Add a doc showing how to run GPT OSS 20B with WebGPU

doc/RunGPTOSSWebGPU.md

@@ -0,0 +1,46 @@
+# Run GPT OSS 20B with Web GPU (preview)
+
+These instructions allow you to experiment with GPT OSS 20B on any GPU.
+
+Note: once the model is officially supported by Foundry Local that will be the easiest way to run it.
+
+## Model
+
+Download the model
+
+```bash
+https://huggingface.co/onnx-community/gpt-oss-20b-ONNX
+```
+
+Copy the files in the onnx folder into the parent folder
+
+Change the name of model_q4f16.onnx into model.onnx (as referenced in genai_config.json)
+
+## Runtime
+
+```bash
+pip install onnxruntime-webgpu
+pip install onnxruntime-genai --no-deps
+```
+
+## Example script
+
+```bash
+curl -O https://raw.githubusercontent.com/microsoft/onnxruntime-genai/refs/heads/main/examples/python/model-qa.py
+```
+
+## Run the model
+
+```bash
+python model-qa.py -m gpt-oss-20b-ONNX
+```
+
+Note that the reasoning tokens are not processed by this script and appear in the output.
+
+```bash
+Prompt (Use quit() to exit): Hello
+
+Output: <|channel|>analysis<|message|>We need to respond to "Hello". Simple greeting. Probably a short hello back and ask how can help.<|end|><|start|>assistant<|channel|>final<|message|>Hello! 👋 How can I help you today?
+
+Prompt (Use quit() to exit): 
+```