End of training

README.md

@@ -8,6 +8,12 @@ tags:
 - diffusers-training
 - stable-diffusion
 - stable-diffusion-diffusers
+- text-to-image
+- diffusers
+- lora
+- diffusers-training
+- stable-diffusion
+- stable-diffusion-diffusers
 base_model: runwayml/stable-diffusion-v1-5
 inference: true
 instance_prompt: a photo of sks dog

checkpoint-100/optimizer.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:646e226e970fccefc1bbeb856d1603c0acaebe38c1c91046d31e5d69226c1aef
-size 3396218
+oid sha256:3fef76dd1515c660279fab0218004aa69f28503f51d852ae91cca05e803d2c9c
+size 6584954

checkpoint-100/pytorch_lora_weights.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9010c3e01fc8a9d70b12418d71c2e96c7fe0bf898ddb3306269d5f2ba6086697
-size 1631680
+oid sha256:38a118bf5f1a6b3e2bf9961d6b0f877d4115480724b8385cc7d663a476bfa2a5
+size 3226184

checkpoint-200/optimizer.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f65026d17e0cabf6c80950e725e65da262f6d3bd76b75c4adf9f3e9b27e1bf8d
-size 3396218
+oid sha256:1deb8bd6df4ae7fcf08371f1db98545fbf342326ddf361ccf9159284eaae8676
+size 6584954

checkpoint-200/pytorch_lora_weights.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9010c3e01fc8a9d70b12418d71c2e96c7fe0bf898ddb3306269d5f2ba6086697
-size 1631680
+oid sha256:58259087cf6305a421542c60428fba4d5a9303d76e728320b80903115f9bf3fc
+size 3226184

checkpoint-300/optimizer.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:29391edc778cb306de5b8613608f602d5a8f2498628e3c9120b97f4e98364630
-size 3396218
+oid sha256:9254592e65ed4498ffcd54ca7577d5a3c88ea0d3f8f238b8e932fbdf4e0d60e7
+size 6584954

checkpoint-300/pytorch_lora_weights.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9010c3e01fc8a9d70b12418d71c2e96c7fe0bf898ddb3306269d5f2ba6086697
-size 1631680
+oid sha256:683502e4639c5a2d8dd42312c31af8083c312a4f696ff41fed1bd7b005170f96
+size 3226184

checkpoint-400/optimizer.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd4eaacd22302b2f78116d0e7f7b9f57529c7cecd31974588bcf61d08f2d5c57
-size 3396218
+oid sha256:059700890bc1c2a7db1642ed88058c52f9ad676ae797bf3e6034ceb529d90b76
+size 6584954

checkpoint-400/pytorch_lora_weights.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9010c3e01fc8a9d70b12418d71c2e96c7fe0bf898ddb3306269d5f2ba6086697
-size 1631680
+oid sha256:8e8407fe57dc47e869f0909b843591c3b7343e91caa866acca3cf8321d0aea28
+size 3226184

checkpoint-500/optimizer.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9bd2ec2952a1e2811f017611d3f1ee172ccd82f309230ec7109e3f50e2dc1388
-size 3396218
+oid sha256:cfb252d5c6851d8f59de49f3d0f65dd422fe4dcf9bccf6fe95a6cb779e830aa6
+size 6584954

checkpoint-500/pytorch_lora_weights.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9010c3e01fc8a9d70b12418d71c2e96c7fe0bf898ddb3306269d5f2ba6086697
-size 1631680
+oid sha256:962469684f6d9dd5033894b8be075ca09ef8a6c439da9bb85c8316c9e53e8166
+size 3226184

examples/dreambooth/train_dreambooth_lora.py

@@ -142,6 +142,7 @@ def log_validation(
     pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config, **scheduler_args)
 
     pipeline = pipeline.to(accelerator.device)
+    pipeline.enable_attention_slicing()
     pipeline.set_progress_bar_config(disable=True)
 
