rev app

app.py

@@ -12,18 +12,21 @@ try:
 except Exception as e:
     print(f"Failed to install packages: {e}")
 
-# [แก้ 1] ย้าย environment variables มาไว้ก่อน imports
+
+###
+# เพิ่มบรรทัดนี้ที่ต้นโค้ด ก่อน import torch
 import os
 os.environ['NVIDIA_VISIBLE_DEVICES'] = ''
+###
 
 import warnings
 import torch
+
+# เปลี่ยนแปลงที่ 1: เพิ่มการตั้งค่า dynamo ก่อน import unsloth
 torch._dynamo.config.suppress_errors = True
 torch._dynamo.config.verbose = False
 
-# [แก้ 2] ย้าย imports มาไว้ที่ module level
-from unsloth import FastVisionModel
-from transformers import AutoModelForVision2Seq
+ 
 from transformers import TextStreamer
 import gradio as gr
 from huggingface_hub import login
@@ -31,55 +34,92 @@ from PIL import Image
 
 warnings.filterwarnings('ignore')
 
+model = None
+tokenizer = None
+
 if 'HUGGING_FACE_HUB_TOKEN' in os.environ:
     print("กำลังเข้าสู่ระบบ Hugging Face Hub...")
     login(token=os.environ['HUGGING_FACE_HUB_TOKEN'])
 else:
     print("คำเตือน: ไม่พบ HUGGING_FACE_HUB_TOKEN")
 
-# [แก้ 3] เพิ่ม @spaces.GPU decorator
+# @spaces.GPU
+# def load_model():
+#     global model, tokenizer
+#     print("กำลังโหลดโมเดล...")
+#     try:
+#         from unsloth import FastVisionModel
+#         # โหลด base model และ tokenizer แบบพื้นฐาน
+#         base_model, tokenizer = FastVisionModel.from_pretrained(
+#             "unsloth/Llama-3.2-11B-Vision-Instruct"
+#         )
+        
+#         print("โหลด base model และ tokenizer สำเร็จ")
+        
+#         # โหลดโมเดล fine-tuned แบบพื้นฐาน
+#         from transformers import AutoModelForVision2Seq
+#         model = AutoModelForVision2Seq.from_pretrained(
+#             "Aekanun/Llama-3.2-11B-Vision-Instruct-XRay"
+#         ).to('cuda')
+        
+#         print("โหลดโมเดลสำเร็จ!")
+#         return True
+#     except Exception as e:
+#         print(f"เกิดข้อผิดพลาดในการโหลดโมเดล: {str(e)}")
+#         import traceback
+#         traceback.print_exc()  # เพิ่มการแสดง stack trace
+#         return False
+
 @spaces.GPU
-def model_context():
-    _tokenizer = None
-    _model = None
-
-    def init_models():
-        nonlocal _tokenizer, _model
-        try:
-            print("กำลังโหลด tokenizer...")
-            # [แก้ 4] ลบ imports ออกจาก function
-            base_model, _tokenizer = FastVisionModel.from_pretrained(
-                "unsloth/Llama-3.2-11B-Vision-Instruct",
-                use_gradient_checkpointing = "unsloth"
-            )
-            print("โหลด tokenizer สำเร็จ")
-
-            print("กำลังโหลดโมเดล fine-tuned...")
-            # [แก้ 5] ลบ import ออกจาก function
-            _model = AutoModelForVision2Seq.from_pretrained(
-                "Aekanun/Llama-3.2-11B-Vision-Instruct-XRay",
-                load_in_4bit=True,
-                torch_dtype=torch.float16
-            ).to('cuda')
-            FastVisionModel.for_inference(_model)
-            print("โหลดโมเดลสำเร็จ!")
-            return True
-        except Exception as e:
-            print(f"เกิดข้อผิดพลาดในการโหลดโมเดล: {str(e)}")
-            return False
-
-    def decorator(func):
-        def wrapper(*args, **kwargs):
-            return func(_model, _tokenizer, *args, **kwargs)
-        return wrapper
-
-    return init_models, decorator
-
-init_models, model_decorator = model_context()
-
-@model_decorator
+def load_model():
+    global model, tokenizer
+    print("กำลังโหลดโมเดล...")
+    try:
+        # โหลด tokenizer จาก base model
+        from unsloth import FastVisionModel
+        from transformers import AutoTokenizer
+        print("กำลังโหลด tokenizer...")
+        base_model, _tokenizer = FastVisionModel.from_pretrained(
+            "unsloth/Llama-3.2-11B-Vision-Instruct",
+            use_gradient_checkpointing = "unsloth"
+        )
+        
+        tokenizer = _tokenizer  # กำหนดค่าให้ตัวแปร global โดยตรง
+        print(f"2. ประเภทของ tokenizer: {type(tokenizer)}")
+        print(f"3. เมธอดที่มีใน tokenizer: {dir(tokenizer)}")
+        print("4. Global tokenizer after assignment:", type(tokenizer))  # เช็คค่า
+        
+        print("โหลด base model และ tokenizer สำเร็จ กำลังโหลดโมเดลที่ fine-tune...")
+        
+        # โหลดโมเดล fine-tuned
+        from transformers import AutoModelForVision2Seq
+        print("กำลังโหลดโมเดล fine-tuned...")
+        model = AutoModelForVision2Seq.from_pretrained(
+            "Aekanun/Llama-3.2-11B-Vision-Instruct-XRay",
+            load_in_4bit=True,
+            torch_dtype=torch.float16
+        ).to('cuda')
+        
+        FastVisionModel.for_inference(model)
+        print("โหลดโมเดลสำเร็จ!")
+        return True
+        
+    except Exception as e:
+        print(f"เกิดข้อผิดพลาดในการโหลดโมเดล: {str(e)}")
+        import traceback
+        traceback.print_exc()
+        return False
+
 @spaces.GPU(duration=30)
-def process_image(_model, _tokenizer, image):
+def process_image(image):
+    global model, tokenizer
+
+    print("Type of model:", type(model))
+    print("\nใน process_image():")
+    print("A. Type of tokenizer:", type(tokenizer))
+    if tokenizer is not None:
+        print("B. Available methods:", dir(tokenizer))
+    
     if image is None:
         return "กรุณาอัพโหลดรูปภาพ"
     
