Uncased SST-2	IoT	Deploy compact AI models on MCUs	Export with ExecuTorch →

Sentiment Analysis

Learning Path	Frameworks & Tools Used	Model(s) Featured	Market Application	Examples	Arm Learning Path
Accelerate NLP models from Hugging Face on Arm servers	PyTorch	DistilBERT Base Uncased SST-2	Cloud & Datacenter	Text classification, sentiment analysis	Accelerate NLP on Arm →

Speed Up AI Model Inference with Arm Kleidi

Arm Kleidi, comprising KleidiAI and KleidiCV, delivers out-of-the-box AI acceleration across popular frameworks – such as Pytorch, llama.cpp, MediaPipe (via XNNPACK), ONNX Runtime, and more – by integrating highly optimised micro-kernels tailored to Arm CPU architectures.

These lightweight libraries use advanced Arm instructions like Neon, SVE, and SME to deliver faster inference - with no code changes, retraining, or extra tooling. Developers get immediate performance gains while continuing to use familiar frameworks.

What You Can Do with Arm Kleidi:

Accelerate Hugging Face models on real hardware
Boost performance for computer vision, NLP, and generative AI workloads
Use your existing models - no retraining required
Integrate with familiar frameworks and runtimes
Optimise for cloud, mobile, edge, and microcontroller platforms

Key Resources:

Arm KleidiAI GitLab repo – Supports general-purpose AI acceleration
Arm KleidiCV GitLab repo – Optimisation for computer vision models
Arm Compute Library on GitHub – Low-level acceleration for AI software

Get started

Note: The data collated here is sourced from Arm and third parties. While Arm uses reasonable efforts to keep this information accurate, Arm does not warrant (express or implied) or provide any guarantee of data correctness due to the ever-evolving AI and software landscape. Any links to third-party sites and resources are provided for ease and convenience. Your use of such third-party sites and resources is subject to the third party’s terms of use, and use is at your own risk.