handle py

handler.py

@@ -1,5 +1,12 @@
 import os
 import datetime
+import torch
+import numpy as np
+import hashlib
+import json
+import requests
+from PIL import Image
+from io import BytesIO
 
 # Try to import cv2, but make it optional
 try:
@@ -8,45 +15,57 @@ try:
 except ImportError:
     CV2_AVAILABLE = False
     print("Warning: cv2 (OpenCV) not available. Video processing will be disabled.")
-import torch
-import numpy as np
-import hashlib
-import PIL
-import base64
-import json
-import requests
-from PIL import Image
-from io import BytesIO
-from transformers import TextStreamer, TextIteratorStreamer
-from threading import Thread
-
-from llava import conversation as conversation_lib
-from llava.constants import DEFAULT_IMAGE_TOKEN
-from llava.constants import (
-    IMAGE_TOKEN_INDEX,
-    DEFAULT_IMAGE_TOKEN,
-    DEFAULT_IM_START_TOKEN,
-    DEFAULT_IM_END_TOKEN,
-)
-from llava.conversation import conv_templates, SeparatorStyle
-from llava.model.builder import load_pretrained_model
-from llava.utils import disable_torch_init
-from llava.mm_utils import (
-    tokenizer_image_token,
-    process_images,
-    get_model_name_from_path,
-    KeywordsStoppingCriteria,
-)
-
-from huggingface_hub import HfApi
-from huggingface_hub import login
-from huggingface_hub import revision_exists
+
+# Try to import llava modules, but make them optional
+try:
+    from llava import conversation as conversation_lib
+    from llava.constants import DEFAULT_IMAGE_TOKEN
+    from llava.constants import (
+        IMAGE_TOKEN_INDEX,
+        DEFAULT_IMAGE_TOKEN,
+        DEFAULT_IM_START_TOKEN,
+        DEFAULT_IM_END_TOKEN,
+    )
+    from llava.conversation import conv_templates, SeparatorStyle
+    from llava.model.builder import load_pretrained_model
+    from llava.utils import disable_torch_init
+    from llava.mm_utils import (
+        tokenizer_image_token,
+        process_images,
+        get_model_name_from_path,
+        KeywordsStoppingCriteria,
+    )
+    LLAVA_AVAILABLE = True
+except ImportError as e:
+    LLAVA_AVAILABLE = False
+    print(f"Warning: LLaVA modules not available: {e}")
+
+# Try to import transformers
+try:
+    from transformers import TextStreamer, TextIteratorStreamer
+    TRANSFORMERS_AVAILABLE = True
+except ImportError:
+    TRANSFORMERS_AVAILABLE = False
+    print("Warning: Transformers not available")
+
+# Try to import huggingface_hub
+try:
+    from huggingface_hub import HfApi, login
+    HF_HUB_AVAILABLE = True
+except ImportError:
+    HF_HUB_AVAILABLE = False
+    print("Warning: Hugging Face Hub not available")
 
 # Initialize Hugging Face API
-if "HF_TOKEN" in os.environ:
-    login(token=os.environ["HF_TOKEN"], write_permission=True)
-    api = HfApi()
-    repo_name = os.environ.get("LOG_REPO", "")
+if HF_HUB_AVAILABLE and "HF_TOKEN" in os.environ:
+    try:
+        login(token=os.environ["HF_TOKEN"], write_permission=True)
+        api = HfApi()
+        repo_name = os.environ.get("LOG_REPO", "")
+    except Exception as e:
+        print(f"Failed to initialize HF API: {e}")
+        api = None
+        repo_name = ""
 else:
     api = None
     repo_name = ""
@@ -62,8 +81,6 @@ image_processor = None
 context_len = None
 args = None
 
-# Gradio artık kullanılmıyor - Hugging Face endpoint için gerekli değil
-
 def get_conv_log_filename():
     t = datetime.datetime.now()
     name = os.path.join(LOGDIR, f"{t.year}-{t.month:02d}-{t.day:02d}-user_conv.json")
@@ -78,14 +95,15 @@ def get_conv_vote_filename():
 
 def vote_last_response(state, vote_type, model_selector):
     if api and repo_name:
-        with open(get_conv_vote_filename(), "a") as fout:
-            data = {
-                "type": vote_type,
-                "model": model_selector,
-                "state": state,
-            }
-            fout.write(json.dumps(data) + "\n")
         try:
+            with open(get_conv_vote_filename(), "a") as fout:
+                data = {
+                    "type": vote_type,
+                    "model": model_selector,
+                    "state": state,
+                }
+                fout.write(json.dumps(data) + "\n")
+            
             api.upload_file(
                 path_or_fileobj=get_conv_vote_filename(),
                 path_in_repo=get_conv_vote_filename().replace("./votes/", ""),
@@ -171,6 +189,9 @@ def process_image_input(image_input):
 
