### Install Model Server from Source

Source: https://developer.furiosa.ai/docs/latest/en/software/python-sdk

Install the model server extra package from the source code. This involves navigating to the appropriate subdirectory within the cloned furiosa-sdk repository and installing the 'furiosa-server' package.

```bash
cd furiosa-sdk/python
pip install furiosa-server

```

--------------------------------

### Install FuriosaAI Server from Source Code

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Installs the model server functionality from the FuriosaAI SDK source code. This involves navigating to the Python subdirectory and installing the 'furiosa-server' package.

```sh
cd furiosa-sdk/python
pip install furiosa-server
```

--------------------------------

### Install C SDK using APT - Shell

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/c-sdk.rst

Installs the FuriosaAI C SDK package 'furiosa-libnux-dev' using the APT package manager. This command first updates the package list and then installs the specified package. Ensure the FuriosaAI APT repository is set up prior to execution.

```sh
apt-get update && apt-get install -y furiosa-libnux-dev
```

--------------------------------

### Install FuriosaAI Python SDK from Source Code

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Installs the FuriosaAI Python SDK by cloning the repository and installing packages in a specific order. This method is useful for development or when the latest unreleased versions are needed.

```sh
git clone https://github.com/furiosa-ai/furiosa-sdk
cd furiosa-sdk/python
pip install furiosa-runtime
pip install furiosa-tools
pip install furiosa-sdk
```

--------------------------------

### Install Model Server Package

Source: https://developer.furiosa.ai/docs/latest/en/software/python-sdk

Install the model server package, which provides functionality to accelerate DNN models using the NPU and serve them via GRPC or Restful API.

```bash
pip install 'furiosa-sdk[server]'

```

--------------------------------

### Install Furiosa Python SDK with Extra Packages

Source: https://developer.furiosa.ai/docs/latest/en/software/python-sdk

Install the Furiosa Python SDK with specific extra packages for additional functionality. For example, to install packages for model serving ('server') and model-SDK compatibility checking ('litmus').

```bash
pip install 'furiosa-sdk[server, litmus]'

```

--------------------------------

### Install furiosa-bench using apt

Source: https://developer.furiosa.ai/docs/latest/en/_sources/releases/0.10.0.rst

Installs the `furiosa-bench` tool using the apt package manager. This is a straightforward command for setting up benchmarking utilities.

```bash
$ apt install furiosa-bench
```

--------------------------------

### Install C SDK using Downloaded DEB Package - Shell

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/c-sdk.rst

Installs the FuriosaAI C SDK from a downloaded .deb package. Replace 'x.y.z' with the actual version number. This method requires manual download from the FuriosaAI download center. Ensure all prerequisites are met before installation.

```sh
apt-get install -y ./furiosa-libnux-dev-x.y.z-?.deb
```

--------------------------------

### Update Pip, Setuptools, and Wheel

Source: https://developer.furiosa.ai/docs/latest/en/software/python-sdk

Before installing the furiosa-sdk, it is recommended to update Python's package installer (pip) and related tools to their latest versions to avoid potential installation errors.

```bash
pip install --upgrade pip setuptools wheel

```

--------------------------------

### Upgrade Pip and Install SDK

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Commands to upgrade the Python package installer (pip) and install the FuriosaAI SDK. It's crucial to upgrade pip first to avoid potential installation errors.

```sh
pip install --upgrade pip setuptools wheel
```

--------------------------------

### Install FuriosaAI Python SDK with Extra Packages (PIP)

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Installs the FuriosaAI Python SDK with specified extra packages for extended functionality. For example, installing 'server' for model serving and 'litmus' for model compatibility checks.

```sh
pip install 'furiosa-sdk[server, litmus]'
```

--------------------------------

### Install furiosa-compiler

Source: https://developer.furiosa.ai/docs/latest/en/software/compiler

Installs the `furiosa-compiler` command-line tool using the APT package manager. This is the primary method for obtaining the compiler.

```bash
$ apt install furiosa-compiler

```

--------------------------------

### Compile C Application with C SDK - Shell

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/c-sdk.rst

Compiles a C application ('example.c') using the GCC compiler and links it with the FuriosaAI C SDK library ('-lnux'). The C header files and static libraries are expected to be in the default system paths. This command assumes the C SDK has been correctly installed.

```sh
gcc example.c -lnux
```

