Python Bindings for ARA-2

Python bindings for the ARA-2 neural network accelerator client library, providing efficient NPU inference from Python via a proxy service running on NXP i.MX platforms with Kinara ARA-2 hardware.

Published to PyPI as edgefirst-ara2.

Architecture

Python Application ──(UNIX/TCP socket)──▶ ara2-proxy ──(PCIe)──▶ ARA-2 NPU
       │                                (system service)        (Kinara hardware)
       │
edgefirst-hal ──(DMA-BUF fd)──▶ GPU preprocessing (zero-copy)

Your Python code connects to the dvproxy system service (not directly to the hardware). The proxy manages device access and must be running before your application starts. The systemd unit name is platform-dependent: ara2.service on EdgeFirst Yocto images, dvproxy.service on other platforms.

Installation

From PyPI

pip install edgefirst-ara2

For zero-copy preprocessing with edgefirst-hal:

pip install edgefirst-ara2[hal]

Prerequisites for Development

• Python 3.11 or higher
• Rust stable toolchain (edition 2024)
• maturin (pip install maturin)
• ARA-2 client library (libaraclient.so.1)

Development Install

cd crates/ara2-py
maturin develop --release --features abi3

Quick Start

import edgefirst_ara2

# Connect to ARA-2 proxy
session = edgefirst_ara2.Session.create_via_unix_socket("/var/run/ara2.sock")

# Get version information
versions = session.versions()
print(f"Proxy version: {versions['proxy']}")

# List endpoints
endpoints = session.list_endpoints()
print(f"Found {len(endpoints)} endpoints")

# Check endpoint status
for endpoint in endpoints:
    state = endpoint.check_status()
    stats = endpoint.dram_statistics()
    print(f"State: {state}, Free DRAM: {stats.free_size / stats.dram_size * 100:.1f}%")

Inference with numpy

import numpy as np
import edgefirst_ara2

session = edgefirst_ara2.Session.create_via_unix_socket("/var/run/ara2.sock")
endpoints = session.list_endpoints()
model = endpoints[0].load_model("model.dvm")

# Allocate tensors and run inference
model.allocate_tensors()
input_data = np.zeros(model.input_size(0), dtype=np.uint8)
model.set_input_tensor(0, input_data)
timing = model.run()

print(f"Inference: {timing.run_time_us} us")
output = model.get_output_tensor(0)
dequantized = model.dequantize(0)

Zero-Copy DMA-BUF Pipeline

For maximum throughput, use DMA-BUF tensors with edgefirst-hal for GPU-accelerated preprocessing. This eliminates CPU memory copies between preprocessing and inference:

How it works: allocate_tensors("dma") allocates the model's input tensor in a DMA-BUF — a Linux kernel buffer accessible by multiple hardware devices. input_tensor_fd(0) returns a file descriptor to that buffer. You pass this FD to edgefirst_hal.import_image(), which maps it as a GPU image surface. The GPU writes the preprocessed frame directly into the NPU's input buffer — no CPU copies involved.

import os
import edgefirst_ara2 as ara2
import edgefirst_hal as hal

session = ara2.Session.create_via_unix_socket(ara2.DEFAULT_SOCKET)
endpoint = session.list_endpoints()[0]

with endpoint.load_model("yolov8s.dvm") as model:
    model.allocate_tensors("dma")  # Must use "dma" for tensor FD access

    # Get DMA-BUF FD for the model's input tensor
    input_fd = model.input_tensor_fd(0)
    c, h, w = model.input_shape(0)
    try:
        # Import as PlanarRgb (CHW layout) to match ARA-2 tensor format
        dst = hal.import_image(input_fd, w, h, hal.PixelFormat.PlanarRgb)
    finally:
        os.close(input_fd)  # FD duplicated by import_image; close original

    # GPU-accelerated convert: camera frame -> model input (zero CPU copies)
    processor = hal.ImageProcessor()
    src = hal.load_image("image.jpg", format=hal.PixelFormat.Rgba, mem=hal.TensorMemory.DMA)
    processor.convert(src, dst)

    # Run inference — NPU reads from the same DMA-BUF
    timing = model.run()
    print(f"Inference: {timing.run_time_us} us")

Performance

Benchmarked on NXP FRDM i.MX 95 + ARA-2 with YOLOv8m-seg (640×640). The Python API adds minimal overhead over native Rust thanks to DMA-BUF zero-copy — GPU and NPU operate on the same physical memory buffers.

Steady-state mean over 30 iterations after warmup. Python overhead is under 1 ms across the entire pipeline.

