Update app.py

app.py

@@ -1,13 +1,19 @@
 import gradio as gr
-from transformers import AutoProcessor, Kosmos2Model
+from transformers import AutoProcessor, LlavaForConditionalGeneration
 import torch
 from PIL import Image
 import io
 
-# 📦 تحميل المعالج والنموذج
-model_id = "microsoft/kosmos-2-patch14-224"
+# 📦 تحميل النموذج والمُعالِج
+model_id = "dima80/llava-7b-sla-lvis4v_train_1k"
+
+# 🔍 تقليل استهلاك الذاكرة باستخدام 8-bit quantization
+model = LlavaForConditionalGeneration.from_pretrained(
+    model_id,
+    device_map="auto",
+    load_in_8bit=True
+)
 processor = AutoProcessor.from_pretrained(model_id)
-model = Kosmos2Model.from_pretrained(model_id).to("cuda" if torch.cuda.is_available() else "cpu")
 
 # 💬 تعليمات الذكاء الاصطناعي لتكون خبير تداول
 SYSTEM_PROMPT = """
@@ -33,22 +39,16 @@ def analyze_chart(image):
         img = Image.open(io.BytesIO(image)).convert("RGB")
 
         # إنشاء prompt شامل
-        text = f"Question: {SYSTEM_PROMPT}\nAnswer:"
+        text = f"<image>\nUSER: {SYSTEM_PROMPT}\nASSISTANT:"
 
         # المعالجة
-        inputs = processor(images=img, text=text, return_tensors="pt").to("cuda" if torch.cuda.is_available() else "cpu")
+        inputs = processor(text, images=img, return_tensors='pt').to(0)
 
         # توليد التحليل
-        generated_ids = model.generate(
-            pixel_values=inputs["pixel_values"],
-            input_ids=inputs["input_ids"],
-            attention_mask=inputs["attention_mask"],
-            max_new_tokens=256
-        )
-
-        response = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
+        output = model.generate(**inputs, max_new_tokens=256)
+        response = processor.decode(output[0], skip_special_tokens=True)
 
-        return response.strip()
+        return response.split("ASSISTANT:")[-1].strip()
 
     except Exception as e:
         return f"❌ حدث خطأ أثناء التحليل: {str(e)}"