Update app.py
Browse files
app.py
CHANGED
|
@@ -1,16 +1,21 @@
|
|
| 1 |
import os
|
|
|
|
| 2 |
from fastapi import FastAPI, HTTPException
|
| 3 |
from pydantic import BaseModel
|
| 4 |
from llama_cpp import Llama
|
| 5 |
from fastapi.middleware.cors import CORSMiddleware
|
| 6 |
|
|
|
|
| 7 |
MODEL_PATH = "Phi-3-mini-4k-instruct-q4.gguf"
|
| 8 |
|
| 9 |
-
print("--- Vérification du modèle
|
| 10 |
-
if os.path.exists(MODEL_PATH):
|
| 11 |
-
print(
|
|
|
|
|
|
|
|
|
|
| 12 |
else:
|
| 13 |
-
print(
|
| 14 |
|
| 15 |
app = FastAPI()
|
| 16 |
|
|
@@ -22,14 +27,10 @@ app.add_middleware(
|
|
| 22 |
allow_headers=["*"],
|
| 23 |
)
|
| 24 |
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
llm = Llama(model_path=MODEL_PATH, n_ctx=2048, n_threads=2)
|
| 30 |
-
print("Modèle chargé en mémoire avec succès.")
|
| 31 |
-
except Exception as e:
|
| 32 |
-
print(f"ERREUR CRITIQUE lors du chargement : {e}")
|
| 33 |
|
| 34 |
class ChatRequest(BaseModel):
|
| 35 |
message: str
|
|
@@ -37,17 +38,10 @@ class ChatRequest(BaseModel):
|
|
| 37 |
|
| 38 |
@app.get("/")
|
| 39 |
def root_check():
|
| 40 |
-
return {"status": "BEGA AI API is running"
|
| 41 |
|
| 42 |
@app.post("/api/chat")
|
| 43 |
def chat_endpoint(req: ChatRequest):
|
| 44 |
-
global llm
|
| 45 |
-
if llm is None:
|
| 46 |
-
raise HTTPException(
|
| 47 |
-
status_code=500,
|
| 48 |
-
detail="Le modèle IA n'a pas pu être chargé au démarrage. Vérifiez les logs de build."
|
| 49 |
-
)
|
| 50 |
-
|
| 51 |
prompt = (
|
| 52 |
"You are BEGA AI, a helpful AI assistant developed by BEGA Company. "
|
| 53 |
"Reply in the same language used in the user's message.\n\n"
|
|
@@ -58,15 +52,11 @@ def chat_endpoint(req: ChatRequest):
|
|
| 58 |
|
| 59 |
prompt += f"User: {req.message}\nAssistant:"
|
| 60 |
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
return {"response": reply}
|
| 70 |
-
|
| 71 |
-
except Exception as e:
|
| 72 |
-
raise HTTPException(status_code=500, detail=str(e))
|
|
|
|
| 1 |
import os
|
| 2 |
+
import urllib.request
|
| 3 |
from fastapi import FastAPI, HTTPException
|
| 4 |
from pydantic import BaseModel
|
| 5 |
from llama_cpp import Llama
|
| 6 |
from fastapi.middleware.cors import CORSMiddleware
|
| 7 |
|
| 8 |
+
MODEL_URL = "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct-gguf/resolve/main/Phi-3-mini-4k-instruct-q4.gguf"
|
| 9 |
MODEL_PATH = "Phi-3-mini-4k-instruct-q4.gguf"
|
| 10 |
|
| 11 |
+
print("--- Vérification / Téléchargement du modèle ---")
|
| 12 |
+
if not os.path.exists(MODEL_PATH):
|
| 13 |
+
print("Le modèle est absent. Téléchargement direct sur le serveur en cours...")
|
| 14 |
+
# Cette ligne télécharge le fichier de 2.4 Go directement sur le disque cloud de Hugging Face
|
| 15 |
+
urllib.request.urlretrieve(MODEL_URL, MODEL_PATH)
|
| 16 |
+
print("Téléchargement terminé avec succès !")
|
| 17 |
else:
|
| 18 |
+
print("Le modèle est déjà présent sur le disque.")
|
| 19 |
|
| 20 |
app = FastAPI()
|
| 21 |
|
|
|
|
| 27 |
allow_headers=["*"],
|
| 28 |
)
|
| 29 |
|
| 30 |
+
# Chargement du modèle en mémoire RAM
|
| 31 |
+
print("Chargement de l'IA en RAM...")
|
| 32 |
+
llm = Llama(model_path=MODEL_PATH, n_ctx=2048, n_threads=2)
|
| 33 |
+
print("IA opérationnelle !")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 34 |
|
| 35 |
class ChatRequest(BaseModel):
|
| 36 |
message: str
|
|
|
|
| 38 |
|
| 39 |
@app.get("/")
|
| 40 |
def root_check():
|
| 41 |
+
return {"status": "BEGA AI API is running"}
|
| 42 |
|
| 43 |
@app.post("/api/chat")
|
| 44 |
def chat_endpoint(req: ChatRequest):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 45 |
prompt = (
|
| 46 |
"You are BEGA AI, a helpful AI assistant developed by BEGA Company. "
|
| 47 |
"Reply in the same language used in the user's message.\n\n"
|
|
|
|
| 52 |
|
| 53 |
prompt += f"User: {req.message}\nAssistant:"
|
| 54 |
|
| 55 |
+
output = llm(
|
| 56 |
+
prompt,
|
| 57 |
+
max_tokens=350,
|
| 58 |
+
stop=["User:", "\nUser:", "Assistant:", "\nAssistant:"],
|
| 59 |
+
temperature=0.3
|
| 60 |
+
)
|
| 61 |
+
reply = output['choices'][0]['text'].strip()
|
| 62 |
+
return {"response": reply}
|
|
|
|
|
|
|
|
|
|
|
|