diff options
Diffstat (limited to 'modules/sd_models.py')
| -rw-r--r-- | modules/sd_models.py | 32 |
1 files changed, 32 insertions, 0 deletions
diff --git a/modules/sd_models.py b/modules/sd_models.py index 3b6cdea1..31bcb913 100644 --- a/modules/sd_models.py +++ b/modules/sd_models.py @@ -392,6 +392,38 @@ def load_model_weights(model, checkpoint_info: CheckpointInfo, state_dict, timer devices.dtype_unet = torch.float16
timer.record("apply half()")
+ if devices.get_optimal_device_name() == "mps":
+ enable_fp8 = False
+ elif shared.cmd_opts.opt_unet_fp8_storage:
+ enable_fp8 = True
+ elif model.is_sdxl and shared.cmd_opts.opt_unet_fp8_storage_xl:
+ enable_fp8 = True
+ else:
+ enable_fp8 = False
+
+ if enable_fp8:
+ devices.fp8 = True
+ if model.is_sdxl:
+ cond_stage = model.conditioner
+ else:
+ cond_stage = model.cond_stage_model
+
+ for module in cond_stage.modules():
+ if isinstance(module, torch.nn.Linear):
+ module.to(torch.float8_e4m3fn)
+
+ if devices.device == devices.cpu:
+ for module in model.model.diffusion_model.modules():
+ if isinstance(module, torch.nn.Conv2d):
+ module.to(torch.float8_e4m3fn)
+ elif isinstance(module, torch.nn.Linear):
+ module.to(torch.float8_e4m3fn)
+ else:
+ model.model.diffusion_model = model.model.diffusion_model.to(torch.float8_e4m3fn)
+ timer.record("apply fp8")
+ else:
+ devices.fp8 = False
+
devices.unet_needs_upcast = shared.cmd_opts.upcast_sampling and devices.dtype == torch.float16 and devices.dtype_unet == torch.float16
model.first_stage_model.to(devices.dtype_vae)
|