Model Anatomy See what's inside open-weight AI models

Can I run Qwen2.5-VL-32B-Instruct on an RTX 3090?

Probably, at 4-bit

The published weights are too big, but a 4-bit version (about 18.8 GB) would fit, with room for ≈ 10K tokens of context. The site hasn't analyzed a GGUF of Qwen2.5-VL-32B-Instruct yet.

Full analysis of Qwen2.5-VL-32B-Instruct Everything that runs on RTX 3090

Files and context on RTX 3090

Each file of Qwen2.5-VL-32B-Instruct (and of GGUF versions the site has analyzed), and how much context fits next to it, with the context cache in 16-bit or 8-bit.

FileSizeOn RTX 3090 Context (16-bit)Context (8-bit)
As published68.3 GBDoesn't fit——
4-bit (estimate)18.8 GBTight10K21K

Qwen2.5-VL-32B-Instruct on other hardware

Popular models that run on RTX 3090

Assumes 90% of the card's memory is usable and at least 4K tokens of context for one conversation, with the context cache in 16-bit unless noted. Fits leaves at least 20% of that memory for context; Tight leaves less.