clip-path

app.py

@@ -34,7 +34,7 @@ image_model_config_dict = OrderedDict({
     "ASLDM-256-obj": {
         # "config": "./configs/image_cond_diffuser_asl/image-ASLDM-256.yaml",
         # "ckpt_path": "./checkpoints/image_cond_diffuser_asl/image-ASLDM-256.ckpt",
-        "config": "/home/user/app/configs/image_cond_diffuser_asl/image-ASLDM-256.yaml",
+        "config": "./configs/image_cond_diffuser_asl/image-ASLDM-256.yaml",
         "ckpt_path": "checkpoints/image_cond_diffuser_asl/image-ASLDM-256.ckpt",
     },
 })
@@ -115,15 +115,17 @@ def load_model(model_name: str, model_config_dict: dict, inference_model: Infere
 
         config_ckpt_path = model_config_dict[model_name]
 
-        raw_config_file = config_ckpt_path["config"]
-        raw_config = OmegaConf.load(raw_config_file)
-        raw_clip_ckpt_path = raw_config['model']['params']['first_stage_config']['params']['aligned_module_cfg']['params']['clip_model_version'] 
-        clip_ckpt_path = os.path.join(model_path, raw_clip_ckpt_path)
-        raw_config['model']['params']['first_stage_config']['params']['aligned_module_cfg']['params']['clip_model_version'] = clip_ckpt_path
-        raw_config['model']['params']['cond_stage_config']['params']['version'] = clip_ckpt_path
-        OmegaConf.save(raw_config, 'current_config.yaml')
-
-        model_config = get_config_from_file('current_config.yaml')
+        # raw_config_file = config_ckpt_path["config"]
+        # raw_config = OmegaConf.load(raw_config_file)
+        # raw_clip_ckpt_path = raw_config['model']['params']['first_stage_config']['params']['aligned_module_cfg']['params']['clip_model_version'] 
+        
+        # clip_ckpt_path = os.path.join(model_path, raw_clip_ckpt_path)
+        # raw_config['model']['params']['first_stage_config']['params']['aligned_module_cfg']['params']['clip_model_version'] = clip_ckpt_path
+        # raw_config['model']['params']['cond_stage_config']['params']['version'] = clip_ckpt_path
+        # OmegaConf.save(raw_config, 'current_config.yaml')
+
+        # model_config = get_config_from_file('current_config.yaml')
+        model_config = get_config_from_file(config_ckpt_path["config"])
         if hasattr(model_config, "model"):
             model_config = model_config.model
 

configs/image_cond_diffuser_asl/image-ASLDM-256.yaml

@@ -24,7 +24,7 @@ model:
           target: michelangelo.models.tsal.clip_asl_module.CLIPAlignedShapeAsLatentModule
           params:
             # clip_model_version: "./checkpoints/clip/clip-vit-large-patch14"
-            clip_model_version: "checkpoints/clip/clip-vit-large-patch14"
+            clip_model_version: "openai/clip-vit-large-patch14"
 
         loss_cfg:
           target: torch.nn.Identity
@@ -33,7 +33,7 @@ model:
       target: michelangelo.models.conditional_encoders.encoder_factory.FrozenCLIPImageGridEmbedder
       params:
         # version: "./checkpoints/clip/clip-vit-large-patch14"
-        version: "checkpoints/clip/clip-vit-large-patch14"
+        version: "openai/clip-vit-large-patch14"
         zero_embedding_radio: 0.1
 
     first_stage_key: "surface"

configs/text_cond_diffuser_asl/text-ASLDM-256.yaml

@@ -23,7 +23,7 @@ model:
         aligned_module_cfg:
           target: michelangelo.models.tsal.clip_asl_module.CLIPAlignedShapeAsLatentModule
           params:
-            clip_model_version: "checkpoints/clip/clip-vit-large-patch14"
+            clip_model_version: "openai/clip-vit-large-patch14"
 
         loss_cfg:
           target: torch.nn.Identity
@@ -31,7 +31,7 @@ model:
     cond_stage_config:
       target: michelangelo.models.conditional_encoders.encoder_factory.FrozenAlignedCLIPTextEmbedder
       params:
-        version: "checkpoints/clip/clip-vit-large-patch14"
+        version: "openai/clip-vit-large-patch14"
         zero_embedding_radio: 0.1
         max_length: 77