AI PORTAL
  • Models
  • Code
  • Tools

About the models on this Arm AI Portal

The Arm AI Portal (the “Portal”) provides details of AI models developed, optimized or tested for use on Arm-based platforms. Models listed on the Portal may have been developed by Arm or by third parties. Unless expressly stated otherwise, Arm did not develop or train the underlying third-party model and is not responsible for its original intended behavior or use.

Information accompanies each model listing and describes the model’s provenance, together with relevant information about the original third-party model, its developer and repository, where applicable. Inclusion on the Portal of a model developed by a third party or based on a third-party model does not constitute Arm’s endorsement or certification of that third-party model or any third-party work relating to it.

Except as expressly stated in the information accompanying a specific model, Arm makes no representation as to the accuracy, safety, security, non-infringement, legal compliance, suitability for production use or fitness for any particular purpose of any model, converted or optimized version, or related outputs.

Inclusion of a model on the Portal does not itself grant you any rights to use that model. Use of each model is subject to the applicable license terms, usage restrictions and documentation. You are responsible for reviewing that information and independently evaluating the model, its outputs and its suitability for your intended use, including compliance with applicable legal, regulatory, safety and security requirements. Any use of or reliance on a model, its outputs or related materials is at your own risk. Where the Portal links to repositories or other materials controlled by third parties, those repositories and materials are outside Arm’s control and may change without notice.

Use of the Portal is subject to the Arm Website Terms and Conditions of Use.

Models

Filters

1

Device Class

Edge LinuxMobile CPUCloud CPU

Evaluation Target

Raspberry Pi 5vivo X300AWS Graviton G4

Arm Technology

SME2

Task

Text GenerationImage ClassificationObject DetectionFeature ExtractionImage-to-ImageAutomatic Speech RecognitionImage SegmentationDepth EstimationText-to-SpeechAudio ClassificationKeypoint DetectionImage-Text-to-TextText-to-ImageTranslationZero-Shot Image Classification

Vendor

Runtime

Quantization

Format

Execution Backend

Model Size

≤135 MB

Average Memory

≤1.4 GB

Peak Memory

≤1.49 GB
    17 results

    YOLO11n INT8

    Raspberry Pi 5

    This is an INT8-quantized version of YOLO11n optimized for edge deployment on ARM devices using ExecuTorch with the XNNPACK backend. The model was quantized using PT2E static symmetric per-channel quantization and exported to the .pte format for efficient inference on ARM Cortex-A processors (AWS Graviton, mobile ARM, embedded, Raspberry Pi).

    object-detection5 MB
    ExecuTorch

    YOLOX-s INT8

    Raspberry Pi 5

    This is an INT8-quantized version of YOLOX-s optimized for edge deployment on ARM devices using ExecuTorch with the XNNPACK backend. The model was quantized using PT2E static quantization and exported to the .pte format for efficient inference on ARM Cortex-A processors (AWS Graviton, mobile ARM, embedded).

    object-detection15 MBExecuTorch

    YOLOX-S INT8

    vivo X300

    An INT8-quantized YOLOX-S single-stage, anchor-free object detection model for locating and classifying objects in images. Exported as an ExecuTorch program using the XNNPACK backend for inference on Arm CPUs.

    object-detection9.7 MBExecuTorch

    YOLO26s INT8 (640x640)

    Raspberry Pi 5

    This is an INT8-quantized version of YOLO26s, Ultralytics' NMS-free, end-to-end object detector with a one2one Detect head. It is optimized for edge deployment on ARM devices using ONNX Runtime, whose CPU execution provider dispatches INT8 GEMMs to MLAS and KleidiAI on aarch64. The detection head is left quantizable; PT2E-forced shared-scale Q/DQ artifacts around concatenation and permute ops in the head are stripped after conversion to preserve mixed-range box/score outputs.

    object-detection11.3 MBONNX

    DETR ResNet-50 INT8

    Raspberry Pi 5

    An INT8-quantized DETR object detection model combining a ResNet-50 backbone with a transformer-based detection architecture. Exported as an ExecuTorch program using the XNNPACK backend for inference on Arm CPUs.

    object-detection135 MBExecuTorch

    YOLOv8s INT8

    Raspberry Pi 5

    An INT8-quantized single-stage object detector for 80 COCO classes, using an anchor-free architecture with 640×640 RGB inputs. It outputs bounding boxes and class scores for common objects, with postprocessing via confidence filtering, box conversion, and per-class non-maximum suppression.

    object-detection13.9 MBExecuTorch

    YOLOX-S INT8

    AWS Graviton G4

    An INT8-quantized YOLOX-S single-stage, anchor-free object detection model for locating and classifying objects in images. Exported as an ExecuTorch program using the XNNPACK backend for inference on Arm CPUs.

    object-detection9.7 MBExecuTorch

    YOLO26n INT8 (640x640)

