I’m running the text-to-text small model using the oga-igpu recipe, and I’m consistently hitting the following error only on an older ThinkPad_T14_AMD, while the exact same code and model work fine on a ThinkPad T14s Gen6 AMD Ryzen AI Pro:
RuntimeError: No adapter is available for DML.
...
model, tokenizer = from_pretrained(local_path, recipe="oga-igpu")
From initial debugging, it looks like ONNX Runtime GenAI cannot find a DirectML-compatible GPU adapter on the older T14 model, but it works correctly on the newer Ryzen AI hardware (which exposes a supported iGPU/NPU path).
Question
Is this older ThinkPad_T14_AMD officially unsupported for the oga-igpu recipe and DML acceleration, or is this a configuration/driver issue I should resolve on the device?
I can force a CPU recipe as a workaround, but I want to confirm whether this behavior is expected on non-Ryzen-AI hardware.