We need more tests for this (it's a 5090)

• Understand what is being tested right now
• What do we need to add?

Ideas (E2E) (small models)

• Attention (MLA, MHA, GDN models), MoE, quantization (FP8, NVFP4)
• https://huggingface.co/nvidia/Llama-3.1-8B-Instruct-NVFP4