Update app.py

app.py

@@ -210,7 +210,7 @@ def trans(text):
     return None
 
     # テキストに言語タグを付与し、バイト列に変換
-    input_text = f"<2ja>{text}".encode('utf-8')
+    input_text = bf"<2ja>{text}"
     
     # トークナイズ
     tokens = llm.tokenize(input_text)
@@ -284,12 +284,12 @@ def respond(
     repeat_penalty: float,
 ):
     llama = Llama("models/madlad400-3b-mt-q8_0.gguf")
-    #tokens = llama.tokenize(f"{message}".encode('utf-8'))#
+    #tokens = llama.tokenize(bf"<2ja>{message}")#
     tokens = llama.tokenize(b"What is the capital of France?")
     llama.encode(tokens)
     tokens = [llama.decoder_start_token()]
     for token in llama.generate(tokens, top_k=40, top_p=0.95, temp=1, repeat_penalty=1.0):
-        yield (llama.detokenize([token]))
+        yield (llama.detokenize([token]).decode())
         if token == llama.token_eos():
             break