Update app.py

app.py

@@ -6,9 +6,25 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 import time, uuid, json, asyncio, requests
 from bs4 import BeautifulSoup
 from typing import Optional, List, Union
-import re # On importe le module pour les expressions régulières
+import re
 
-# --- AGENT BRAIN (SYSTEM PROMPT) ---
+# ==============================================================================
+# ===                BLOC DE CONFIGURATION DE L'AGENT                        ===
+# === Un jour, changez les valeurs ici pour utiliser un nouveau modèle.      ===
+# ==============================================================================
+
+MODEL_CONFIG = {
+    # L'identifiant du modèle depuis le Hub Hugging Face (pas de GGUF).
+    "MODEL_ID": "deepseek-ai/deepseek-coder-1.3b-instruct",
+
+    # Paramètres de génération (ajustez si nécessaire pour le nouveau modèle)
+    "MAX_TOKENS_PLAN": 1024,      # Tokens max pour la phase de réflexion/planification.
+    "MAX_TOKENS_ANSWER": 1024,    # Tokens max pour la réponse finale.
+    "TEMPERATURE": 0.1,           # Contrôle la créativité (plus bas = plus déterministe).
+}
+
+# --- CERVEAU DE L'AGENT (SYSTEM PROMPT) ---
+# Si vous changez de modèle, vous devrez peut-être adapter ce prompt à son format.
 SYSTEM_PROMPT = """
 You are a highly advanced AI agent specializing in WordPress & WooCommerce development. You must follow a strict "Think, Act, Answer" workflow for every user request. Your primary directive is to be transparent, showing your thought process before taking any action.
 
@@ -32,20 +48,30 @@ You MUST structure your response within the following XML tags. This is not opti
     -   If you can answer the user's request WITHOUT using a tool, formulate the complete and final answer here.
     -   If you used a tool, leave this tag empty in your first response. You will be given the tool's output and asked to generate the final answer in a second step.
     </final_answer>
+
+### AVAILABLE TOOLS ###
+-   **Web Browser:** To use it, populate the `<tool_code>` tag with a JSON object: `{"tool": "browse", "url": "your_url_here"}`
+
+### CODING RULES (For the content inside <final_answer>) ###
+-   Always provide secure, efficient, and standard-compliant code.
+-   Explain where to place the code (`functions.php`, custom plugin, etc.).
 """
 
-# --- Configuration & Model Loading --- (Identique)
-MODEL_ID = "deepseek-ai/deepseek-coder-1.3b-instruct"
+# ==============================================================================
+# ===          FIN DU BLOC DE CONFIGURATION - NE PAS MODIFIER CI-DESSOUS     ===
+# ==============================================================================
+
+# --- Configuration & Model Loading ---
 DEVICE = "cpu"
-print(f"Loading model: {MODEL_ID}")
-model = AutoModelForCausalLM.from_pretrained(MODEL_ID, torch_dtype=torch.bfloat16, device_map=DEVICE)
-tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, padding_side='left')
+print(f"Loading model: {MODEL_CONFIG['MODEL_ID']}")
+model = AutoModelForCausalLM.from_pretrained(MODEL_CONFIG['MODEL_ID'], torch_dtype=torch.bfloat16, device_map=DEVICE)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_CONFIG['MODEL_ID'], padding_side='left')
 tokenizer.pad_token = tokenizer.eos_token
 print("Model and tokenizer loaded successfully.")
 
 app = FastAPI()
 
-# --- Tool Execution Functions --- (Identique)
+# --- Tool Execution Functions ---
 def execute_browse_tool(url: str) -> str:
     try:
         headers = {'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36'}
@@ -58,7 +84,7 @@ def execute_browse_tool(url: str) -> str:
     except Exception as e:
         return f"Error browsing {url}: {str(e)}"
 
-# --- Pydantic Models --- (Identique)
+# --- Pydantic Models ---
 class ContentPart(BaseModel): type: str; text: str
 class ChatMessage(BaseModel): role: str; content: Union[str, List[ContentPart]]
 class ChatCompletionRequest(BaseModel):
@@ -77,7 +103,7 @@ def parse_tag(tag: str, text: str) -> str:
 # --- API Endpoints ---
 @app.get("/models", response_model=ModelList)
 async def list_models():
-    return ModelList(data=[ModelData(id=MODEL_ID)])
+    return ModelList(data=[ModelData(id=MODEL_CONFIG['MODEL_ID'])])
 
