Update dependencies in requirements.txt

logs/api.log

@@ -0,0 +1,5 @@
+2024-11-09 07:49:24.302 | INFO     | __main__:<module>:271 - Starting server on port 8000
+2024-11-09 08:14:16.781 | INFO     | __main__:<module>:271 - Starting server on port 8000
+2024-11-09 08:19:05.939 | INFO     | __main__:<module>:271 - Starting server on port 8000
+2024-11-09 08:31:58.329 | INFO     | __main__:<module>:271 - Starting server on port 8000
+2024-11-09 08:41:50.867 | INFO     | __main__:<module>:271 - Starting server on port 8000

requirements.txt

@@ -7,3 +7,6 @@ wandb
 setuptools
 pydantic-settings
 boto3
+prometheus-client
+loguru
+bitsandbytes

src/api/mochi_serve.py

@@ -160,7 +160,7 @@ def predict(self, inputs: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
 
                         # Upload to S3
                         with open(temp_video_path, "rb") as video_file:
-                            s3_response = mp4_to_s3_json(video_file, "video.mp4")
+                            s3_response = mp4_to_s3_json(video_file, f"mochi_{int(time.time())}.mp4")
 
                         result = {
                             "status": "success",
@@ -241,7 +241,7 @@ def encode_response(self, output: Union[Dict[str, Any], List[Any]]) -> Dict[str,
 if __name__ == "__main__":
     import sys
     prometheus_logger = PrometheusLogger()
-    prometheus_logger.mount(path="/metrics", app=make_asgi_app(registry=registry))
+    prometheus_logger.mount(path="/api/v1/metrics", app=make_asgi_app(registry=registry))
     # Configure logging
     logger.remove()
     logger.add(
@@ -266,7 +266,9 @@ def encode_response(self, output: Union[Dict[str, Any], List[Any]]) -> Dict[str,
             devices="auto",
             max_batch_size=1,
             track_requests=True,
-            loggers=prometheus_logger
+            loggers=prometheus_logger,
+            generate_client_file=False
+
         )
         logger.info("Starting server on port 8000")
         server.run(port=8000)

src/configs/mochi_settings.py

@@ -38,21 +38,20 @@ class MochiSettings(BaseSettings):
     model_name: str = 'Genmo-Mochi'
     transformer_path: str = "imnotednamode/mochi-1-preview-mix-nf4"
     pipeline_path: str = "VikramSingh178/mochi-diffuser-bf16"
-    dtype: torch.dtype = torch.bfloat16
+    dtype: torch.dtype = torch.float16
     device: str = "cuda"
 
     # Optimization Settings
     enable_vae_tiling: bool = True
     enable_model_cpu_offload: bool = True
-    enable_attention_slicing: bool = False
-    attention_slice_size: Optional[int] = None
+
 
     # Video Generation Settings
     num_inference_steps: int = 20
     guidance_scale: float = 7.5
-    height: int = 480
-    width: int = 848
-    num_frames: int = 150
+    height: int = 640
+    width: int = 480
+    num_frames: int = 60
     fps: int = 10
 
     class Config:

src/scripts/mochi_diffusers.py

@@ -170,4 +170,40 @@ def clear_memory(self) -> None:
         if torch.cuda.is_available():
             logger.info("Clearing CUDA memory cache")
             torch.cuda.empty_cache()
-            torch.cuda.reset_peak_memory_stats()
+            torch.cuda.reset_peak_memory_stats()
+
+if __name__ == "__main__":
+    from configs.mochi_settings import MochiSettings
+
+    settings = MochiSettings()
+
+    # Initialize inference class
+    mochi_inference = MochiInference(settings)
+
+    # Define prompt and output path
+    prompt = "Close-up of a chameleon's eye, with its scaly skin changing color. Ultra high resolution 4k."
+    output_path = "/home/ubuntu/Minimochi/outputs/output.mp4"
+
+    # Generate video
+    try:
+        video_path = mochi_inference.generate(
+            prompt=prompt,
+            negative_prompt='((((ugly)))), (((duplicate))), ((morbid)), ((mutilated)), out of frame, extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck)))',
+            output_path=output_path,
+            num_inference_steps=30,  
+            guidance_scale=3.5,
+            height=480,
+            width=848,
+            num_frames=150,
+            fps=30,
+        )
+        print(f"Video saved to: {video_path}")
+    except RuntimeError as e:
+        print(f"Failed to generate video: {e}")
+
+    # Display GPU memory usage for debugging
+    allocated, max_allocated = mochi_inference.get_memory_usage()
+    print(f"Memory usage: {allocated:.2f}GB (peak: {max_allocated:.2f}GB)")
+
+    # Clear memory cache after inference
+    mochi_inference.clear_memory()