Merge pull request #10 from shota3506/update-readme

update readme

Author: shota3506

README.md

@@ -1,14 +1,22 @@
 # onnxruntime-purego
+[![Go Reference](https://pkg.go.dev/badge/github.com/shota3506/onnxruntime-purego.svg)](https://pkg.go.dev/github.com/shota3506/onnxruntime-purego)
 
 Pure Go bindings for [ONNX Runtime](https://github.com/microsoft/onnxruntime) using [ebitengine/purego](https://github.com/ebitengine/purego).
 
 This library provides a pure Go interface to ONNX Runtime without requiring cgo, enabling cross-platform machine learning inference in Go applications.
 
 NOTE: This project is currently unstable. APIs may change without notice.
 
-## ONNX Runtime GenAI Support (Experimental)
+## Supported Versions
 
-This library also includes experimental support for [ONNX Runtime GenAI](https://github.com/microsoft/onnxruntime-genai), enabling text generation with large language models. See `examples/genai/` for usage examples.
+| Library | Supported Version |
+|---------|-------------------|
+| ONNX Runtime | 1.23.x |
+| ONNX Runtime GenAI | 0.11.x |
+
+## ONNX Runtime GenAI Support
+
+This library also includes experimental support for [ONNX Runtime GenAI](https://github.com/microsoft/onnxruntime-genai), enabling text generation with large language models. See [`examples/`](./examples/) for usage examples.
 
 ## Prerequisites
 
@@ -36,13 +44,4 @@ go get github.com/shota3506/onnxruntime-purego
 
 ## Examples
 
-The `examples/` directory contains complete examples:
-
-- **resnet**: Image classification using ResNet
-- **roberta-sentiment**: Sentiment analysis using RoBERTa
-- **yolov10**: Object detection using YOLOv10
-
-- **genai/phi3**: Text generation using Phi-3
-- **genai/phi3.5-vision**: Multimodal vision-language processing using Phi-3.5-Vision
-
-See each example's README for detailed instructions.
+See the [`examples/`](./examples/) directory for complete usage examples.

examples/README.md

@@ -0,0 +1,36 @@
+# Examples
+
+This directory contains example applications demonstrating the usage of onnxruntime-purego.
+
+## ONNX Runtime Examples
+
+- [**resnet**](./resnet/): Image classification using ResNet
+- [**roberta-sentiment**](./roberta-sentiment/): Sentiment analysis using RoBERTa
+- [**yolov10**](./yolov10/): Object detection using YOLOv10
+
+## ONNX Runtime GenAI Examples
+
+- [**genai/phi3**](./genai/phi3/): Text generation using Phi-3
+- [**genai/phi3.5-vision**](./genai/phi3.5-vision/): Multimodal vision-language processing using Phi-3.5-Vision
+
+### GenAI Prerequisites
+
+1. **ONNX Runtime GenAI Library**
+
+   Download from [ONNX Runtime GenAI Releases](https://github.com/microsoft/onnxruntime-genai/releases):
+   - macOS: `libonnxruntime-genai.dylib`
+   - Linux: `libonnxruntime-genai.so`
+   - Windows: `onnxruntime-genai.dll`
+
+2. **Environment Variable**
+
+   Set the library path before running GenAI examples:
+   ```bash
+   export ONNXRUNTIME_GENAI_LIB_PATH=/path/to/libonnxruntime-genai.dylib
+   ```
+
+### GenAI Troubleshooting
+
+1. **Library not found**: Ensure `ONNXRUNTIME_GENAI_LIB_PATH` is set correctly and the library is accessible.
+
+2. **macOS ARM64 crash on library load**: See [onnxruntime-genai issues](https://github.com/microsoft/onnxruntime-genai/issues).

examples/genai/phi3.5-vision/README.md

@@ -2,11 +2,11 @@
 
 This example demonstrates multimodal vision-language processing using ONNX Runtime GenAI with the Phi-3.5-Vision model. The model can analyze images and answer questions about their content.
 
-## Prerequisites
+See [examples/README.md](../README.md) for common GenAI prerequisites and troubleshooting.
 
