[Feature]: mutiple GPU = total VRAM

### Problem or motivation

Hi,

I have 2 GPUs, 5090 and 5080, and I run inferences with the total VRAM. I see who is not counted and does not limit itself to a single GPU.

### Proposed solution

Calculate the total VRAM + propose the llms that are capable of being played

### Alternatives considered

_No response_

### Feature area

Hardware detection (new GPU vendor, new hardware type)

### Would you be willing to contribute this?

No, just suggesting

### Additional context

_No response_

Start at the hardware-detection entry point and trace how VRAM is currently counted and how model recommendations are produced. Define done as aggregating VRAM from multiple GPUs and limiting recommendations to models that fit the aggregate; add coverage for a two-GPU setup such as a 5090 and 5080 if the project’s existing test structure supports it.

索引モデルが issue の本文から書いたものです。