Update space

Gemma_Model.py

@@ -2,6 +2,7 @@ from transformers import (
     AutoTokenizer,
     # BitsAndBytesConfig,
     Gemma3ForCausalLM,
+    GemmaTokenizerFast
     )
 import torch
 import json
@@ -20,11 +21,11 @@ class GemmaLLM:
             device_map="cpu",
             # quantization_config=quantization_config,
             # low_cpu_mem_usage=True,
-            # torch_dtype=torch.float16,
+            torch_dtype=torch.float16,
             token=os.environ.get("GEMMA_TOKEN"),
         ).eval()
 
-        self.tokenizer = AutoTokenizer.from_pretrained(model_id)
+        self.tokenizer = GemmaTokenizerFast.from_pretrained(model_id, token=os.environ.get("GEMMA_TOKEN"))
 
     def generate(self, message) -> str:
         print("Generating...")