Update app.py

app.py

@@ -2,8 +2,15 @@
 # FILE: app.py
 # Description: Image-to-Video generation server with Gradio UI and FastAPI for Hugging Face Spaces
 # Version: 1.2.8
-# Timestamp: 2025-07-02 03:00 CDT
-
+# Timestamp: 2025-07-02 03:20 CDT
+# Author: Grok 3, built by xAI (based on GhostAI's ghostpack_gradio_f1.py)
+# NOTE: Optimized for Hugging Face Spaces with H200 GPU, 25 min/day render time
+#       Loads models from /data/models (pre-downloaded)
+#       Uses /data for persistent storage, /tmp for temporary files
+#       API key authentication for /generate endpoint (off-site use)
+#       Base64-encoded video responses
+#       Gradio UI matches original ghostpack_gradio_f1.py
+#       Idle until triggered by API or Gradio
 #       Requires custom diffusers_helper package in /diffusers_helper
 
 import os
@@ -107,7 +114,7 @@ args = parser.parse_args()
 render_on_off = True
 
 BASE = os.path.abspath(os.path.dirname(__file__))
-os.environ["HF_HOME"] = "/tmp/hf_cache"  # Cache models in /tmp
+MODEL_DIR = "/data/models"
 
 # Check if ports are available
 def is_port_in_use(port):
@@ -134,7 +141,7 @@ INSTALL_LOG_FILE = "/data/install_logs.txt"
 LAST_CLEANUP_FILE = "/data/last_cleanup.txt"
 
 # Initialize directories
-for d in (DATA_DIR, TMP_DIR, VIDEO_OUTPUT_DIR, VIDEO_IMG_DIR, VIDEO_TMP_DIR):
+for d in (DATA_DIR, TMP_DIR, VIDEO_OUTPUT_DIR, VIDEO_IMG_DIR, VIDEO_TMP_DIR, MODEL_DIR):
     if not os.path.exists(d):
         try:
             os.makedirs(d, exist_ok=True)
@@ -413,36 +420,36 @@ logger.info(f"VRAM available: {free_mem:.2f} GB, High VRAM mode: {hv}")
 print(f"{yellow(f'VRAM available: {free_mem:.2f} GB, High VRAM mode: {hv}')}")
 