-1. **ONNX Runtime GenAI Library**: Download from [releases](https://github.com/microsoft/onnxruntime-genai/releases)
+## Prerequisites
 
-2. **Phi-3.5-Vision Model** (~3.2GB):
+1. **Phi-3.5-Vision Model** (~3.2GB):
    ```bash
    huggingface-cli download microsoft/Phi-3.5-vision-instruct-onnx \
      --include "cpu_and_mobile/cpu-int4-rtn-block-32-acc-level-4/*" \
@@ -49,8 +49,6 @@ export ONNXRUNTIME_GENAI_LIB_PATH=/path/to/libonnxruntime-genai.dylib
 
    Reference: [GitHub Issue #1146](https://github.com/microsoft/onnxruntime-genai/issues/1146)
 
-2. **Library not found**: Ensure `ONNXRUNTIME_GENAI_LIB_PATH` is set correctly.
-
-3. **Model loading fails**: Verify the model directory contains all required files (genai_config.json, processor_config.json, and three ONNX models).
+2. **Model loading fails**: Verify the model directory contains all required files (genai_config.json, processor_config.json, and three ONNX models).
 
-4. **Out of memory during generation**: Reduce `-max-length` or close other applications.
+3. **Out of memory during generation**: Reduce `-max-length` or close other applications.

examples/genai/phi3/README.md

@@ -2,16 +2,11 @@
 
 This example demonstrates basic text generation using ONNX Runtime GenAI with the Phi-3-mini model.
 
-## Prerequisites
-
-1. **ONNX Runtime GenAI Library**
+See [examples/README.md](../README.md) for common GenAI prerequisites and troubleshooting.
 
-   Download from [ONNX Runtime GenAI Releases](https://github.com/microsoft/onnxruntime-genai/releases):
-   - macOS: `libonnxruntime-genai.dylib`
-   - Linux: `libonnxruntime-genai.so`
-   - Windows: `onnxruntime-genai.dll`
+## Prerequisites
 
-2. **Phi-3-mini Model**
+1. **Phi-3-mini Model**
 
    Download using Hugging Face CLI:
    ```bash
@@ -48,10 +43,6 @@ export ONNXRUNTIME_GENAI_LIB_PATH=/path/to/libonnxruntime-genai.dylib
 - `-model`: Path to model directory containing `genai_config.json` (required)
 - `-max-length`: Maximum tokens to generate (default: 200)
 
-## Environment Variables
-
-- `ONNXRUNTIME_GENAI_LIB_PATH`: Path to ONNX Runtime GenAI shared library (required)
-
 ## Expected Output
 
 ```
@@ -65,26 +56,6 @@ What is the capital of France? The capital of France is Paris.
 
 ## Troubleshooting
 
-1. **Library not found**: Ensure `ONNXRUNTIME_GENAI_LIB_PATH` is set correctly and the library is accessible.
-
-2. **Model loading fails**: Verify the model directory contains `genai_config.json` and the ONNX model files.
-
-3. **Slow generation**: The INT4 CPU model runs at ~20 tokens/sec. Consider using GPU acceleration for faster inference.
+1. **Model loading fails**: Verify the model directory contains `genai_config.json` and the ONNX model files.
 
-4. **macOS ARM64 crash on library load**: The pre-built macOS ARM64 binaries have a known issue where they crash during static initialization with `std::runtime_error` in `Ort::InitApi()`. This is a problem with the released binaries, not with the Go bindings.
-
-   **Workarounds:**
-   - Build ONNX Runtime GenAI from source for macOS ARM64
-   - Use Linux x64 environment for testing
-   - Check [GitHub Issues](https://github.com/microsoft/onnxruntime-genai/issues) for updates
-
-   **Debug output:**
-   ```
-   libc++abi: terminating due to uncaught exception of type std::runtime_error
-   SIGABRT: abort
-   ```
-
-   This crash occurs in:
-   ```
-   libonnxruntime-genai.dylib`Ort::InitApi()
-   ```
+2. **Slow generation**: The INT4 CPU model runs at ~20 tokens/sec. Consider using GPU acceleration for faster inference.