Legacy CPU support (AVX without AVX2) + new architecture loading failures

[l08850-alt]

Hello LM Studio team,

First, thank you for the wonderful product — it made local LLMs accessible to millions.

I'm running into two related issues on an older but functional laptop and would like to understand if there is a path forward:

Hardware:

• CPU: Intel Core i7-2720QM (Sandy Bridge, 2011) — supports AVX but NOT AVX2
• RAM: 16 GB DDR3
• Storage: SSD
• GPU: NVIDIA GT 555M 3 GB (Fermi, not usable for inference)
• OS: Windows 11

Issues encountered:

1. Latest LM Studio (Bionic 1.1.5): No compatible runtimes at all. All available engines (llama.cpp variants, Harmony) report "Needs attention" / "Non Compatible".

2. LM Studio 0.4.23: CPU llama.cpp runtime reports "Invalid CPU architecture".

3. LM Studio 0.2.10 AVX-Beta-V4: Runtime launches, but fails to load modern models with error:

   • "arch": "qwen2vl" → Failed to load model
   • "arch": "qwen35" → "The model architecture is not supported"

   Interestingly, the diagnostic JSON reports "supports_avx2": true for this CPU, which appears incorrect (Sandy Bridge does not have AVX2).

Request:
Would it be possible to ship (or offer as a separate download) an AVX-only llama.cpp runtime that also supports current model architectures (qwen2, qwen2vl, qwen3, qwen35)? There is a meaningful base of users on 2011–2012 Sandy/Ivy Bridge machines who cannot upgrade hardware immediately but would love to keep using LM Studio locally.

I'm happy to provide full diagnostic JSON or test any beta build on this hardware.

Thank you,
Valery