Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+images/fix-fp16.png filter=lfs diff=lfs merge=lfs -text
+images/fix-fp32.png filter=lfs diff=lfs merge=lfs -text
+images/orig-fp32.png filter=lfs diff=lfs merge=lfs -text

config.json

@@ -0,0 +1,32 @@
+{
+  "_class_name": "AutoencoderKLHunyuanVideo",
+  "_diffusers_version": "0.32.0.dev0",
+  "act_fn": "silu",
+  "block_out_channels": [
+    128,
+    256,
+    512,
+    512
+  ],
+  "down_block_types": [
+    "HunyuanVideoDownBlock3D",
+    "HunyuanVideoDownBlock3D",
+    "HunyuanVideoDownBlock3D",
+    "HunyuanVideoDownBlock3D"
+  ],
+  "in_channels": 3,
+  "latent_channels": 16,
+  "layers_per_block": 2,
+  "mid_block_add_attention": true,
+  "norm_num_groups": 32,
+  "out_channels": 3,
+  "scaling_factor": 0.476986,
+  "spatial_compression_ratio": 8,
+  "temporal_compression_ratio": 4,
+  "up_block_types": [
+    "HunyuanVideoUpBlock3D",
+    "HunyuanVideoUpBlock3D",
+    "HunyuanVideoUpBlock3D",
+    "HunyuanVideoUpBlock3D"
+  ]
+}

diffusion_pytorch_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c68a6295f9034a88225fbafb1f3258291a08d57a1fdb938233fa57b1b8f4883
+size 985943868

handler.py

@@ -0,0 +1,64 @@
+from typing import Dict, List, Any
+import torch
+from base64 import b64decode
+from diffusers import AutoencoderKLHunyuanVideo
+from diffusers.video_processor import VideoProcessor
+from diffusers.utils import export_to_video
+
+
+class EndpointHandler:
+    def __init__(self, path=""):
+        self.device = "cpu"
+        self.dtype = torch.float32
+        self.vae = (
+            AutoencoderKLHunyuanVideo.from_pretrained(
+                path, subfolder="vae", torch_dtype=self.dtype
+            )
+            .to(self.device, self.dtype)
+            .eval()
+        )
+
+        self.vae_scale_factor_spatial = self.vae.spatial_compression_ratio
+        self.video_processor = VideoProcessor(
+            vae_scale_factor=self.vae_scale_factor_spatial
+        )
+
+    @torch.no_grad()
+    def __call__(self, data: Any) -> List[List[Dict[str, float]]]:
+        """
+        Args:
+            data (:obj:):
+                includes the input data and the parameters for the inference.
+        """
+        tensor = data["inputs"]
+        tensor = b64decode(tensor.encode("utf-8"))
+        parameters = data.get("parameters", {})
+        if "shape" not in parameters:
+            raise ValueError("Expected `shape` in parameters.")
+        if "dtype" not in parameters:
+            raise ValueError("Expected `dtype` in parameters.")
+
+        DTYPE_MAP = {
+            "float16": torch.float16,
+            "float32": torch.float32,
+            "bfloat16": torch.bfloat16,
+        }
+
+        shape = parameters.get("shape")
+        dtype = DTYPE_MAP.get(parameters.get("dtype"))
+        tensor = torch.frombuffer(bytearray(tensor), dtype=dtype).reshape(shape)
+
+        tensor = tensor.to(self.device, self.dtype)
+
+        tensor = tensor / self.vae.config.scaling_factor
+
+        with torch.no_grad():
+            frames = self.vae.decode(tensor, return_dict=False)[0]
+
+        frames = self.video_processor.postprocess_video(frames, output_type="pil")[0]
+
+        path = export_to_video(frames, fps=15)
+        with open(path, "rb") as f:
+            video = f.read()
+
+        return video

requirements.txt

@@ -0,0 +1,3 @@
+huggingface_hub
+diffusers
+imageio-ffmpeg