--------------------------------

### Install Specific Package Version (Shell)

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/installation.rst

This command installs a specific version of a package. First, you list available versions using 'apt list -a <package-name>', then you specify the desired version in the 'apt-get install' command. The '-y' flag automatically confirms the installation.

```sh
sudo apt list -a furiosa-libnux
```

```sh
sudo apt-get install -y furiosa-libnux=0.9.1-?
```

--------------------------------

### Install FuriosaAI Server Package (PIP)

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Installs the FuriosaAI Server package, enabling the acceleration of DNN models with the NPU and serving them via GRPC or Restful API. This is installed as an extra package for the SDK.

```sh
pip install 'furiosa-sdk[server]'
```

--------------------------------

### Real Example: Model Input Type Optimization (Python)

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/performance.rst

Demonstrates a practical application of the ModelEditor API, specifically using the convert_input_type method to optimize model input types. This example showcases how to integrate the optimization into a workflow.

```python
import onnx
from furiosa_quantizer.onnx.importer.model_editor import ModelEditor, TensorType

def optimize_model_input_type(model_path: str, output_path: str):
    model = onnx.load_model(model_path)
    # Assuming optimize_model is defined elsewhere and returns an optimized model
    # model = optimize_model(model)

    editor = ModelEditor(model)
    input_tensor_names = editor.get_pure_input_names()

    if input_tensor_names:
        # Convert the first input tensor to UINT8
        editor.convert_input_type(input_tensor_names[0], TensorType.UINT8)
        print(f"Converted input tensor '{input_tensor_names[0]}' to UINT8.")
    else:
        print("No input tensors found to convert.")

    # Save the modified model
    editor.save(output_path)
    print(f"Optimized model saved to {output_path}")

# Example usage:
# optimize_model_input_type("path/to/your/model.onnx", "path/to/optimized_model.onnx")
```

--------------------------------

### Install Quantizer Package for Model Quantization

Source: https://developer.furiosa.ai/docs/latest/en/software/python-sdk

Install the quantizer package, which offers a set of APIs for converting models into a quantized format. This is essential for optimizing models for the Furiosa NPU.

```bash
pip install 'furiosa-sdk[quantizer]'

```

--------------------------------

### Install FuriosaAI SDK with Legacy Runtime/API (PIP)

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Installs the FuriosaAI SDK with the legacy runtime and API, which was adopted before version 0.10.0. This is useful for compatibility with older projects or specific requirements.

```sh
pip install 'furiosa-sdk[legacy]'
```

--------------------------------

### Install furiosa-toolkit using APT

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/cli.rst

Installs the `furiosa-toolkit` package using the APT package manager. This is a prerequisite for using other `furiosactl` commands. Ensure the kernel driver is installed beforehand.

```sh
sudo apt-get install -y furiosa-toolkit
```

--------------------------------

### Install FuriosaAI Litmus Package (PIP)

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Installs the FuriosaAI Litmus tool, which verifies model compatibility with the Furiosa SDK. It simulates processes like model quantization and compilation to ensure smooth execution.

```sh
pip install 'furiosa-sdk[litmus]'
```

--------------------------------

### Install Furiosa SDK with Legacy Runtime/API

Source: https://developer.furiosa.ai/docs/latest/en/software/python-sdk

Install the Furiosa Python SDK with the legacy runtime and API. This option is available for compatibility with older versions or specific project requirements.

```bash
pip install 'furiosa-sdk[legacy]'

```

--------------------------------

### Install APT Packages for HTTPS Access

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/installation.rst

Installs essential packages required to access HTTPS-based APT repositories. These include tools for updating package lists, installing packages, managing GPG keys, and transferring files.

```sh
sudo apt update
sudo apt install -y ca-certificates apt-transport-https gnupg wget
```

--------------------------------

### Model Summary Example

Source: https://developer.furiosa.ai/docs/latest/en/_sources/api/python/furiosa.runtime.rst

This example demonstrates how to print a human-readable summary of a model using the `print_summary` method. The output typically includes details about the input and output tensors, such as their shapes, data types, formats, sizes, and lengths.

```python
print(model.summary())
# Expected output format:
# Inputs:
# {0: TensorDesc(shape=(1, 1, 28, 28), dtype=FLOAT32, format=NCHW, size=3136, len=784)}
# Outputs:
# {0: TensorDesc(shape=(1, 10), dtype=FLOAT32, format=??, size=40, len=10)}
```

