speed optim
This commit is contained in:
@@ -40,6 +40,10 @@ torch.backends.cudnn.benchmark = True
|
||||
torch.backends.cuda.matmul.allow_tf32 = True
|
||||
torch.backends.cudnn.allow_tf32 = True
|
||||
torch.set_float32_matmul_precision('medium')
|
||||
torch.backends.cuda.sdp_kernel("flash")
|
||||
torch.backends.cuda.enable_flash_sdp(True)
|
||||
torch.backends.cuda.enable_mem_efficient_sdp(True)
|
||||
torch.backends.cuda.enable_math_sdp(True)
|
||||
global_step = 0
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user