@@ -87,6 +127,7 @@ def process_image(_model, _tokenizer, image):
         if not isinstance(image, Image.Image):
             image = Image.fromarray(image)
 
+        print("0. Image info:", type(image), image.size)  # เพิ่ม debug ข้อมูลรูปภาพ
         instruction = "You are an expert radiographer. Describe accurately what you see in this image."
         messages = [
             {"role": "user", "content": [
@@ -95,17 +136,29 @@ def process_image(_model, _tokenizer, image):
             ]}
         ]
 
-        input_text = _tokenizer.apply_chat_template(messages, add_generation_prompt=True)
-        inputs = _tokenizer(
+        # input_text = tokenizer.apply_chat_template(messages, add_generation_prompt=True)
+        # inputs = tokenizer(
+        #     image,
+        #     input_text,
+        #     add_special_tokens=False,
+        #     return_tensors="pt",
+        # ).to("cuda")
+        print("1. Messages:", messages)  
+
+        print("2. Tokenizer type:", type(tokenizer))
+        input_text = tokenizer.apply_chat_template(messages, add_generation_prompt=True)
+        print("3. Chat template success:", input_text[:100])
+        inputs = tokenizer(
             image,
             input_text,
             add_special_tokens=False,
             return_tensors="pt",
         ).to("cuda")
+        print("3. Tokenizer inputs:", inputs.keys())  # Debug 3
 
-        text_streamer = TextStreamer(_tokenizer, skip_prompt=True)
-        outputs = _model.generate(
-            **inputs,
+        text_streamer = TextStreamer(tokenizer, skip_prompt=True)
+        outputs = model.generate(
+            **inputs, 
             streamer=text_streamer,
             max_new_tokens=128,
             use_cache=True,
@@ -113,13 +166,13 @@ def process_image(_model, _tokenizer, image):
             min_p=0.1
         )
         
-        return _tokenizer.decode(outputs[0], skip_special_tokens=True).strip()
+        return tokenizer.decode(outputs[0], skip_special_tokens=True).strip()
         
     except Exception as e:
         return f"เกิดข้อผิดพลาด: {str(e)}"
 
 print("กำลังเริ่มต้นแอปพลิเคชัน...")
-if init_models():
+if load_model():
     demo = gr.Interface(
         fn=process_image,
         inputs=gr.Image(type="pil"),

app.py.origi

@@ -1,41 +0,0 @@
-import os
-import sys
-import subprocess
-
-def install_packages():
-    print("Installing packages...")
-    subprocess.check_call([sys.executable, '-m', 'pip', 'install', '--upgrade', 'pip'])
-    subprocess.check_call([sys.executable, '-m', 'pip', 'install', 'huggingface_hub'])
-    subprocess.check_call([sys.executable, '-m', 'pip', 'install', 'transformers'])
-    subprocess.check_call([sys.executable, '-m', 'pip', 'install', 'torch', '--index-url', 'https://download.pytorch.org/whl/cpu'])
-    subprocess.check_call([sys.executable, '-m', 'pip', 'install', 'gradio'])
-    subprocess.check_call([sys.executable, '-m', 'pip', 'install', 'Pillow'])
-    subprocess.check_call([sys.executable, '-m', 'pip', 'install', 'bitsandbytes'])
-    subprocess.check_call([sys.executable, '-m', 'pip', 'install', 'accelerate'])
-
-if __name__ == "__main__":
-    try:
-        install_packages()
-        print("Package installation completed")
-        
-        import gradio as gr
-        import torch
-        from transformers import AutoProcessor
-        
-        def process_handwriting(image):
-            if image is None:
-                return "กรุณาอัพโหลดรูปภาพ"
-            return f"ทดสอบระบบ: Torch version: {torch.__version__}, Transformers installed"
-        
-        demo = gr.Interface(
-            fn=process_handwriting,
-            inputs=gr.Image(type="pil", label="อัพโหลดรูปภาพ"),
-            outputs=gr.Textbox(label="ผลลัพธ์"),
-            title="Test Installation",
-            description="ทดสอบการติดตั้ง libraries"
-        )
-        
-        demo.launch()
-    except Exception as e:
-        print(f"Error occurred: {str(e)}")
-        raise e