--------------------------------

### Install FuriosaAI Packages using APT

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/installation.rst

Installs the NPU kernel driver and runtime library using the APT package manager. This command updates the package list and then installs the specified FuriosaAI packages.

```sh
sudo apt-get update && sudo apt-get install -y furiosa-driver-warboy furiosa-libnux
```

--------------------------------

### Install furiosa-toolkit using apt

Source: https://developer.furiosa.ai/docs/latest/en/_sources/releases/0.10.0.rst

Installs the `furiosa-toolkit` package using the apt package manager. This command ensures that the toolkit and its associated tools are available on the system.

```bash
apt install furiosa-toolkit
```

--------------------------------

### Install Furiosa SDK with Serving Support

Source: https://developer.furiosa.ai/docs/latest/en/_sources/releases/0.6.0.rst

Command to install the Furiosa SDK with support for the Furiosa Serving library, which is based on FastAPI for advanced model serving. This installation includes all components needed for serving functionalities.

```sh
$ pip install 'furiosa-sdk[serving]'
```

--------------------------------

### Install FuriosaAI Packages from .deb Files

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/installation.rst

Installs FuriosaAI packages by directly referencing downloaded .deb files. This method requires specifying the exact filenames, including version numbers, for the driver, firmware, runtime, and onnxruntime.

```sh
sudo apt-get install -y ./furiosa-driver-warboy-x.y.z-?.deb
sudo apt-get install -y ./furiosa-libhal-warboy-x.y.z-?.deb
sudo apt-get install -y ./libonnxruntime-x.y.z-?.deb
sudo apt-get install -y ./furiosa-libnux-x.y.z-?.deb
```

--------------------------------

### Real Example: Model Output Type Optimization (Python)

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/performance.rst

Provides a concrete implementation for using the convert_output_type method of ModelEditor. This example demonstrates how to modify model outputs to INT8 or UINT8 to potentially improve inference performance.

```python
import onnx
from furiosa_quantizer.onnx.importer.model_editor import ModelEditor, TensorType

def optimize_model_output_type(model_path: str, output_path: str):
    model = onnx.load_model(model_path)
    # Assuming optimize_model is defined elsewhere and returns an optimized model
    # model = optimize_model(model)

    editor = ModelEditor(model)
    output_tensor_names = editor.get_pure_output_names()

    if output_tensor_names:
        # Convert the first output tensor to UINT8
        editor.convert_output_type(output_tensor_names[0], TensorType.UINT8)
        print(f"Converted output tensor '{output_tensor_names[0]}' to UINT8.")
    else:
        print("No output tensors found to convert.")

    # Save the modified model
    editor.save(output_path)
    print(f"Optimized model saved to {output_path}")

# Example usage:
# optimize_model_output_type("path/to/your/model.onnx", "path/to/optimized_model.onnx")
```

--------------------------------

### Install Furiosa Model Server from Source

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/serving.rst

Installs the Furiosa Model Server by cloning the source code repository and running the pip install command. This method is useful for developers who need to modify or contribute to the server's codebase.

```sh
$ git clone https://github.com/furiosa-ai/furiosa-sdk.git
$ cd furiosa-sdk/python/furiosa-server
$ pip install .
```

--------------------------------

### Install Specific furiosa-libnux Version with apt-get

Source: https://developer.furiosa.ai/docs/latest/en/software/installation

This command installs a specific version of `furiosa-libnux`. The `-y` flag automatically confirms the installation. Users must replace `0.9.1-?` with the exact version string obtained from `apt list`.

```bash
sudo apt-get install -y furiosa-libnux=0.9.1-?
```

--------------------------------

### Async/Await Syntax for Serving Applications (Python)

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/performance.rst

Shows how to implement a serving application using the async/await API provided by `NPUModel` from `furiosa-server`. This approach simplifies handling multiple concurrent requests asynchronously, improving serving performance.

```python
from furiosa.server.model import NPUModel

# Assuming 'model_path' is the path to your model
model = NPUModel(model_path)

async def handle_request(request):
    # Process request and prepare input data
    input_data = ...
    # Run inference asynchronously
    outputs = await model.run(input_data)
    # Process outputs and return response
    return ...

# Example of how to use it within an async web framework (e.g., FastAPI)
# @app.post('/predict')
# async def predict(request: Request):
#     data = await request.json()
#     return await handle_request(data)
...
```

