Temporarily disable HLFB for stable diffusion (#104)

* Tmp disable SDPA

* Update

Author: yichunk

ai_edge_torch/generative/examples/stable_diffusion/decoder.py

@@ -293,6 +293,7 @@ def get_model_config() -> unet_cfg.AutoEncoderConfig:
           qkv_fused_interleaved=False,
           rotary_percentage=0.0,
       ),
+      enable_hlfb=False,
   )
 
   mid_block_config = unet_cfg.MidBlock2DConfig(

ai_edge_torch/generative/examples/stable_diffusion/diffusion.py

@@ -294,13 +294,15 @@ def __init__(self, config: unet_cfg.DiffusionModelConfig):
                             attention_batch_size=config.transformer_batch_size,
                             normalization_config=config.transformer_norm_config,
                             attention_config=attention_config,
+                            enable_hlfb=False,
                         ),
                         cross_attention_block_config=unet_cfg.CrossAttentionBlock2DConfig(
                             query_dim=output_channel,
                             cross_dim=config.transformer_cross_attention_dim,
                             attention_batch_size=config.transformer_batch_size,
                             normalization_config=config.transformer_norm_config,
                             attention_config=attention_config,
+                            enable_hlfb=False,
                         ),
                         pre_conv_normalization_config=config.transformer_pre_conv_norm_config,
                         feed_forward_block_config=unet_cfg.FeedForwardBlock2DConfig(
@@ -354,13 +356,15 @@ def __init__(self, config: unet_cfg.DiffusionModelConfig):
                     attention_batch_size=config.transformer_batch_size,
                     normalization_config=config.transformer_norm_config,
                     attention_config=attention_config,
+                    enable_hlfb=False,
                 ),
                 cross_attention_block_config=unet_cfg.CrossAttentionBlock2DConfig(
                     query_dim=mid_block_channels,
                     cross_dim=config.transformer_cross_attention_dim,
                     attention_batch_size=config.transformer_batch_size,
                     normalization_config=config.transformer_norm_config,
                     attention_config=attention_config,
+                    enable_hlfb=False,
                 ),
                 pre_conv_normalization_config=config.transformer_pre_conv_norm_config,
                 feed_forward_block_config=unet_cfg.FeedForwardBlock2DConfig(
@@ -415,13 +419,15 @@ def __init__(self, config: unet_cfg.DiffusionModelConfig):
                             attention_batch_size=config.transformer_batch_size,
                             normalization_config=config.transformer_norm_config,
                             attention_config=attention_config,
+                            enable_hlfb=False,
                         ),
                         cross_attention_block_config=unet_cfg.CrossAttentionBlock2DConfig(
                             query_dim=output_channel,
                             cross_dim=config.transformer_cross_attention_dim,
                             attention_batch_size=config.transformer_batch_size,
                             normalization_config=config.transformer_norm_config,
                             attention_config=attention_config,
+                            enable_hlfb=False,
                         ),
                         pre_conv_normalization_config=config.transformer_pre_conv_norm_config,
                         feed_forward_block_config=unet_cfg.FeedForwardBlock2DConfig(