disable diffusion to avoid hf version bug

Anymate/model.py

@@ -4,11 +4,11 @@ from ThirdParty.michelangelo.utils.misc import get_config_from_file, instantiate
 # from ThirdParty.PointLLM.pointllm.model.pointllm import PointLLMLlamaForCausalLM
 from ThirdParty.michelangelo.models.modules.distributions import DiagonalGaussianDistribution
 from ThirdParty.michelangelo.models.modules.embedder import components_from_spherical_harmonics
-from Anymate.utils.diffusion_encoder import TransformerEncoder
+# from Anymate.utils.diffusion_encoder import TransformerEncoder
 from Anymate.models.joint import TransformerDecoder, ImplicitTransformerDecoder, TriPlaneDecoder
 from Anymate.models.conn import AttendjointsDecoder_con_combine, AttendjointsDecoder_con_token
 from Anymate.models.skin import AttendjointsDecoder_combine, AttendjointsDecoder_multi
-from Anymate.models.diffusion import Pointe_Diffusion, Cross_Attention_Diffusion
+# from Anymate.models.diffusion import Pointe_Diffusion, Cross_Attention_Diffusion
 
 class Encoder(nn.Module):
     def __init__(self, 
@@ -172,15 +172,15 @@ class EncoderDecoder(nn.Module):
                 synthesis_kwargs = {'num_fp16_res': 0, 'conv_clamp': None, 'fused_modconv_default': 'inference_only'}
             )
             
-        elif decoder == 'Pointe_Diffusion':
-            self.only_embed = False
-            self.return_latents = True
-            self.decoder = Pointe_Diffusion(**kwargs)
+        # elif decoder == 'Pointe_Diffusion':
+        #     self.only_embed = False
+        #     self.return_latents = True
+        #     self.decoder = Pointe_Diffusion(**kwargs)
             
-        elif decoder == 'Cross_Attention_Diffusion':
-            self.only_embed = False
-            self.return_latents = True
-            self.decoder = Cross_Attention_Diffusion(**kwargs)
+        # elif decoder == 'Cross_Attention_Diffusion':
+        #     self.only_embed = False
+        #     self.return_latents = True
+        #     self.decoder = Cross_Attention_Diffusion(**kwargs)
             
         elif decoder == 'attendjoints_combine':
             self.only_embed = False
@@ -294,11 +294,11 @@ class EncoderDecoder(nn.Module):
                     ),
                     nn.Linear(513, 1, dtype=dtype)
                 ])
-        if encoder == 'transformer':
-            self.points_cloud_embed = nn.Linear(
-            768, 768, device=device, dtype=dtype
-        )
-            self.encoder = TransformerEncoder(device=device,dtype=dtype, num_latents=kwargs['num_latents'])
+        # if encoder == 'transformer':
+        #     self.points_cloud_embed = nn.Linear(
+        #     768, 768, device=device, dtype=dtype
+        # )
+        #     self.encoder = TransformerEncoder(device=device,dtype=dtype, num_latents=kwargs['num_latents'])