--------------------------------

### Install Python Packages on Linux

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Commands to install Python 3, pip, and ensure 'python' command points to Python 3 on a Linux system. This is an alternative to using Conda for environment setup.

```bash
sudo apt install -y python3 python3-pip python-is-python3
```

--------------------------------

### Install Miniconda

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Steps to download and install the Miniconda installer for Python environment management. This is the recommended method for setting up isolated Python environments.

```bash
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
sh ./Miniconda3-latest-Linux-x86_64.sh
source ~/.bashrc
conda --version
```

--------------------------------

### Check Python Version

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Command to check the currently installed Python version on your system. This is a prerequisite for installing the FuriosaAI Python SDK.

```bash
python --version
Python 3.8.10
```

--------------------------------

### Install Furiosa SDK Server from Source

Source: https://developer.furiosa.ai/docs/latest/en/software/serving

Installs the Furiosa SDK server by cloning the source code and building it locally. This method is useful for development or when specific build configurations are needed.

```bash
git clone https://github.com/furiosa-ai/furiosa-sdk.git
cd furiosa-sdk/python/furiosa-server
pip install .
```

--------------------------------

### Full ResNet50 Optimization Example (Python)

Source: https://developer.furiosa.ai/docs/latest/en/software/performance

Demonstrates a complete workflow for optimizing a ResNet50 model. It includes exporting a PyTorch model to ONNX, optimizing the ONNX model, calibrating it, and then using `ModelEditor` to convert the input type.

```python
#!/usr/bin/env python

import time
import numpy as np
import onnx
import torch
import torchvision
from torchvision import transforms
import tqdm

from furiosa.optimizer import optimize_model
from furiosa.quantizer import get_pure_input_names, quantize, Calibrator, CalibrationMethod, ModelEditor, TensorType
from furiosa.runtime import session
from furiosa.runtime.profiler import profile


torch_model = torchvision.models.resnet50(weights='DEFAULT')
torch_model = torch_model.eval()

dummy_input = (torch.randn(1, 3, 224, 224),)

torch.onnx.export(
    torch_model,  # PyTorch model to export
    dummy_input,  # model input
    "resnet50.onnx",  # where to save the exported ONNX model
    opset_version=13,  # the ONNX OpSet version to export the model to
    do_constant_folding=True,  # whether to execute constant folding for optimization
    input_names=["input"],  # the ONNX model's input names
    output_names=["output"],  # the ONNX model's output names
)

onnx_model = onnx.load_model("resnet50.onnx")
onnx_model = optimize_model(onnx_model)

calibrator = Calibrator(onnx_model, CalibrationMethod.MIN_MAX_ASYM)
calibrator.collect_data([[torch.randn(1, 3, 224, 224).numpy()]])
ranges = calibrator.compute_range()

editor = ModelEditor(onnx_model)
input_tensor_name = get_pure_input_names(onnx_model)[0]

# Note: The actual conversion call editor.convert_input_type(input_tensor_name, TensorType.UINT8) is missing in the provided snippet but would follow here.

```

--------------------------------

### List Available furiosa-libnux Versions with apt

Source: https://developer.furiosa.ai/docs/latest/en/software/installation

This command lists all available versions of the `furiosa-libnux` package. It helps users identify the exact version string required for installation. No specific inputs are needed beyond the package name.

```bash
sudo apt list -a furiosa-libnux
```

--------------------------------

### Install FuriosaAI Models Package (PIP)

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Installs the FuriosaAI Models package, which provides optimized DNN model architectures and pre-trained model images executable on the NPU. This is installed as an extra package for the SDK.

```sh
pip install 'furiosa-sdk[models]'
```