 try:
-    print(f"{yellow('Loading models...')}")
+    print(f"{yellow('Loading models from /data/models...')}")
     text_encoder = LlamaModel.from_pretrained(
-        "hunyuanvideo-community/HunyuanVideo", subfolder="text_encoder", torch_dtype=torch.float16, token=HF_TOKEN, cache_dir="/tmp/hf_cache"
+        os.path.join(MODEL_DIR, "hunyuanvideo-community/HunyuanVideo/text_encoder"), torch_dtype=torch.float16
     ).cpu().eval()
     text_encoder_2 = CLIPTextModel.from_pretrained(
-        "hunyuanvideo-community/HunyuanVideo", subfolder="text_encoder_2", torch_dtype=torch.float16, token=HF_TOKEN, cache_dir="/tmp/hf_cache"
+        os.path.join(MODEL_DIR, "hunyuanvideo-community/HunyuanVideo/text_encoder_2"), torch_dtype=torch.float16
     ).cpu().eval()
     tokenizer = LlamaTokenizerFast.from_pretrained(
-        "hunyuanvideo-community/HunyuanVideo", subfolder="tokenizer", token=HF_TOKEN, cache_dir="/tmp/hf_cache"
+        os.path.join(MODEL_DIR, "hunyuanvideo-community/HunyuanVideo/tokenizer")
     )
     tokenizer_2 = CLIPTokenizer.from_pretrained(
-        "hunyuanvideo-community/HunyuanVideo", subfolder="tokenizer_2", token=HF_TOKEN, cache_dir="/tmp/hf_cache"
+        os.path.join(MODEL_DIR, "hunyuanvideo-community/HunyuanVideo/tokenizer_2")
     )
     vae = AutoencoderKLHunyuanVideo.from_pretrained(
-        "hunyuanvideo-community/HunyuanVideo", subfolder="vae", torch_dtype=torch.float16, token=HF_TOKEN, cache_dir="/tmp/hf_cache"
+        os.path.join(MODEL_DIR, "hunyuanvideo-community/HunyuanVideo/vae"), torch_dtype=torch.float16
     ).cpu().eval()
     feature_extractor = SiglipImageProcessor.from_pretrained(
-        "lllyasviel/flux_redux_bfl", subfolder="feature_extractor", token=HF_TOKEN, cache_dir="/tmp/hf_cache"
+        os.path.join(MODEL_DIR, "lllyasviel/flux_redux_bfl/feature_extractor")
     )
     image_encoder = SiglipVisionModel.from_pretrained(
-        "lllyasviel/flux_redux_bfl", subfolder="image_encoder", torch_dtype=torch.float16, token=HF_TOKEN, cache_dir="/tmp/hf_cache"
+        os.path.join(MODEL_DIR, "lllyasviel/flux_redux_bfl/image_encoder"), torch_dtype=torch.float16
     ).cpu().eval()
     transformer = HunyuanVideoTransformer3DModelPacked.from_pretrained(
-        "lllyasviel/FramePack_F1_I2V_HY_20250503", torch_dtype=torch.bfloat16, token=HF_TOKEN, cache_dir="/tmp/hf_cache"
+        os.path.join(MODEL_DIR, "lllyasviel/FramePack_F1_I2V_HY_20250503"), torch_dtype=torch.bfloat16
     ).cpu().eval()
-    logger.info("Models loaded successfully")
-    print(f"{green('Models loaded successfully')}")
+    logger.info("Models loaded successfully from /data/models")
+    print(f"{green('Models loaded successfully from /data/models')}")
 except Exception as e:
     logger.error(f"Failed to load models: {e}", exc_info=True)
-    print(f"{red(f'Error: Failed to load models: {e}')}")
+    print(f"{red(f'Error: Failed to load models from /data/models: {e}')}")
     raise
 
 if not hv:
@@ -539,6 +546,7 @@ async def test_server():
             },
             "paths": {
                 "base": BASE,
+                "models": MODEL_DIR,
                 "images": VIDEO_IMG_DIR,
                 "videos": VIDEO_OUTPUT_DIR,
                 "temp": VIDEO_TMP_DIR,
@@ -553,6 +561,7 @@ async def test_server():
                 "videos_writable": os.access(VIDEO_OUTPUT_DIR, os.W_OK),
                 "temp_writable": os.access(VIDEO_TMP_DIR, os.W_OK),
                 "data_writable": os.access(DATA_DIR, os.W_OK),
+                "models_writable": os.access(MODEL_DIR, os.W_OK),
             },
             "dependencies": {
                 "xformers": status_xformers(),
@@ -613,7 +622,7 @@ async def stop_render(job_id: str, api_key: str = Depends(verify_api_key)):
         logger.info(f"No active job {job_id} to stop")
         print(f"{yellow(f'No active job {job_id} to stop')}")
         return JSONResponse(content={"message": f"No active job {job_id}"})
-    stream = active_jobs[job_id]
+    stream = active_jobs[jid]
     stream.stop()
     active_jobs.pop(job_id, None)
     job_status[job_id]["status"] = "stopped"
@@ -1365,10 +1374,4 @@ if __name__ == "__main__":
                 logger.info(f"Public Gradio URL: {server.share_url}")
                 print(f"{yellow(f'Public Gradio URL: {server.share_url}')}")
             logger.info(f"Gradio UI running on http://{args.server}:7860")
-            print(f"{green(f'Gradio UI running on http://{args.server}:7860')}")
-        while True:
-            time.sleep(1)
-    except KeyboardInterrupt:
-        logger.info("Shutting down gracefully")
-        print(f"{green('Shutting down gracefully')}")
-        sys.exit(0)
+            print