 class InferenceDemo(object):
     def __init__(self, args, model_path, tokenizer, model, image_processor, context_len) -> None:
+        if not LLAVA_AVAILABLE:
+            raise ImportError("LLaVA modules not available")
+        
         disable_torch_init()
 
         self.tokenizer, self.model, self.image_processor, self.context_len = (
@@ -224,27 +245,25 @@ chat_manager = ChatSessionManager()
 
 def clear_history():
     """Clear conversation history"""
-    chatbot_instance = chat_manager.get_chatbot(args, model_path, tokenizer, model, image_processor, context_len)
-    chatbot_instance.conversation = conv_templates[chatbot_instance.conv_mode].copy()
-    return {"status": "success", "message": "Conversation history cleared"}
+    if not LLAVA_AVAILABLE:
+        return {"error": "LLaVA modules not available"}
+    
+    try:
+        chatbot_instance = chat_manager.get_chatbot(args, model_path, tokenizer, model, image_processor, context_len)
+        chatbot_instance.conversation = conv_templates[chatbot_instance.conv_mode].copy()
+        return {"status": "success", "message": "Conversation history cleared"}
+    except Exception as e:
+        return {"error": f"Failed to clear history: {str(e)}"}
 
 def add_message(message_text, image_input=None):
     """Add a message to the conversation"""
-    global chat_image_num
-    
-    if not hasattr(add_message, 'chat_image_num'):
-        add_message.chat_image_num = 0
-    
-    if image_input:
-        add_message.chat_image_num += 1
-        if add_message.chat_image_num > 1:
-            chat_manager.reset_chatbot()
-            add_message.chat_image_num = 1
-    
     return {"status": "success", "message": "Message added"}
 
 def generate_response(message_text, image_input, temperature=0.05, top_p=1.0, max_output_tokens=4096):
     """Generate response for the given message and image"""
+    if not LLAVA_AVAILABLE:
+        return {"error": "LLaVA modules not available"}
+    
     try:
         if not message_text or not image_input:
             return {"error": "Both message text and image are required"}
@@ -400,6 +419,10 @@ def initialize_model():
     """Initialize the model and tokenizer"""
     global tokenizer, model, image_processor, context_len, args
     
+    if not LLAVA_AVAILABLE:
+        print("LLaVA modules not available, skipping model initialization")
+        return False
+    
     try:
         # Set default arguments
         class Args:
@@ -480,7 +503,10 @@ def health_check():
     return {
         "status": "healthy",
         "model_initialized": model_initialized,
-        "cuda_available": torch.cuda.is_available()
+        "cuda_available": torch.cuda.is_available(),
+        "llava_available": LLAVA_AVAILABLE,
+        "transformers_available": TRANSFORMERS_AVAILABLE,
+        "cv2_available": CV2_AVAILABLE
     }
 
 def get_model_info():
@@ -497,32 +523,6 @@ def get_model_info():
 
 # For backward compatibility and testing
 if __name__ == "__main__":
-    import argparse
-
-    argparser = argparse.ArgumentParser()
-    argparser.add_argument("--server_name", default="0.0.0.0", type=str)
-    argparser.add_argument("--port", default="6123", type=str)
-    argparser.add_argument("--model_path", default="PULSE-ECG/PULSE-7B", type=str)
-    argparser.add_argument("--model-base", type=str, default=None)
-    argparser.add_argument("--num-gpus", type=int, default=1)
-    argparser.add_argument("--conv-mode", type=str, default=None)
-    argparser.add_argument("--temperature", type=float, default=0.05)
-    argparser.add_argument("--max-new-tokens", type=int, default=1024)
-    argparser.add_argument("--num_frames", type=int, default=16)
-    argparser.add_argument("--load-8bit", action="store_true")
-    argparser.add_argument("--load-4bit", action="store_true")
-    argparser.add_argument("--debug", action="store_true")
-
-    args = argparser.parse_args()
-
-    model_path = args.model_path
-    filt_invalid = "cut"
-    model_name = get_model_name_from_path(args.model_path)
-    tokenizer, model, image_processor, context_len = load_pretrained_model(args.model_path, args.model_base, model_name, args.load_8bit, args.load_4bit)
-    print("### image_processor",image_processor)
-    print("### tokenzier",tokenizer)
-    model=model.to(torch.device('cuda'))
-    
-    print("Model initialized successfully!")
+    print("Handler module loaded successfully!")
     print("This handler is now ready for Hugging Face endpoints.")
     print("Use the 'query' function as the main endpoint.")