Merge pull request #4271 from MarkovInequality/racecond_fix

Fixes #4137 caused by race condition in training when VAE is unloaded

Author: AUTOMATIC1111

modules/hypernetworks/hypernetwork.py

@@ -433,7 +433,10 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, gradient_step,
 
     dl = modules.textual_inversion.dataset.PersonalizedDataLoader(ds, latent_sampling_method=latent_sampling_method, batch_size=ds.batch_size, pin_memory=pin_memory)
 
+    old_parallel_processing_allowed = shared.parallel_processing_allowed
+
     if unload:
+        shared.parallel_processing_allowed = False
         shared.sd_model.cond_stage_model.to(devices.cpu)
         shared.sd_model.first_stage_model.to(devices.cpu)
 
@@ -612,10 +615,12 @@ def train_hypernetwork(hypernetwork_name, learn_rate, batch_size, gradient_step,
     if shared.opts.save_optimizer_state:
         hypernetwork.optimizer_state_dict = optimizer.state_dict()
     save_hypernetwork(hypernetwork, checkpoint, hypernetwork_name, filename)
+
     del optimizer
     hypernetwork.optimizer_state_dict = None  # dereference it after saving, to save memory.
     shared.sd_model.cond_stage_model.to(devices.device)
     shared.sd_model.first_stage_model.to(devices.device)
+    shared.parallel_processing_allowed = old_parallel_processing_allowed
 
     return hypernetwork, filename
 

modules/textual_inversion/textual_inversion.py

@@ -269,6 +269,7 @@ def train_embedding(embedding_name, learn_rate, batch_size, gradient_step, data_
 
    # dataset loading may take a while, so input validations and early returns should be done before this
     shared.state.textinfo = f"Preparing dataset from {html.escape(data_root)}..."
+    old_parallel_processing_allowed = shared.parallel_processing_allowed
 
     pin_memory = shared.opts.pin_memory
 
@@ -279,6 +280,7 @@ def train_embedding(embedding_name, learn_rate, batch_size, gradient_step, data_
     dl = modules.textual_inversion.dataset.PersonalizedDataLoader(ds, latent_sampling_method=latent_sampling_method, batch_size=ds.batch_size, pin_memory=pin_memory)
 
     if unload:
+        shared.parallel_processing_allowed = False
         shared.sd_model.first_stage_model.to(devices.cpu)
 
     embedding.vec.requires_grad = True
@@ -450,6 +452,7 @@ def train_embedding(embedding_name, learn_rate, batch_size, gradient_step, data_
         pbar.leave = False
         pbar.close()
         shared.sd_model.first_stage_model.to(devices.device)
+        shared.parallel_processing_allowed = old_parallel_processing_allowed
 
     return embedding, filename