Merge branch 'Ai' into main

app.py

@@ -1,25 +1,18 @@
 # ===============================
 # 1️⃣ استيراد المكتبات
 # ===============================
-import os
-import threading, torch, json
+from transformers import AutoModelForCausalLM, AutoTokenizer, Trainer, TrainingArguments
+from peft import LoraConfig, get_peft_model, TaskType, PeftModel
 from flask import Flask, request, jsonify
 from flask_cors import CORS
+import threading, torch, os, json
 from datasets import Dataset
-from transformers import AutoModelForCausalLM, AutoTokenizer, Trainer, TrainingArguments
-from peft import LoraConfig, get_peft_model, TaskType, PeftModel
-
-# ===============================
-# 🟢 إصلاح مشكلة الكاش في Hugging Face Spaces
-# ===============================
-os.environ["HF_HOME"] = "/app/cache"
-os.environ["TRANSFORMERS_CACHE"] = "/app/cache"
 
 # ===============================
 # 2️⃣ إعداد المتغيرات
 # ===============================
 MODEL_NAME = "Qwen/Qwen2.5-0.5B-Instruct"
-FLASK_PORT = 7860   # Spaces بيشتغل على 7860
+FLASK_PORT = 7860   # 🚨 Hugging Face Spaces يشتغل على 7860
 SAVE_DIR = "./lora_model"
 DATASET_FILE = "dataset.json"
 
@@ -31,7 +24,7 @@ tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_NAME,
     device_map="auto",
-    torch_dtype=torch.float16
+    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32
 )
 print("✅ Base model loaded!")
 
@@ -56,7 +49,7 @@ if os.path.exists(SAVE_DIR):
         base_model = AutoModelForCausalLM.from_pretrained(
             MODEL_NAME,
             device_map="auto",
-            torch_dtype=torch.float16
+            torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32
         )
         model = PeftModel.from_pretrained(base_model, SAVE_DIR)
         print("✅ LoRA weights auto-loaded from disk! [Fine-tuned model running]")
@@ -74,7 +67,66 @@ CORS(app)
 trainer = None
 current_dataset = None
 
+# ===============================
+# 5a️⃣ توليد نصوص (generate)
+# ===============================
+@app.route("/generate", methods=["POST"])
+def generate():
+    try:
+        data = request.json
+        prompt = data.get("prompt", "").strip()
+
+        if current_dataset is None:
+            return jsonify({
+                "messages": [
+                    {
+                        "role": "assistant",
+                        "content": "⚖️ أنا زين، مساعد ذكي للاستشارات القانونية. برجاء طرح سؤال قانوني."
+                    }
+                ]
+            })
+
+        # البحث في الداتا
+        for item in current_dataset:
+            lines = item["text"].split("\n")
+            for line in lines:
+                if line.lower().startswith("user:") and prompt in line:
+                    response_messages = []
+                    for l in lines:
+                        if ":" in l:
+                            role, content = l.split(":", 1)
+                            response_messages.append({
+                                "role": role.strip().lower(),
+                                "content": content.strip()
+                            })
+                    return jsonify({"messages": response_messages})
+
+        # 🟡 fallback (لو مفيش في الداتا)
+        inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=150,
+            temperature=0.7,
+            do_sample=True
+        )
+        reply = tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+        return jsonify({
+            "messages": [
+                {"role": "assistant", "content": reply}
+            ]
+        })
+
+    except Exception as e:
+        return jsonify({
+            "messages": [
+                {"role": "assistant", "content": f"خطأ: {str(e)}"}
+            ]
+        })
+
+# ===============================
 # 🟢 تحميل الداتا من الملف عند البداية
+# ===============================
 def load_dataset():
     global current_dataset
     if os.path.exists(DATASET_FILE):
@@ -104,7 +156,99 @@ def load_dataset():
 load_dataset()
 
 # ===============================
-# 6️⃣ تشغيل Flask (بدون ngrok في Spaces)
+# 5b️⃣ تدريب LoRA (Async) + حفظ
+# ===============================
+@app.route("/train", methods=["POST"])
+def train():
+    global trainer, current_dataset, model
+    try:
+        data = []
+        epochs = 1
+        batch_size = 1
+
+        # 📌 1) JSON request
+        if request.is_json:
+            body = request.get_json()
+            data = body.get("data", [])
+            epochs = int(body.get("epochs", 1))
+            batch_size = int(body.get("batch_size", 1))
+
+        # 📌 2) Form-data request
+        else:
+            epochs = int(request.form.get("epochs", 1))
+            batch_size = int(request.form.get("batch_size", 1))
+            file = request.files.get("file")
+            if file:
+                data = json.load(file)
+            else:
+                return jsonify({"error": "No file uploaded"}), 400
+
+        # دمج وإزالة التكرارات
+        old_data = []
+        if os.path.exists(DATASET_FILE):
+            with open(DATASET_FILE, "r", encoding="utf-8") as f:
+                try:
+                    old_data = json.load(f)
+                except:
+                    old_data = []
+
+        combined = old_data + data
+        unique_data, seen = [], set()
+        for item in combined:
+            key = json.dumps(item, ensure_ascii=False)
+            if key not in seen:
+                seen.add(key)
+                unique_data.append(item)
+
+        # الحفظ
+        with open(DATASET_FILE, "w", encoding="utf-8") as f:
+            json.dump(unique_data, f, ensure_ascii=False, indent=2)
+
+        # تجهيز الداتا
+        formatted = []
+        for item in unique_data:
+            if "messages" in item:
+                conversation = ""
+                for msg in item["messages"]:
+                    conversation += f"{msg['role'].capitalize()}: {msg['content']}\n"
+                formatted.append({"text": conversation.strip()})
+            elif "text" in item:
+                formatted.append({"text": item["text"]})
+
+        current_dataset = Dataset.from_list(formatted)
+
+        def tokenize_function(examples):
+            return tokenizer(examples["text"], truncation=True, padding="max_length", max_length=256)
+
+        tokenized_dataset = current_dataset.map(tokenize_function, batched=True, remove_columns=["text"])
+
+        training_args = TrainingArguments(
+            output_dir=SAVE_DIR,
+            per_device_train_batch_size=batch_size,
+            num_train_epochs=epochs,
+            logging_steps=2,
+            save_strategy="epoch",
+            fp16=torch.cuda.is_available(),
+            report_to=None
+        )
+
+        trainer = Trainer(model=model, args=training_args, train_dataset=tokenized_dataset)
+
+        def run_training():
+            print("🔄 Training started...")
+            trainer.train()
+            trainer.save_model(SAVE_DIR)
+            print("✅ Training completed & model saved!")
+
+        threading.Thread(target=run_training, daemon=True).start()
+
+        return jsonify({"status": "training started", "epochs": epochs, "batch_size": batch_size, "records": len(unique_data)})
+
+    except Exception as e:
+        return jsonify({"error": str(e)})
+
+# ===============================
+# 6️⃣ تشغيل Flask (بدون ngrok)
 # ===============================
 if __name__ == "__main__":
     app.run(host="0.0.0.0", port=FLASK_PORT, threaded=True)

requirements.txt

@@ -1,6 +1,6 @@
-flask
-flask-cors
 transformers
-torch
 peft
+torch
 datasets
+flask
+flask-cors