- Normal small model on a benchmark
2) Big models - 2 GPU
3) Lora or similar models
4) Run small models with long context like open rag on both single gpu an 2 gpu
Both on ccc interactive job and non-interactive job (this sometime requires different config)