Deploy FastAPI Swagger Space (Docker)

app.py

@@ -7,6 +7,7 @@ from pydantic import BaseModel, Field
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
 MODEL_REPO = "aagzamov/search-query-parser"
+TOKENIZER_REPO = "Qwen/Qwen2.5-1.5B-Instruct"
 MAX_NEW_TOKENS = 192
 
 SortType = Literal["relevance", "price_asc", "price_desc", "newest"]
@@ -87,7 +88,7 @@ model = None
 
 def load_model():
     global tokenizer, model
-    tokenizer = AutoTokenizer.from_pretrained(MODEL_REPO, use_fast=False)
+    tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_REPO, use_fast=False)
     model = AutoModelForCausalLM.from_pretrained(
         MODEL_REPO,
         torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,