> For the complete documentation index, see [llms.txt](https://nx.docs.scailable.net/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://nx.docs.scailable.net/ai-manager-v6.1.4/ai-accelerators-support/introduction-to-oaxs.md).

# Accelerators Overview

How the AI Manager converts models into artifacts for each accelerator runtime.

The AI Manager accelerates AI model inference using built-in runtimes, each dedicated to a specific platform and AI accelerator (CPU, GPU, NPU, etc.). These runtimes integrate directly into the AI Manager, so new runtimes and acceleration types can be added.

When a user uploads a [supported model](https://nx.docs.scailable.net/ai-manager-v6.1.4/ai-models-support/importing-models.md) to the AI Manager Cloud, the platform runs several conversion processes on the file, each generating a model artifact for one of the [runtimes](https://nx.docs.scailable.net/ai-manager-v6.1.4/ai-accelerators-support/supported-ai-accelerators.md).

## Example of a Teachable Machine Model

Suppose you've trained a teachable machine model, and exported it according to the guidelines mentioned [here](https://nx.docs.scailable.net/ai-manager-v6.1.4/ai-models-support/importing-models/from-teachable-machine.md), the following model artifacts are stored to be used by the AI Manager when needed:

<figure><img src="https://2982163321-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKGfC0TFSmuMDjbpLErO3%2Fuploads%2Fgit-blob-15345680b78e6d713262294280c408b340fd4f1a%2Fmodel-mimetypes.png?alt=media" alt="Model detail page listing six file downloads: Hailo 8 ONNX, Default ONNX, Intel ZIP fp32, Intel ZIP fp16, Original ZIP and Hailo 8L ONNX"><figcaption><p>Model artifact types: each runtime uses a different artifact generated from the uploaded model.</p></figcaption></figure>

| Artifact type                                      | What it is                                                                                                                                                                                                                                                                 |
| -------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `application/zip; source=original`                 | The original ZIP archive that's uploaded in the interface.                                                                                                                                                                                                                 |
| `application/x-onnx`                               | Converts the TFLite generated earlier to ONNX, validated and optimized to run on CPU and NVIDIA hardware. The Intel runtime uses the OpenVINO artifact that the Cloud generates automatically from this ONNX file. *This step does not perform quantization on the model.* |
| `application/zip; device=intel; precision=fp32`    | The OpenVINO version of the model, generated automatically for the Intel CPU, GPU and NPU runtime. The Cloud also stores `fp16` (shown in the screenshot) and `int8` variants, but AI Manager 6.1.4 only uses `fp32`.                                                      |
| `application/x-hailo8` and `application/x-hailo8l` | Custom ONNX files generated specifically for Hailo-8 and Hailo-8L chips.                                                                                                                                                                                                   |
| `application/x-dxnn; version=2`                    | The compiled model generated by the DEEPX conversion for DEEPX chips. It is not part of the screenshot above.                                                                                                                                                              |

MemryX and EdgeCortix artifacts (`application/zip; device=memryx` and `application/zip; device=edge-cortix`) are generated by the Cloud from dedicated input archives, but the AI Manager 6.1.4 does not load these runtimes.

{% hint style="info" %}
Any runtime or toolchain combination that follows the [Open AI Accelerator eXchange (OAAX)](https://www.oaax.org/) standard is compatible with the AI Manager, so any installed runtime can be swapped for a new one.\
AI chip makers integrating with the AI Manager can refer to the [OAAX documentation](https://docs.oaax.org/).
{% endhint %}