--------------------------------

### Configure Device Plugin via Command Line Arguments

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/kubernetes_support.rst

Demonstrates how to configure the k8s-device-plugin using command-line arguments. This example sets the interval for device searching to 10 seconds. These arguments can be specified within a Pod or DaemonSet definition.

```sh
$ k8s-device-plugin --interval 10
```

```yaml
apiVersion: v1
kind: Pod
metadata:
  name: furiosa-device-plugin
  namespace: kube-system
spec:
  containers:
    - name: device-plugin
      image: ghcr.io/furiosa-ai/k8s-device-plugin:latest
      command: ["/usr/bin/k8s-device-plugin"]
      args: ["--interval", "10"]
```

--------------------------------

### Install FuriosaAI Python SDK with PIP

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Installs the core FuriosaAI Python SDK package using pip. This package includes the compiler CLI and inference API. Additional functionalities can be installed as Python extra packages.

```sh
pip install furiosa-sdk
```

--------------------------------

### Install and Run Furiosa Server (Shell)

Source: https://developer.furiosa.ai/docs/latest/en/_sources/releases/0.5.0.rst

Provides the command to install Furiosa Server, a serving framework supporting GRPC and REST API, and instructions on how to run it to serve models on the NPU. This feature simplifies model deployment and inference serving.

```sh
pip install furiosa-sdk[server]
# Command to run the server (specific command not provided in text, but implied)
```

--------------------------------

### Install FuriosaAI Quantizer Package (PIP)

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Installs the FuriosaAI Quantizer package, which offers APIs for converting models into a quantized format. This package is essential for optimizing models for the NPU.

```sh
pip install 'furiosa-sdk[quantizer]'
```

--------------------------------

### Install FuriosaAI Models Package

Source: https://developer.furiosa.ai/docs/latest/en/software/python-sdk

Install the FuriosaAI Models Python module, which provides optimized DNN model architectures and pre-trained model images that can be executed directly on the NPU.

```bash
pip install 'furiosa-sdk[models]'

```

--------------------------------

### Install Litmus Package for Model Compatibility

Source: https://developer.furiosa.ai/docs/latest/en/software/python-sdk

Install the Litmus package, a tool designed to verify the compatibility of a specified model with the Furiosa SDK. It simulates processes like model quantization and compilation.

```bash
pip install 'furiosa-sdk[litmus]'

```

--------------------------------

### Run Latency Benchmark with furiosa-bench

Source: https://developer.furiosa.ai/docs/latest/en/_sources/releases/0.10.0.rst

This command shows an example of using `furiosa-bench` for a latency-oriented benchmark. It includes the model path, workload type, number of iterations, and the specific devices to be used for measurement.

```bash
$ furiosa-bench ./model.onnx --workload latency -n 10000 --devices "warboy(2)*1"
```

--------------------------------

### Configure APT Authentication for FuriosaAI

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/installation.rst

Sets up authentication for the FuriosaAI APT server using an API key. It creates a configuration file with machine, login (key ID), and password (API key) details and sets restrictive read permissions.

```sh
sudo tee -a /etc/apt/auth.conf.d/furiosa.conf > /dev/null <<EOT
    machine archive.furiosa.ai
    login [KEY (ID)]
    password [PASSWORD]
  EOT

  sudo chmod 400 /etc/apt/auth.conf.d/furiosa.conf
```

--------------------------------

### Check Node Labels for Node Feature Discovery Installation

Source: https://developer.furiosa.ai/docs/latest/en/software/kubernetes_support

Queries Kubernetes nodes to check for existing labels, specifically looking for labels starting with `feature.node.kubernetes.io/`. This indicates whether the Node Feature Discovery DaemonSet is already installed and running.

```bash
$ kubectl get no -o json | jq '.items[].metadata.labels'
{
  "beta.kubernetes.io/arch": "amd64",
  "beta.kubernetes.io/os": "linux",
  "feature.node.kubernetes.io/cpu-cpuid.ADX": "true",
  "feature.node.kubernetes.io/cpu-cpuid.AESNI": "true",
  ...