    Raspberry Pi 5

    This is an INT8-quantized version of YOLO26n, Ultralytics' NMS-free, end-to-end object detector with a one2one Detect head. It is optimized for edge deployment on ARM devices using ONNX Runtime, whose CPU execution provider dispatches INT8 GEMMs to MLAS and KleidiAI on aarch64. The detection head is left quantizable; PT2E-forced shared-scale Q/DQ artifacts around concatenation and permute ops in the head are stripped after conversion to preserve mixed-range box/score outputs.

    object-detection4.2 MBONNX

    YOLOv5s INT8

    Raspberry Pi 5

    An INT8-quantized single-stage object detection model for 80 COCO classes, using an anchor-based architecture with 640×640 RGB inputs. It provides bounding boxes, objectness scores, and class probabilities, with postprocessing via confidence filtering, coordinate conversion, and per-class non-maximum suppression.

    object-detection7.5 MBExecuTorch

    YOLO12-L INT8

    Raspberry Pi 5

    This is an INT8-quantized version of YOLO12-L optimized for edge deployment on ARM devices using ExecuTorch with the XNNPACK backend. The model was quantized with PT2E static post-training quantization (symmetric INT8, per-channel weights) and exported to the .pte format for efficient inference on ARM Cortex-A processors (Raspberry Pi, AWS Graviton, mobile ARM, embedded).

    object-detection35.5 MBExecuTorch

    YOLO26n INT8 weight-only — Compute Flow

    vivo X300

    Object detection model bundle optimized for LiteRT edge inference, using INT8 weight-only quantization for Conv2D and Fully Connected layers while retaining floating-point activations and biases. It includes a detection manifest, COCO class labels, sample image, and reproducibility checksums, targeting efficient ARM and edge AI use cases.

    object-detection3.5 MBLiteRT

    Deformable DETR INT8 (ONNX Runtime)

    AWS Graviton G4

    An INT8 hybrid-quantized version of Deformable DETR, a transformer-based object detector with multi-scale deformable attention, for 80-class COCO-style detection from 640x640 RGB images on ONNX Runtime's CPU execution provider (MLAS NEON dotprod and i8mm kernels) on Arm server CPUs. The backbone convolutions are quantized statically per channel, the FFN matmuls dynamically, and the attention, normalization and prediction-head layers stay in floating point.

    object-detection57.7 MBONNX

    rtdetr l INT8

    Raspberry Pi 5

    This is an INT8-quantized version of RT-DETR-L, a real-time end-to-end object detector that pairs a convolutional backbone with a transformer encoder-decoder and predicts a fixed set of objects without non-maximum suppression. It is optimized for edge deployment on ARM devices using ONNX Runtime, whose CPU execution provider dispatches INT8 GEMMs to MLAS and KleidiAI on aarch64. Quantization is applied per region: the convolutional backbone and neck are quantized statically against calibration images, the transformer's matrix multiplications dynamically, and the prediction heads, deformable-attention sampling components and the first stem convolution are left in floating point.

    object-detection40.3 MBONNX

    UltraFace RFB-320 INT8

    Raspberry Pi 5

    This is an INT8-quantized version of UltraFace RFB-320, an Ultra-Light-Fast-Generic-Face-Detector built on an SSD-style architecture with an RFB backbone. Every convolution, including the detection heads, is quantized statically and exported to ONNX Runtime for edge deployment on ARM CPUs, where the CPU execution provider dispatches quantized convolutions to MLAS and KleidiAI on aarch64.

    object-detection798 KBONNX

    YOLO11n INT8

    vivo X300SME2

    This is an INT8-quantized version of YOLO11n optimized for edge deployment on ARM devices using LiteRT with the XNNPACK backend and KleidiAI acceleration. The model was quantized using post-training static INT8 quantization via the AI Edge Quantizer and exported to the .tflite format for efficient inference on ARM Cortex-A processors (mobile ARM, embedded).

    object-detection4.6 MBLiteRT

    YOLO26n FP16 — Compute Flow

    vivo X300

    Compact object detection model bundle in LiteRT/TFLite format, optimized for FP16 Conv2D execution via XNNPACK on ARM and edge devices. It includes a detection manifest, COCO class labels, sample image, and reproducibility checksums, supporting efficient detection across standard COCO object categories.

    object-detection10.9 MBLiteRT

    YOLOv9-S INT8

    Raspberry Pi 5

    An INT8-quantized single-stage object detection family for efficient edge inference on ARM devices. It targets 80-class COCO-style detection with 640×640 RGB inputs, producing bounding boxes and class scores for postprocessing with thresholding and NMS. The quantized design reduces model size, latency, and memory use while retaining strong detection accuracy, making it suitable for resource-constrained edge AI applications.

    object-detection12.4 MBExecuTorch

    Develop on Arm with your AI coding assistant

    The Arm AI Portal MCP Server makes it faster to find optimized models, integrate code examples and deploy to your target device from your agentic coding assistant.

    User guide

    Run the following command in your terminal:

    codex mcp add arm-ai --url https://mcp.api.devplatform.arm.com/ai-portal