Comment out unused imports.

Author: monorimet

src/diffusers/models/attention_processor.py

@@ -19,13 +19,14 @@
 import torch
 import torch.nn.functional as F
 from torch import nn
+import numpy as np
 
 from ..image_processor import IPAdapterMaskProcessor
 from ..utils import deprecate, logging
 from ..utils.import_utils import is_torch_npu_available, is_xformers_available
 from ..utils.torch_utils import maybe_allow_in_graph
 from .lora import LoRALinearLayer
-from shark_turbine.ops.iree import trace_tensor
+#from shark_turbine.ops.iree import trace_tensor
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -1116,10 +1117,13 @@ def __call__(
         batch_size = encoder_hidden_states.shape[0]
 
         # `sample` projections.
+        #trace_tensor("hidden_states", hidden_states[0,0,0])
         query = attn.to_q(hidden_states)
         key = attn.to_k(hidden_states)
         value = attn.to_v(hidden_states)
-
+        #trace_tensor("query_pre_proj", query[0,0,0])
+        #trace_tensor("key_pre_proj", key[0,0,0])
+        #trace_tensor("value_pre_proj", value[0,0,0])
         # `context` projections.
         encoder_hidden_states_query_proj = attn.add_q_proj(encoder_hidden_states)
         encoder_hidden_states_key_proj = attn.add_k_proj(encoder_hidden_states)
@@ -1129,20 +1133,18 @@ def __call__(
         query = torch.cat([query, encoder_hidden_states_query_proj], dim=1)
         key = torch.cat([key, encoder_hidden_states_key_proj], dim=1)
         value = torch.cat([value, encoder_hidden_states_value_proj], dim=1)
-
         inner_dim = key.shape[-1]
         head_dim = inner_dim // attn.heads
         query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
         key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
         value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-        # trace_tensor("query", query[0,0,0])
-        # trace_tensor("key", key[0,0,0])
-        # trace_tensor("value", value[0,0,0])
+        # np.save("q.npy", query.detach().cpu().numpy())
+        # np.save("k.npy", key.detach().cpu().numpy())
+        # np.save("v.npy", value.detach().cpu().numpy())
         hidden_states = hidden_states = F.scaled_dot_product_attention(
             query, key, value, dropout_p=0.0, is_causal=False
         )
         #trace_tensor("attn_out", hidden_states[0,0,0,0])
-
         hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
         hidden_states = hidden_states.to(query.dtype)
 
@@ -1152,7 +1154,7 @@ def __call__(
             hidden_states[:, residual.shape[1] :],
         )
         hidden_states_cl = hidden_states.clone()
-        trace_tensor("attn_out", hidden_states_cl[0,0,0])
+        #trace_tensor("attn_out", hidden_states_cl[0,0,0])
         # linear proj
         hidden_states = attn.to_out[0](hidden_states_cl)
         # dropout

src/diffusers/models/embeddings.py

@@ -23,6 +23,8 @@
 from .activations import FP32SiLU, get_activation
 from .attention_processor import Attention
 
+from shark_turbine.ops.iree import trace_tensor
+
 
 def get_timestep_embedding(
     timesteps: torch.Tensor,

src/diffusers/models/transformers/transformer_sd3.py

@@ -278,13 +278,13 @@ def forward(
         else:
             lora_scale = 1.0
 
-        # if USE_PEFT_BACKEND:
-        #     # weight the lora layers by setting `lora_scale` for each PEFT layer
-        #     scale_lora_layers(self, lora_scale)
-        # else:
-        #     logger.warning(
-        #         "Passing `scale` via `joint_attention_kwargs` when not using the PEFT backend is ineffective."
-        #     )
+        if USE_PEFT_BACKEND:
+            # weight the lora layers by setting `lora_scale` for each PEFT layer
+            scale_lora_layers(self, lora_scale)
+        else:
+            logger.warning(
+                "Passing `scale` via `joint_attention_kwargs` when not using the PEFT backend is ineffective."
+            )
 
         height, width = hidden_states.shape[-2:]
 

src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py

@@ -795,7 +795,8 @@ def __call__(
         if self.do_classifier_free_guidance:
             prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
             pooled_prompt_embeds = torch.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds], dim=0)
-
+        print("prompt_embeds", prompt_embeds)
+        print("pooled_prompt_embeds", pooled_prompt_embeds)
         # 4. Prepare timesteps
         timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
         num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
@@ -813,6 +814,8 @@ def __call__(
             generator,
             latents,
         )
+        print(latents)
+        print(timesteps)
 
         # 6. Denoising loop
         with self.progress_bar(total=num_inference_steps) as progress_bar:

src/diffusers/schedulers/scheduling_euler_ancestral_discrete.py

@@ -23,7 +23,7 @@
 from ..utils import BaseOutput, logging
 from ..utils.torch_utils import randn_tensor
 from .scheduling_utils import KarrasDiffusionSchedulers, SchedulerMixin
-from shark_turbine.ops.iree import trace_tensor
+#from shark_turbine.ops.iree import trace_tensor
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 

src/diffusers/schedulers/scheduling_pndm.py

@@ -22,7 +22,7 @@
 
 from ..configuration_utils import ConfigMixin, register_to_config
 from .scheduling_utils import KarrasDiffusionSchedulers, SchedulerMixin, SchedulerOutput
-from shark_turbine.ops.iree import trace_tensor
+#from shark_turbine.ops.iree import trace_tensor
 
 # Copied from diffusers.schedulers.scheduling_ddpm.betas_for_alpha_bar
 def betas_for_alpha_bar(