Update FantasyTalking/infer.py

FantasyTalking/infer.py

@@ -127,38 +127,35 @@ def parse_args():
 
 
 def load_models(args):
-    # Load Wan I2V models
-    model_manager = ModelManager(device="cpu")
-    model_manager.load_models(
-        [
-            [
-                f"{args.wan_model_dir}/diffusion_pytorch_model-00001-of-00007.safetensors",
-                f"{args.wan_model_dir}/diffusion_pytorch_model-00002-of-00007.safetensors",
-                f"{args.wan_model_dir}/diffusion_pytorch_model-00003-of-00007.safetensors",
-                f"{args.wan_model_dir}/diffusion_pytorch_model-00004-of-00007.safetensors",
-                f"{args.wan_model_dir}/diffusion_pytorch_model-00005-of-00007.safetensors",
-                f"{args.wan_model_dir}/diffusion_pytorch_model-00006-of-00007.safetensors",
-                f"{args.wan_model_dir}/diffusion_pytorch_model-00007-of-00007.safetensors",
-            ],
-            f"{args.wan_model_dir}/models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth",
-            f"{args.wan_model_dir}/models_t5_umt5-xxl-enc-bf16.pth",
-            f"{args.wan_model_dir}/Wan2.1_VAE.pth",
-        ],
-        # torch_dtype=torch.float8_e4m3fn, # You can set `torch_dtype=torch.bfloat16` to disable FP8 quantization.
-        torch_dtype=torch.bfloat16,  # You can set `torch_dtype=torch.bfloat16` to disable FP8 quantization.
-    )
-    pipe = WanVideoPipeline.from_model_manager(
-        model_manager, torch_dtype=torch.bfloat16, device="cuda"
-    )
+    # Load Wan I2V models (تم التعليق عليها مؤقتاً)
+    # model_manager = ModelManager(device="cpu")
+    # model_manager.load_models(
+    #     [
+    #         [
+    #             f"{args.wan_model_dir}/diffusion_pytorch_model-00001-of-00007.safetensors",
+    #             f"{args.wan_model_dir}/diffusion_pytorch_model-00002-of-00007.safetensors",
+    #             f"{args.wan_model_dir}/diffusion_pytorch_model-00003-of-00007.safetensors",
+    #             f"{args.wan_model_dir}/diffusion_pytorch_model-00004-of-00007.safetensors",
+    #             f"{args.wan_model_dir}/diffusion_pytorch_model-00005-of-00007.safetensors",
+    #             f"{args.wan_model_dir}/diffusion_pytorch_model-00006-of-00007.safetensors",
+    #             f"{args.wan_model_dir}/diffusion_pytorch_model-00007-of-00007.safetensors",
+    #         ],
+    #         f"{args.wan_model_dir}/models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth",
+    #         f"{args.wan_model_dir}/models_t5_umt5-xxl-enc-bf16.pth",
+    #         f"{args.wan_model_dir}/Wan2.1_VAE.pth",
+    #     ],
+    #     torch_dtype=torch.bfloat16,
+    # )
+    # pipe = WanVideoPipeline.from_model_manager(
+    #     model_manager, torch_dtype=torch.bfloat16, device="cuda"
+    # )
+
+    # مبدئياً نحط pipe بـ None أو تقدرِ تحطي أي Placeholder مؤقت
+    pipe = None
 
     # Load FantasyTalking weights
-    fantasytalking = FantasyTalkingAudioConditionModel(pipe.dit, 768, 2048).to("cuda")
-    fantasytalking.load_audio_processor(args.fantasytalking_model_path, pipe.dit)
-
-    # You can set `num_persistent_param_in_dit` to a small number to reduce VRAM required.
-    pipe.enable_vram_management(
-        num_persistent_param_in_dit=args.num_persistent_param_in_dit
-    )
+    fantasytalking = FantasyTalkingAudioConditionModel(None, 768, 2048).to("cuda")
+    fantasytalking.load_audio_processor(args.fantasytalking_model_path, None)
 
     # Load wav2vec models
     wav2vec_processor = Wav2Vec2Processor.from_pretrained(args.wav2vec_model_dir)
@@ -167,6 +164,7 @@ def load_models(args):
     return pipe, fantasytalking, wav2vec_processor, wav2vec
 
 
+
 def main(args, pipe, fantasytalking, wav2vec_processor, wav2vec):
     os.makedirs(args.output_dir, exist_ok=True)