Initial commit

app.py

@@ -4,7 +4,7 @@ import sqlite3
 from datetime import datetime
 import torch
 import gradio as gr
-from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
+from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
 from huggingface_hub import login
 
 # تنظیم logging
@@ -20,28 +20,17 @@ if not hf_token:
 login(token=hf_token)
 logger.debug("Hugging Face login successful")
 
-# لود مدل کدنویسی
-logger.debug("Loading CodeT5-base model")
+# لود مدل DeepSeek
+logger.debug("Loading DeepSeek-Coder-6.7B-Instruct model")
 try:
-    code_tokenizer = AutoTokenizer.from_pretrained("Salesforce/codeT5-base", cache_dir="/tmp/hf_cache")
-    code_model = AutoModelForSeq2SeqLM.from_pretrained("Salesforce/codeT5-base", cache_dir="/tmp/hf_cache")
-    code_gen = pipeline("text2text-generation", model=code_model, tokenizer=code_tokenizer, max_length=768, device=-1)
-    logger.debug("CodeT5-base model loaded")
+    tokenizer = AutoTokenizer.from_pretrained("DeepSeek/DeepSeek-Coder-6.7B-Instruct", cache_dir="/tmp/hf_cache")
+    model = AutoModelForCausalLM.from_pretrained("DeepSeek/DeepSeek-Coder-6.7B-Instruct", cache_dir="/tmp/hf_cache", torch_dtype=torch.float16, device_map="auto")
+    code_gen = pipeline("text-generation", model=model, tokenizer=tokenizer, max_new_tokens=1024, temperature=0.7, top_p=0.9)
+    logger.debug("DeepSeek-Coder-6.7B-Instruct model loaded")
 except Exception as e:
-    logger.error(f"Failed to load CodeT5-base: {e}")
+    logger.error(f"Failed to load DeepSeek model: {e}")
     raise
 
-# لود مدل ترجمه
-logger.debug("Loading Helsinki-NLP/opus-mt-fas-en model")
-try:
-    translate_tokenizer = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-fas-en", cache_dir="/tmp/hf_cache")
-    translate_model = AutoModelForSeq2SeqLM.from_pretrained("Helsinki-NLP/opus-mt-fas-en", cache_dir="/tmp/hf_cache")
-    translate_gen = pipeline("translation", model=translate_model, tokenizer=translate_tokenizer, max_length=300)
-    logger.debug("Helsinki-NLP/opus-mt-fas-en model loaded")
-except Exception as e:
-    logger.error(f"Failed to load translation model: {e}")
-    translate_gen = None  # Fallback: اگه مدل ترجمه لود نشد، از ورودی خام استفاده می‌کنیم
-
 # آزادسازی حافظه
 import gc
 gc.collect()
@@ -108,71 +97,67 @@ def get_history():
         logger.error(f"History retrieval error: {e}")
         return f"خطا در دریافت تاریخچه: {e}"
 
-# تابع دیباگ برای تست مدل‌ها
-def debug_models(text, language):
-    logger.debug(f"Debugging models with input: {text}, language: {language}")
+# تابع دیباگ برای تست مدل
+def debug_model(text, language):
+    logger.debug(f"Debugging model with input: {text}, language: {language}")
     try:
-        if translate_gen:
-            translated = translate_gen(text, max_length=300)[0]['translation_text']
-        else:
-            translated = text  # Fallback: اگه ترجمه کار نکرد، از ورودی خام استفاده کن
-        logger.debug(f"Translated text: {translated}")
-        prompt = f"Write a complete, correct, and well-explained code in {language} to: {translated}"
+        prompt = f"""You are a coding assistant. The user provided this instruction in Persian: "{text}". Translate it to English, then write a complete, correct, and well-explained code in {language} based on the translated instruction. Return the English translation, the prompt, and the generated code."""
         logger.debug(f"Generated prompt: {prompt}")
-        code_output = code_gen(prompt, max_length=768, num_beams=5, do_sample=False)[0]['generated_text']
-        logger.debug(f"Generated code: {code_output}")
-        return f"Debug: Translation: {translated}\nPrompt: {prompt}\nCode: {code_output}"
+        result = code_gen(prompt, max_new_tokens=1024, temperature=0.7, top_p=0.9, do_sample=True)[0]['generated_text']
+        logger.debug(f"Generated output: {result}")
+        return f"Debug: {result}"
     except Exception as e:
         logger.error(f"Debug error: {e}")
         return f"Debug error: {e}"
 