     # run inference
@@ -151,7 +152,9 @@ def log_validation(
         images = []
         for _ in range(args.num_validation_images):
             with torch.cuda.amp.autocast():
+                print("Gen Image")
                 image = pipeline(**pipeline_args, generator=generator).images[0]
+
                 images.append(image)
     else:
         images = []
@@ -748,7 +751,7 @@ def main(args):
         log_with=args.report_to,
         project_config=accelerator_project_config,
     )
-
+    print("Accelerator setup")
     # Disable AMP for MPS.
     if torch.backends.mps.is_available():
         accelerator.native_amp = False
@@ -784,8 +787,10 @@ def main(args):
     if args.seed is not None:
         set_seed(args.seed)
 
+    print("Before prior preservation")
     # Generate class images if prior preservation is enabled.
     if args.with_prior_preservation:
+        print("In prior preservation")
         class_images_dir = Path(args.class_data_dir)
         if not class_images_dir.exists():
             class_images_dir.mkdir(parents=True)
@@ -815,6 +820,8 @@ def main(args):
             sample_dataloader = torch.utils.data.DataLoader(sample_dataset, batch_size=args.sample_batch_size)
 
             sample_dataloader = accelerator.prepare(sample_dataloader)
+            print("printing accelerator defined device")
+            print(accelerator.device)
             pipeline.to(accelerator.device)
 
             for example in tqdm(
@@ -882,11 +889,14 @@ def main(args):
     # For mixed precision training we cast all non-trainable weights (vae, non-lora text_encoder and non-lora unet) to half-precision
     # as these weights are only used for inference, keeping weights in full precision is not required.
     weight_dtype = torch.float32
+    print("What is accelerator mixed precision")
+    print(accelerator.mixed_precision)
+    print(accelerator.device)
     if accelerator.mixed_precision == "fp16":
         weight_dtype = torch.float16
     elif accelerator.mixed_precision == "bf16":
         weight_dtype = torch.bfloat16
-
+    print(weight_dtype)
     # Move unet, vae and text_encoder to device and cast to weight_dtype
     unet.to(accelerator.device, dtype=weight_dtype)
     if vae is not None:
@@ -1091,6 +1101,7 @@ def main(args):
         validation_prompt_negative_prompt_embeds = None
         pre_computed_class_prompt_encoder_hidden_states = None
 
+    print("Getting dataset")
     # Dataset and DataLoaders creation:
     train_dataset = DreamBoothDataset(
         instance_data_root=args.instance_data_dir,
@@ -1204,7 +1215,9 @@ def main(args):
     )
 
     for epoch in range(first_epoch, args.num_train_epochs):
+        print("Unet train start")
         unet.train()
+        print("Unet train done")
         if args.train_text_encoder:
             text_encoder.train()
         for step, batch in enumerate(train_dataloader):
@@ -1335,6 +1348,7 @@ def main(args):
                 break
 
         if accelerator.is_main_process:
+            print("Accelerator is main process")
             if args.validation_prompt is not None and epoch % args.validation_epochs == 0:
                 # create pipeline
                 pipeline = DiffusionPipeline.from_pretrained(
@@ -1345,6 +1359,8 @@ def main(args):
                     variant=args.variant,
                     torch_dtype=weight_dtype,
                 )
+                pipeline = pipeline.to(accelerator.device)
+                pipeline.enable_attention_slicing()
 
                 if args.pre_compute_text_embeddings:
                     pipeline_args = {
@@ -1353,7 +1369,7 @@ def main(args):
                     }
                 else:
                     pipeline_args = {"prompt": args.validation_prompt}
-
+                print("Going for images")
                 images = log_validation(
                     pipeline,
                     args,
@@ -1391,6 +1407,8 @@ def main(args):
         # load attention processors
         pipeline.load_lora_weights(args.output_dir, weight_name="pytorch_lora_weights.safetensors")
 
+        pipeline = pipeline.to(accelerator.device)
+        pipeline.enable_attention_slicing()
         # run inference
         images = []
         if args.validation_prompt and args.num_validation_images > 0:

pytorch_lora_weights.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:278b3ce5d13d106dfe44fda29e478d65a45d3a1b7d767300088014b90147aaaf
+oid sha256:962469684f6d9dd5033894b8be075ca09ef8a6c439da9bb85c8316c9e53e8166
 size 3226184