Run the benchmark yourself. Create a virtual environment on the target and install the packages from PyPI:

python3 -m venv ~/venv
~/venv/bin/pip install edgefirst-ara2 edgefirst-hal
~/venv/bin/python3 yolov8.py model.dvm image.jpg --benchmark 30 --save

DVM Metadata

Read model metadata without loading onto the NPU:

import edgefirst_ara2

metadata = edgefirst_ara2.read_metadata("model.dvm")
if metadata:
    print(f"Task: {metadata.task}")
    print(f"Classes: {metadata.classes}")
    if metadata.compilation and metadata.compilation.ppa:
        print(f"IPS: {metadata.compilation.ppa.ips}")

labels = edgefirst_ara2.read_labels("model.dvm")

Async Inference

The submit() / wait() API enables overlapping CPU work with NPU execution. This is the building block for pipeline parallelism — while the NPU runs inference on one frame, the CPU can preprocess the next.

import edgefirst_ara2 as ara2

session = ara2.Session.create_via_unix_socket(ara2.DEFAULT_SOCKET)
endpoint = session.list_endpoints()[0]
model = endpoint.load_model("model.dvm")
model.allocate_tensors()

# Submit — returns immediately while NPU works
request = model.submit()
print(f"Request #{request.request_id} submitted")

# CPU is free to do other work here...
# The GIL is NOT held during wait(), so other Python threads can run

timing = request.wait()  # blocks until NPU finishes
print(f"NPU inference: {timing.run_time_us} µs")

# Monitor pipeline depth
print(f"In-flight: {session.inflight_count()}")

Warning: Do not call model.allocate_tensors() while an InferRequest is still pending — the NPU is actively reading/writing the tensor buffers.

API Reference

Session

Connection to the ARA-2 proxy service.

Static Methods:

• create_via_unix_socket(socket_path: str) -> Session
• create_via_tcp_ipv4_socket(ip: str, port: int) -> Session

Methods:

• versions() -> dict[str, str] - Get component versions
• list_endpoints() -> list[Endpoint] - List available endpoints
• inflight_count() -> int - Number of pending async inference requests

Properties:

• socket_type: str - "unix" or "tcp"

Endpoint

Represents an ARA-2 accelerator device.

Methods:

• check_status() -> State - Get device state
• dram_statistics() -> DramStatistics - Get memory usage
• load_model(model_path: str) -> Model - Load a .dvm model

Model

Loaded neural network model.

Lifecycle:

• allocate_tensors(memory: str | None = None) - Allocate tensors ("dma", "shm", "mem", or None)
• set_timeout_ms(timeout_ms: int) - Set inference timeout
• run() -> ModelTiming - Execute inference synchronously
• submit() -> InferRequest - Submit inference asynchronously (returns immediately)

Tensor I/O (numpy):

• set_input_tensor(index: int, data: np.ndarray) - Copy data into input
• get_output_tensor(index: int) -> np.ndarray - Copy output data out
• dequantize(index: int) -> np.ndarray - Dequantize output to float32

DMA-BUF Zero-Copy:

• input_tensor_fd(index: int) -> int - Get input tensor FD (pass to hal.ImageProcessor.import_image, which dups it — close after)
• output_tensor_fd(index: int) -> int - Get output tensor FD (pass to hal.Tensor.from_fd, which takes ownership — do not close after)
• input_tensor_memory(index: int) -> str - Input memory type
• output_tensor_memory(index: int) -> str - Output memory type

Introspection:

• n_inputs: int, n_outputs: int - Tensor counts
• input_shape(i) -> (C, H, W), output_shape(i) -> (C, H, W)
• input_size(i) -> int, output_size(i) -> int - Size in bytes
• input_bpp(i) -> int, output_bpp(i) -> int - Bytes per element
• input_info(i) -> InputTensorInfo, output_info(i) -> OutputTensorInfo
• input_quants(i) -> InputQuantization, output_quants(i) -> OutputQuantization

InferRequest

Pending asynchronous inference request, created by Model.submit().

Methods:

• wait(timeout_ms: int = 1000) -> ModelTiming - Block until complete (GIL released)

Properties:

• request_id: int - Proxy-assigned ID for log correlation

Metadata Functions

• read_metadata(path: str) -> DvmMetadata | None
• read_labels(path: str) -> list[str]
• has_metadata(path: str) -> bool

Supporting Types

• State (enum): Init, Idle, Active, ActiveSlow, ActiveBoosted, ThermalInactive, ThermalUnknown, Inactive, Fault
• ModelOutputType (enum): Classification, Detection, SemanticSegmentation, Raw
• DramStatistics: dram_size, free_size, model_occupancy_size, ...
• ModelTiming: run_time_us, input_time_us, output_time_us
• InputQuantization: qn, scale, mean, is_signed
• OutputQuantization: qn, scale, offset, is_signed

Exceptions

Ara2Error (RuntimeError)
 +-- LibraryError       - libaraclient.so loading failures
 +-- HardwareError      - NPU faults, endpoint errors
 +-- ProxyError         - Proxy connection failures
 +-- ModelError         - Model load/inference failures
 +-- TensorError        - Tensor allocation, DMA-BUF errors
 +-- MetadataError      - DVM metadata parsing errors

Building Wheels

cd crates/ara2-py
maturin build --release --features abi3

Wheels are created in target/wheels/.

Stable ABI

The bindings use PyO3's stable ABI (abi3-py311):

• A single wheel works across Python 3.11, 3.12, 3.13, and future versions
• Minimum supported Python version is 3.11

Troubleshooting

"libaraclient.so.1 not found"

export LD_LIBRARY_PATH=/path/to/ara2/lib:$LD_LIBRARY_PATH

Verify Installation

python -c "import edgefirst_ara2; print(edgefirst_ara2.__version__)"

License

Licensed under the Apache License 2.0.

Copyright 2025 Au-Zone Technologies. All Rights Reserved.