Arifzyn commited on
Commit
d3914ef
·
verified ·
1 Parent(s): e713c44

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +6 -7
app.py CHANGED
@@ -91,13 +91,12 @@ class ChatResponse(BaseModel):
91
  response: str
92
  usage: Dict[str, Any]
93
 
 
 
 
94
 
95
  @app.post("/chat", response_model=ChatResponse)
96
  async def chat(req: ChatRequest):
97
- # Pastikan model dimuat sebelum digunakan
98
- if model is None:
99
- load_model()
100
-
101
  if model is None:
102
  raise HTTPException(status_code=500, detail="Gagal memuat model")
103
 
@@ -150,9 +149,9 @@ async def chat(req: ChatRequest):
150
  # Set parameter generasi yang lebih sesuai
151
  generation_config = {
152
  'max_new_tokens': req.max_tokens,
153
- 'temperature': req.temperature,
154
- 'top_p': req.top_p,
155
- 'do_sample': True,
156
  'pad_token_id': tokenizer.eos_token_id
157
  }
158
 
 
91
  response: str
92
  usage: Dict[str, Any]
93
 
94
+ @app.on_event("startup")
95
+ async def startup_event():
96
+ load_model()
97
 
98
  @app.post("/chat", response_model=ChatResponse)
99
  async def chat(req: ChatRequest):
 
 
 
 
100
  if model is None:
101
  raise HTTPException(status_code=500, detail="Gagal memuat model")
102
 
 
149
  # Set parameter generasi yang lebih sesuai
150
  generation_config = {
151
  'max_new_tokens': req.max_tokens,
152
+ 'temperature': 0.7,
153
+ 'top_p': 0.9,
154
+ 'do_sample': False,
155
  'pad_token_id': tokenizer.eos_token_id
156
  }
157