Add Ryzen AI NPU/Hybrid model selection for Hy-MT2-7B

[HoangVung]

Title: Add Ryzen AI NPU/Hybrid support for Hy-MT2-7B model testing

Problem

The app currently works with the "AMD Lemonade / TurnkeyML" backend and can detect the available Lemonade model:

Hy-MT2-7B-GGUF-Q4_K_M-VulkanClean

However, this model appears to be a GGUF/VulkanClean variant, which likely runs on Vulkan/iGPU rather than the Ryzen AI NPU.

I want to test and customize the Hy-MT2-7B model path so the app can use Ryzen AI NPU acceleration when available.

Current Behavior

The app shows:

Backend: AMD Lemonade / TurnkeyML
Model: Hy-MT2-7B-GGUF-Q4_K_M-VulkanClean (Lemonade)
Base URL: http://127.0.0.1:13305/api/v1

Translation can be started through the Local LLM Translation page, but there is no clear way to confirm or select whether the model is running on:

• CPU
• Vulkan/iGPU
• Ryzen AI NPU
• Hybrid NPU + iGPU

Expected Behavior

Add support for testing Hy-MT2-7B with Ryzen AI NPU/Hybrid execution.

The app should allow the user to:

1. Detect available Lemonade models from:

http://127.0.0.1:13305/api/v1/models

2. Distinguish model/runtime variants, for example:

GGUF / Vulkan / iGPU
ONNX / OGA / NPU
Hybrid / Ryzen AI

3. Select a Hy-MT2-7B NPU or Hybrid model if available.

4. Save the selected model into the runtime config.

5. Show clearly on the UI which runtime is being used:

AMD Lemonade - Vulkan/iGPU
AMD Lemonade - Ryzen AI NPU
AMD Lemonade - Hybrid NPU+iGPU

6. Provide a small test button to verify the selected model with a short Chinese-to-Vietnamese translation, for example:

今天我们学习导数。

Expected output:

Hôm nay chúng ta học đạo hàm.

Suggested Implementation

Please keep the change minimal and avoid refactoring unrelated parts of the app.

Suggested steps:

1. Only inspect the files related to runtime model configuration, Lemonade backend selection, and the UI model settings page.
2. Add model variant detection based on model name keywords such as:

NPU
Hybrid
OGA
ONNX
RyzenAI
Vulkan
GGUF

3. Add or improve the model selector so the user can select the exact Lemonade model returned by "/models".
4. Persist the selected model in the existing runtime config.
5. Update the Local LLM Translation page header to display the selected backend/model/runtime clearly.
6. Add a quick test action that sends one short request to the selected backend/model.
7. Do not change the translation pipeline, parser, subtitle logic, or file handling unless strictly necessary.

Validation

Run only the relevant checks.

Manual test:

1. Start Lemonade server at:

http://127.0.0.1:13305/api/v1

2. Open the app:

http://localhost:3000

3. Go to the model configuration page.
4. Refresh or load the Lemonade model list.
5. Select a Hy-MT2-7B NPU/Hybrid model if available.
6. Save configuration.
7. Return to the Local LLM Translation page.
8. Confirm the header displays the selected runtime/model.
9. Run a short translation test.
10. Check Windows Task Manager Performance tab to see whether NPU usage increases when using an NPU/Hybrid model.

Notes

If only the following model is available:

Hy-MT2-7B-GGUF-Q4_K_M-VulkanClean

then the app should clearly indicate that this is likely a GGUF/Vulkan/iGPU path, not an NPU path.

The app should not assume that any Hy-MT2-7B model automatically uses NPU. It should display the detected runtime type based on model name or backend metadata when available.