-# ترجمه و ساخت پرامپت
-def translate_and_prompt(text, language):
-    logger.debug(f"Translating input: {text}")
+# تولید پرامپت و کد
+def generate_code_and_prompt(text, language):
+    logger.debug(f"Processing input: {text}, language: {language}")
     try:
-        if translate_gen:
-            translated = translate_gen(text, max_length=300)[0]['translation_text']
-        else:
-            translated = text  # Fallback: اگه ترجمه کار نکرد، از ورودی خام استفاده کن
-        logger.debug(f"Translated text: {translated}")
-        prompt = f"Write a complete, correct, and well-explained code in {language} to: {translated}"
-        return prompt
-    except Exception as e:
-        logger.error(f"Translation error: {e}")
-        prompt = f"Write a complete, correct, and well-explained code in {language} to: {text}"
-        logger.debug(f"Fallback prompt: {prompt}")
-        return prompt
-
-# تولید کد
-def generate_code(prompt):
-    logger.debug(f"Generating code with prompt: {prompt}")
-    try:
-        result = code_gen(prompt, max_length=768, num_beams=5, do_sample=False)
-        code_output = result[0]['generated_text']
-        if not code_output.strip():
+        # پرامپت برای DeepSeek که هم ترجمه کنه و هم کد تولید کنه
+        prompt = f"""You are a coding assistant. The user provided this instruction in Persian: "{text}". First, translate the instruction to English. Then, write a complete, correct, and well-explained code in {language} based on the translated instruction. Return the English translation followed by the code."""
+        logger.debug(f"Generated prompt: {prompt}")
+        result = code_gen(prompt, max_new_tokens=1024, temperature=0.7, top_p=0.9, do_sample=True)[0]['generated_text']
+        
+        # جدا کردن ترجمه و کد از خروجی
+        lines = result.split('\n')
+        translation = ""
+        code = []
+        is_code = False
+        for line in lines:
+            if line.startswith("Translation:"):
+                translation = line.replace("Translation:", "").strip()
+            elif line.startswith("```"):
+                is_code = not is_code
+                continue
+            elif is_code:
+                code.append(line)
+        
+        code_output = "\n".join(code).strip()
+        if not code_output:
             logger.warning("Generated code is empty")
-            return "خطا: کد تولیدشده خالی است"
+            code_output = "خطا: کد تولیدشده خالی است"
+        
+        logger.debug(f"Translation: {translation}")
         logger.debug(f"Generated code: {code_output}")
         torch.cuda.empty_cache()
-        return code_output
+        return translation, prompt, code_output
     except Exception as e:
         logger.error(f"Code generation error: {e}")
-        return f"Code generation error: {e}"
+        return "", f"Error prompt: {text}", f"Code generation error: {e}"
 
 # تابع اصلی
 def christopher(user_input, language, show_history):
     logger.debug(f"Processing input: {user_input}, language: {language}, show_history: {show_history}")
     try:
         task_type = "نوشتن کد کامل"
-        # ترجمه و ساخت پرامپت
-        prompt = translate_and_prompt(user_input, language)
-        if "error" in prompt.lower():
+        # تولید ترجمه، پرامپت و کد
+        translation, prompt, response = generate_code_and_prompt(user_input, language)
+        if "error" in response.lower():
             history = get_history() if show_history else ""
-            return prompt, history
-        # تولید کد
-        response = generate_code(prompt)
+            return response, history
         # ذخیره تو دیتابیس (اختیاری)
-        save_to_memory(user_input, task_type, prompt, response)
+        save_to_memory(user_input, task_type, f"Translation: {translation}\nPrompt: {prompt}", response)
         history = get_history() if show_history else ""
         return response, history
     except Exception as e: