options to modify torch for maximum quality possible

[cerulliber]

Proposal

asked deepseek what can i do to maximize image quality. said add this to torch:

torch.set_float32_matmul_precision('high') # Maximize FP32 precision torch.backends.cuda.matmul.allow_tf32 = False # Disable TF32 acceleration

can you inject them into torch, please? or add options to be checked at launch ? thanks !!!