app.py.origi1

@@ -0,0 +1,133 @@
+import spaces
+import os
+import sys
+import subprocess
+
+def install_packages():
+    subprocess.check_call([sys.executable, "-m", "pip", "install", "unsloth-zoo"])
+    subprocess.check_call([sys.executable, "-m", "pip", "install", "--no-deps", "git+https://github.com/unslothai/unsloth.git"])
+
+try:
+    install_packages()
+except Exception as e:
+    print(f"Failed to install packages: {e}")
+
+# [แก้ 1] ย้าย environment variables มาไว้ก่อน imports
+import os
+os.environ['NVIDIA_VISIBLE_DEVICES'] = ''
+
+import warnings
+import torch
+torch._dynamo.config.suppress_errors = True
+torch._dynamo.config.verbose = False
+
+# [แก้ 2] ย้าย imports มาไว้ที่ module level
+from unsloth import FastVisionModel
+from transformers import AutoModelForVision2Seq
+from transformers import TextStreamer
+import gradio as gr
+from huggingface_hub import login
+from PIL import Image
+
+warnings.filterwarnings('ignore')
+
+if 'HUGGING_FACE_HUB_TOKEN' in os.environ:
+    print("กำลังเข้าสู่ระบบ Hugging Face Hub...")
+    login(token=os.environ['HUGGING_FACE_HUB_TOKEN'])
+else:
+    print("คำเตือน: ไม่พบ HUGGING_FACE_HUB_TOKEN")
+
+# [แก้ 3] เพิ่ม @spaces.GPU decorator
+@spaces.GPU
+def model_context():
+    _tokenizer = None
+    _model = None
+
+    def init_models():
+        nonlocal _tokenizer, _model
+        try:
+            print("กำลังโหลด tokenizer...")
+            # [แก้ 4] ลบ imports ออกจาก function
+            base_model, _tokenizer = FastVisionModel.from_pretrained(
+                "unsloth/Llama-3.2-11B-Vision-Instruct",
+                use_gradient_checkpointing = "unsloth"
+            )
+            print("โหลด tokenizer สำเร็จ")
+
+            print("กำลังโหลดโมเดล fine-tuned...")
+            # [แก้ 5] ลบ import ออกจาก function
+            _model = AutoModelForVision2Seq.from_pretrained(
+                "Aekanun/Llama-3.2-11B-Vision-Instruct-XRay",
+                load_in_4bit=True,
+                torch_dtype=torch.float16
+            ).to('cuda')
+            FastVisionModel.for_inference(_model)
+            print("โหลดโมเดลสำเร็จ!")
+            return True
+        except Exception as e:
+            print(f"เกิดข้อผิดพลาดในการโหลดโมเดล: {str(e)}")
+            return False
+
+    def decorator(func):
+        def wrapper(*args, **kwargs):
+            return func(_model, _tokenizer, *args, **kwargs)
+        return wrapper
+
+    return init_models, decorator
+
+init_models, model_decorator = model_context()
+
+@model_decorator
+@spaces.GPU(duration=30)
+def process_image(_model, _tokenizer, image):
+    if image is None:
+        return "กรุณาอัพโหลดรูปภาพ"
+    
+    try:
+        if not isinstance(image, Image.Image):
+            image = Image.fromarray(image)
+
+        instruction = "You are an expert radiographer. Describe accurately what you see in this image."
+        messages = [
+            {"role": "user", "content": [
+                {"type": "image"},
+                {"type": "text", "text": instruction}
+            ]}
+        ]
+
+        input_text = _tokenizer.apply_chat_template(messages, add_generation_prompt=True)
+        inputs = _tokenizer(
+            image,
+            input_text,
+            add_special_tokens=False,
+            return_tensors="pt",
+        ).to("cuda")
+
+        text_streamer = TextStreamer(_tokenizer, skip_prompt=True)
+        outputs = _model.generate(
+            **inputs,
+            streamer=text_streamer,
+            max_new_tokens=128,
+            use_cache=True,
+            temperature=1.5,
+            min_p=0.1
+        )
+        
+        return _tokenizer.decode(outputs[0], skip_special_tokens=True).strip()
+        
+    except Exception as e:
+        return f"เกิดข้อผิดพลาด: {str(e)}"
+
+print("กำลังเริ่มต้นแอปพลิเคชัน...")
+if init_models():
+    demo = gr.Interface(
+        fn=process_image,
+        inputs=gr.Image(type="pil"),
+        outputs=gr.Textbox(),
+        title="Medical Vision Analysis"
+    )
+    
+    if __name__ == "__main__":
+        demo.launch()
+else:
+    print("ไม่สามารถเริ่มต้นแอปพลิเคชันได้")