 @app.post("/chat/completions")
 async def create_chat_completion(request: ChatCompletionRequest):
@@ -94,26 +120,23 @@ async def create_chat_completion(request: ChatCompletionRequest):
         response_id = f"chatcmpl-{uuid.uuid4()}"
 
         def stream_chunk(content: str):
-            chunk = {"id": response_id, "object": "chat.completion.chunk", "created": int(time.time()), "model": MODEL_ID, "choices": [{"index": 0, "delta": {"content": content}, "finish_reason": None}]}
+            chunk = {"id": response_id, "object": "chat.completion.chunk", "created": int(time.time()), "model": MODEL_CONFIG['MODEL_ID'], "choices": [{"index": 0, "delta": {"content": content}, "finish_reason": None}]}
             return f"data: {json.dumps(chunk)}\n\n"
 
-        # --- STEP 1: Planification ---
         initial_messages = [{'role': 'system', 'content': SYSTEM_PROMPT}, {'role': 'user', 'content': user_prompt}]
         formatted_prompt = tokenizer.apply_chat_template(initial_messages, tokenize=False, add_generation_prompt=True)
         inputs = tokenizer(formatted_prompt, return_tensors="pt", padding=True).to(DEVICE)
-        outputs = model.generate(**inputs, max_new_tokens=1024, eos_token_id=tokenizer.eos_token_id)
+        outputs = model.generate(**inputs, max_new_tokens=MODEL_CONFIG['MAX_TOKENS_PLAN'], eos_token_id=tokenizer.eos_token_id)
         agent_plan = tokenizer.decode(outputs[0][len(inputs['input_ids'][0]):], skip_special_tokens=True)
 
         thinking_text = parse_tag("thinking", agent_plan)
         tool_code_text = parse_tag("tool_code", agent_plan)
         final_answer_text = parse_tag("final_answer", agent_plan)
 
-        # --- STEP 2: Diffusion de la pensée ---
         if thinking_text:
             yield stream_chunk(f"🤔 **Thinking...**\n```thought\n{thinking_text}\n```\n\n")
             await asyncio.sleep(0.1)
 
-        # --- STEP 3: Action & Synthèse ---
         tool_call = None
         if tool_code_text:
             try:
@@ -122,7 +145,6 @@ async def create_chat_completion(request: ChatCompletionRequest):
                 pass
         
         if tool_call and 'tool' in tool_call:
-            # --- Exécution et Diffusion de l'Action ---
             if tool_call['tool'] == 'browse' and 'url' in tool_call:
                 url = tool_call['url']
                 yield stream_chunk(f"🔎 **Action:** Browsing `{url}`...\n\n")
@@ -131,7 +153,6 @@ async def create_chat_completion(request: ChatCompletionRequest):
             else:
                 tool_context = "Unknown tool requested."
             
-            # --- Appel de synthèse ---
             synthesis_messages = [
                 {'role': 'system', 'content': SYSTEM_PROMPT},
                 {'role': 'user', 'content': user_prompt},
@@ -140,18 +161,16 @@ async def create_chat_completion(request: ChatCompletionRequest):
             ]
             synthesis_prompt = tokenizer.apply_chat_template(synthesis_messages, tokenize=False, add_generation_prompt=True)
             synthesis_inputs = tokenizer(synthesis_prompt, return_tensors="pt", padding=True).to(DEVICE)
-            synthesis_outputs = model.generate(**synthesis_inputs, max_new_tokens=1024, do_sample=True, temperature=0.1, top_k=50, top_p=0.95, eos_token_id=tokenizer.eos_token_id)
+            synthesis_outputs = model.generate(**synthesis_inputs, max_new_tokens=MODEL_CONFIG['MAX_TOKENS_ANSWER'], do_sample=True, temperature=MODEL_CONFIG['TEMPERATURE'], top_k=50, top_p=0.95, eos_token_id=tokenizer.eos_token_id)
             final_response = tokenizer.decode(synthesis_outputs[0][len(synthesis_inputs['input_ids'][0]):], skip_special_tokens=True)
             final_answer_text = parse_tag("final_answer", final_response)
 
-        # --- STEP 4: Diffusion de la Réponse Finale ---
         if final_answer_text:
             yield stream_chunk(f"✅ **Final Answer:**\n{final_answer_text}")
         else:
             yield stream_chunk("Agent could not generate a final answer.")
         
-        # --- Fin du stream ---
-        final_chunk = {"id": response_id, "object": "chat.completion.chunk", "created": int(time.time()), "model": MODEL_ID, "choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}]}
+        final_chunk = {"id": response_id, "object": "chat.completion.chunk", "created": int(time.time()), "model": MODEL_CONFIG['MODEL_ID'], "choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}]}
         yield f"data: {json.dumps(final_chunk)}\n\n"
         yield "data: [DONE]\n\n"
 
@@ -159,4 +178,4 @@ async def create_chat_completion(request: ChatCompletionRequest):
 
 @app.get("/")
 def root():
-    return {"status": "Transparent Reasoning Agent for WordPress/WooCommerce is online", "model_id": MODEL_ID}
+    return {"status": "Configurable Reasoning Agent is online", "model_id": MODEL_CONFIG['MODEL_ID']}