```

--------------------------------

### Deactivate Conda Python Environment

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Command to exit the currently active Conda Python environment. This allows you to return to your base environment or switch to another.

```bash
conda deactivate
```

--------------------------------

### Full Example: Creating a Pod with NPU Allocation

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/kubernetes_support.rst

This shell script demonstrates the complete process of creating a Kubernetes Pod that requests NPU resources. It defines the Pod specification, including resource limits for CPU, memory, and NPUs, and then applies the configuration using `kubectl`.

```sh
$ cat > npu-pod.yaml <<EOL
apiVersion: v1
kind: Pod
metadata:
  name: npu-pod
spec:
  containers:
    - name: npu-pod
      image: ubuntu:focal
      resources:
        limits:
          cpu: "4"
          memory: "8Gi"
          beta.furiosa.ai/npu: "1"
        requests:
          cpu: "4"
          memory: "8Gi"
          beta.furiosa.ai/npu: "1"
EOL

$ kubectl apply -f npu-pod.yaml
```

--------------------------------

### Get Python Version

Source: https://developer.furiosa.ai/docs/latest/en/customer-support/bugs

This command-line snippet demonstrates how to check the installed Python version. It's a basic but essential step for debugging and reporting issues, ensuring compatibility with the Furiosa SDK.

```bash
python --version
```

--------------------------------

### Legacy Interface: create_async

Source: https://developer.furiosa.ai/docs/latest/en/_sources/api/python/furiosa.runtime.rst

Compiles a model if necessary, allocates an NPU device, and initializes a new asynchronous inference session. Returns an `AsyncSession` and a `CompletionQueue`.

```APIDOC
## POST /legacy/create_async

### Description
Initializes an asynchronous inference session by compiling the model, allocating an NPU device, and setting up queues.

### Method
POST

### Endpoint
/legacy/create_async

### Parameters
#### Request Body
- **model** (ModelSource) - Required - Model path or data.
- **device** (str or None) - Optional - Textual device specification.
- **worker_num** (int or None) - Optional - The number of worker threads.
- **batch_size** (int or None) - Optional - The number of batches per run.
- **compiler_hints** (bool or None) - Optional - If True, compiler prints additional hint messages.
- **input_queue_size** (int or None) - Optional - The size of the input queue.
- **output_queue_size** (int or None) - Optional - The size of the output queue.

### Request Example
```json
{
  "model": "path/to/model.onnx",
  "device": "npu0",
  "worker_num": 4,
  "batch_size": 1,
  "compiler_hints": true,
  "input_queue_size": 10,
  "output_queue_size": 10
}
```

### Response
#### Success Response (200)
- **async_session** (AsyncSession) - The asynchronous session object.
- **completion_queue** (CompletionQueue) - The completion queue for receiving results.

#### Response Example
```json
{
  "async_session": {},
  "completion_queue": {}
}
```

### Version History
- **0.10**
    - All optional arguments are now keyword-only.
    - Positional arguments are still accepted and behave identically for now, but will warn against future incompatibility.
    - Legacy queues were unbounded; now use much larger default sizes for migration facilitation.
```

--------------------------------

### Verify NPU Node Labels with kubectl

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/kubernetes_support.rst

This command retrieves all labels for Kubernetes nodes, allowing verification that labels starting with 'beta.furiosa.ai' are present, indicating successful installation of Node Feature Discovery.

```sh
kubectl get nodes --show-labels
```

--------------------------------

### Optimize Model and Initialize ModelEditor (Python)

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/performance.rst

Loads an ONNX model, optimizes it using optimize_model, and then initializes the ModelEditor with the optimized model. This is a prerequisite for using other ModelEditor functionalities.

```python
import onnx
from furiosa_quantizer.onnx.importer.model_editor import ModelEditor

model = onnx.load_model("yolox_l.onnx")
model = optimize_model(model)

editor = ModelEditor(model)
```

--------------------------------

### Create and Activate Conda Python Environment

Source: https://developer.furiosa.ai/docs/latest/en/_sources/software/python-sdk.rst

Commands to create a new isolated Python environment named 'furiosa-3.8' with Python version 3.8 and then activate it. This ensures a clean environment for the SDK.

```bash
conda create -n furiosa-3.8 python=3.8
conda activate furiosa-3.8
# version check
python --version
```

--------------------------------

### Implement OpenTelemetry Tracing in Python

Source: https://developer.furiosa.ai/docs/latest/en/_sources/releases/0.8.0.rst

Python code demonstrating how to integrate OpenTelemetry for tracing specific code sections within a serving application. It shows how to get a tracer, start a span, and manage its lifecycle.

```python
from opentelemetry import trace

tracer = trace.get_tracer(__name__)

class Application:

        async def process(self, image: Image.Image) -> int:

```