# hexagon_tpu v0.1.2 - Table of Contents

> QNN (Qualcomm AI Engine Direct) NIF bindings for the Hexagon NPU (QCS6490/HTP)

## Modules

- [HexagonTpu](HexagonTpu.md): Elixir bindings for the Qualcomm Hexagon NPU (HTP) via the QNN
(Qualcomm AI Engine Direct) runtime.
- [HexagonTpu.Context](HexagonTpu.Context.md): A loaded QNN context: a pre-compiled context binary (produced on the host
with `qnn-context-binary-generator`) deserialized onto the backend.
- [HexagonTpu.Graph](HexagonTpu.Graph.md): An executable graph retrieved from a loaded context.
- [HexagonTpu.GraphInfo](HexagonTpu.GraphInfo.md): Metadata for one graph in a loaded context: its name and I/O tensors.

- [HexagonTpu.Monitor](HexagonTpu.Monitor.md): Periodic guard for the NIF layer. Add it to your supervision tree
- [HexagonTpu.Quant](HexagonTpu.Quant.md): Quantize/dequantize helpers driven by `HexagonTpu.TensorInfo` metadata.
- [HexagonTpu.Runtime](HexagonTpu.Runtime.md): A QNN backend runtime: the dlopen'd backend library plus its backend and
device handles.
- [HexagonTpu.Stats](HexagonTpu.Stats.md): Observability for the NIF layer: native resource accounting (leak guard),
dirty-scheduler utilization, and memory snapshots.
- [HexagonTpu.TensorInfo](HexagonTpu.TensorInfo.md): Metadata for one graph input/output tensor, extracted from the QNN
context binary.

