Model Zoo — Pre-trained Models

The Voyager Model Zoo is a collection of pre-trained AI models ready to run on Axelera hardware. When you run inference.py yolov5s-v7-coco usb:0, the model name (yolov5s-v7-coco) comes from the Model Zoo.

• Model Zoo — Pre-trained Models
  • Listing available models
  • How model names work
  • Convolutional Neural Networks (CNNs)
    • Task types
    • Running a CNN model
    • Datasets
    • Non-redistributable datasets
    • CNN Performance characteristics
    • Image Classification
    • Object Detection
    • Semantic Segmentation
    • Instance Segmentation
    • Keypoint Detection
    • Depth Estimation
    • License Plate Recognition
    • Face Recognition
    • Re-identification
  • Vision Transformers (ViTs)
    • ViT Performance characteristics
    • ViT Reference Platform
  • Vision Language Models (VLMs)
    • VLM Performance characteristics
    • VLM Reference platforms
  • Large Language Models (LLMs)
    • LLM Performance characteristics
    • LLM Reference platform
  • Experimenting with optimized input shapes
    • How to test
    • Expected results
  • Custom weights
  • See also

Listing available models​

From the SDK root directory (with environment activated):

make

This lists three categories:

How model names work​

Model names follow a pattern:

<architecture>-<dataset>[-<variant>]

Examples:

Convolutional Neural Networks (CNNs)​

Task types​

Running a CNN model​

# Object detection with USB camera
./inference.py yolov5s-v7-coco usb:0

# Classification with a video file
./inference.py resnet50-imagenet media/traffic1_1080p.mp4

# Headless benchmarking (no display)
./inference.py yolov8s-coco-onnx usb:0 --no-display --frames 1000

The first time you run a model, the SDK:

1. Downloads the pre-trained weights (if not cached)
2. Compiles the model for the AIPU
3. Caches the compiled model for subsequent runs
4. Runs inference

Subsequent runs skip steps 1-3 and start immediately.

Datasets​

Models are trained on specific datasets. The dataset name in the model identifier tells you what the model can recognize:

Non-redistributable datasets​

Most datasets download automatically when you first run a model. A few datasets require manual registration and download due to licensing restrictions. These must be downloaded by hand from the links below, then placed in the specified directory within your SDK installation. The SDK raises an error with the expected path if a required dataset is missing.

You are responsible for adhering to the terms and conditions of each dataset's license.

CNN Performance characteristics​

The tables below list all Model Zoo models for this SDK release. Columns:

• Ref FP32 — accuracy of the original floating-point model
• Accuracy loss — FP32 accuracy minus quantized int8 accuracy (lower is better)
• Ref PCIe FPS — host throughput on Intel Core Ultra 7 265 + Axelera Edge 130p (1× PCIe) card
• Ref M.2 FPS — host throughput on Intel Core i5-1145G7E + Axelera Embedded 113m (1× M.2) card with the default power limit of 8.5W

Accuracy is measured using:

./inference.py <model> dataset --pipe=torch-aipu --no-display

Throughput is measured using a 720p h.264 video file:

./inference.py <model> media/traffic2_720p.mp4 --pipe=gst --no-display

Image Classification​

Object Detection​

Semantic Segmentation​

Instance Segmentation​

Keypoint Detection​

Depth Estimation​

License Plate Recognition​

Face Recognition​

Re-identification​

Vision Transformers (ViTs)​

ViT Performance characteristics​

• Resolution — input resolution of the model.
• benchmark ms / fps — latency per frame and throughput.
• License — model license.

Latency and throughput are measured with:

python -m model_axscript.vision benchmark <model> <image>

ViT Reference Platform​

• PCIe — Intel Core i7-13700 32GB RAM running Ubuntu 24.04.5 with 1x Axelera Edge 130p PCIe card.

Vision Language Models (VLMs)​

VLM Performance characteristics​

• Resolution — input image resolution of the vision encoder.
• Ctx — maximum context window, in tokens.
• Prompt tokens — prompt length used for the measurement
• PCIe / M.2 TTFT (s) — time to first token, in seconds (lower is better).
• PCIe / M.2 Decode (tok/s) — token generation rate after the first token (higher is better).
• PCIe / M.2 Vision (fps) — vision encoder throughput, frames per second.
• License — model license.

TTFT, decode rate and vision-encoder throughput are measured with the model's default prompt:

python -m model_axscript.llm benchmark <model>

To run a VLM, see Serve an AxScript LLM through HuggingFace.

VLM Reference platforms​

• PCIe — Intel Core Ultra 9 285 128GB RAM running Ubuntu 24.04 with 1x Axelera Edge 16GB 130p PCIe card.
• M.2 — Intel Core Ultra 5 125H 32GB RAM running Ubuntu 24.04 with 1x Axelera Embedded 113m M.2 card.
• — — Model not supported on this form factor.

Large Language Models (LLMs)​

LLM Performance characteristics​

• Ctx — maximum context window, in tokens.
• Prompt tokens — prompt length used for the measurement.
• PCIe TTFT (s) — time to first token, in seconds (lower is better).
• PCIe Decode (tok/s) — token generation rate after the first token (higher is better).
• License — model license.

TTFT and decode rate are measured with the model's default prompt:

python -m model_axscript.llm benchmark <model>

To run an LLM, see Serve an AxScript LLM through HuggingFace.

LLM Reference platform​

PCIe — Intel Core Ultra 9 285 (24 cores) 128GB RAM running Ubuntu 24.04 with 1x Axelera Edge 16GB 130p PCIe card.

Experimenting with optimized input shapes​

Most models are trained on square inputs (640×640), but real-world video is often rectangular (16:9). Standard pipelines pad the input ("letterboxing"), forcing the model to process empty pixels.

By switching to a rectangular input shape that matches your video's aspect ratio, you can often achieve significant speedups with minimal accuracy impact. This is especially effective for fixed-camera applications like surveillance or traffic monitoring.

How to test​

Export models with dynamic input shapes, then compare:

# Standard 640×640
./inference.py yolox-m-coco-onnx dataset --pipe=torch-aipu --no-display

# Rectangular 640×480
./inference.py yolox-m-coco-onnx-rect dataset --pipe=torch-aipu --no-display

Expected results​

Custom weights​

You can use your own trained weights with any model architecture. This involves updating the model's YAML configuration to point to your custom weight file. See Deploy Custom Weights for the full walkthrough.

See also​

• First Inference — run your first model
• Measure Accuracy — benchmark model performance
• inference.py — full command reference
• Glossary — definitions of COCO, YOLO, mAP, and other terms