Update app.py

app.py

@@ -5,9 +5,14 @@ import sys
 import time
 import os
 import random
-from skyreelsinfer.offload import OffloadConfig
+
+from skyreelsinfer.offload import Offload, OffloadConfig
+from skyreelsinfer.pipelines import SkyreelsVideoPipeline
+
 from skyreelsinfer import TaskType
-from skyreelsinfer.skyreels_video_infer import SkyReelsVideoSingleGpuInfer
+
+#from skyreelsinfer.skyreels_video_infer import SkyReelsVideoSingleGpuInfer
+
 from diffusers.utils import export_to_video
 from diffusers.utils import load_image
 from PIL import Image
@@ -29,19 +34,29 @@ os.environ["SAFETENSORS_FAST_GPU"] = "1"
 
 os.putenv("TOKENIZERS_PARALLELISM","False")
 
+model_id = "Skywork/SkyReels-V1-Hunyuan-I2V"
+base_model_id = "hunyuanvideo-community/HunyuanVideo"
+
 def init_predictor():
     global pipe
-    pipe = SkyReelsVideoSingleGpuInfer(
-        task_type= TaskType.I2V,
-        model_id="Skywork/SkyReels-V1-Hunyuan-I2V",
-        quant_model=False,
-        is_offload=False,
-        offload_config=OffloadConfig(
-            high_cpu_memory=True,
-            parameters_level=True,
-            compiler_transformer=False,
-        )
-    )
+    text_encoder = LlamaModel.from_pretrained(
+        base_model_id,
+        subfolder="text_encoder",
+        torch_dtype=torch.bfloat16,
+    ).to("cpu")
+    transformer = HunyuanVideoTransformer3DModel.from_pretrained(
+        model_id,
+        # subfolder="transformer",
+        torch_dtype=torch.bfloat16,
+        device="cpu",
+    ).to("cpu").eval()
+    pipe = SkyreelsVideoPipeline.from_pretrained(
+        base_model_id,
+        transformer=transformer,
+        text_encoder=text_encoder,
+        torch_dtype=torch.bfloat16,
+    ).to("cpu")
+    pipe.to(torch.device('cuda'))
 
 @spaces.GPU(duration=60)
 def generate(segment, image, prompt, size, guidance_scale, num_inference_steps, frames, seed, progress=gr.Progress(track_tqdm=True) ):