Update app.py

app.py

@@ -6,17 +6,21 @@ import torch
 
 app = FastAPI(title="Assistant IA Education Marocaine")
 
-# Chargement du modèle
 MODEL_NAME = "unsloth/mistral-7b-v0.3"
 LORA_NAME = "dohael/mistral-7b-education-maroc"
 
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_NAME,
-    torch_dtype=torch.float16,
-    device_map="auto"
+    dtype=torch.float16,
+    device_map="cpu"
 )
-model = PeftModel.from_pretrained(model, LORA_NAME)
+model = PeftModel.from_pretrained(
+    model, 
+    LORA_NAME,
+    is_trainable=False
+)
+model.eval()
 
 class Question(BaseModel):
     question: str
@@ -28,14 +32,15 @@ def root():
 @app.post("/ask")
 def ask(q: Question):
     prompt = f"<s>[INST] {q.question} [/INST]"
-    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
-    outputs = model.generate(
-        **inputs,
-        max_new_tokens=512,
-        temperature=0.7,
-        do_sample=True,
-        repetition_penalty=1.1
-    )
+    inputs = tokenizer(prompt, return_tensors="pt")
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=512,
+            temperature=0.7,
+            do_sample=True,
+            repetition_penalty=1.1
+        )
     response = tokenizer.decode(outputs[0], skip_special_tokens=True)
     response = response.split("[/INST]")[-1].strip()
     return {"question": q.question, "reponse": response}