Update app.py

app.py

@@ -6,59 +6,56 @@ os.system("pip install transformers torch psutil")
 
 # コマンドの実行結果を取得する（stdoutとstderrは出力されない）
 result = os.system("pip install transformers")
+以下のコードについて、テキスト生成のトークン生成数を3に設定し、生成されたテキストをinputとしてそれをx回繰り返すように設定。
+import os
 
+# コマンドを実行する
+os.system("pip install transformers torch psutil")
 
-import os
-from transformers import AutoModel, AutoTokenizer, trainer_utils, AutoConfig
+# コマンドの実行結果を取得する（stdoutとstderrは出力されない）
+result = os.system("pip install transformers")
+
+
+
+from transformers import AutoModel, AutoTokenizer, trainer_utils
 import gradio as gr
 import psutil
 
 device = "cpu"
-model_directory="./"
-model_path = "./pytorch_model.bin"  # モデルのパス
-conf_path="./config.json"
-if not os.path.exists(model_path):  # モデルが存在しない場合のみダウンロード
-    model_url = "https://huggingface.co/Tanrei/GPTSAN-japanese/resolve/main/pytorch_model.bin"
-    os.system(f"wget -O {model_path} {model_url}")
-    conf_url = "https://huggingface.co/Tanrei/GPTSAN-japanese/resolve/main/config.json"
-    os.system(f"wget -O {conf_path} {conf_url}")
-
-config = AutoConfig.from_pretrained(conf_path)
-model = AutoModel.from_pretrained(model_directory,config=config).to(device)
-tokenizer = AutoTokenizer.from_pretrained(Tanrei/GPTSAN-japanese)
+model = AutoModel.from_pretrained("Tanrei/GPTSAN-japanese").to(device)
+tokenizer = AutoTokenizer.from_pretrained("Tanrei/GPTSAN-japanese")
 trainer_utils.set_seed(30)
 
 def get_memory_usage():
     process = psutil.Process()
     memory_usage = process.memory_info().rss / 1024 / 1024  # メモリ使用量をMB単位で取得
     return f"Memory Usage: {memory_usage:.2f} MB"
-    
-def generate_text(input_text, repetition_count):
-    usag = get_memory_usage()
-    repetition_count = int(repetition_count)  # 数値型を整数型に変換
+  def generate_text(input_text, num_repeats):
+    usag=get_memory_usage()
     x_token = tokenizer("", prefix_text=input_text, return_tensors="pt")
     input_ids = x_token.input_ids.to(device)
     token_type_ids = x_token.token_type_ids.to(device)
-    generated_tokens = []
-    
-    for _ in range(repetition_count):
-        gen_token = model.generate(input_ids, token_type_ids=token_type_ids, max_new_tokens=3)
-        generated_tokens.append(gen_token)
-        input_ids = gen_token.clone().detach()
-    
-    output_text = tokenizer.decode(torch.cat(generated_tokens))
-    return output_text
-    
+    gen_token = model.generate(input_ids, token_type_ids=token_type_ids, max_new_tokens=10)
+    output_text = tokenizer.decode(gen_token[0])
+    repeated_text = output_text
+    for _ in range(num_repeats):
+        x_token = tokenizer("", prefix_text=repeated_text, return_tensors="pt")
+        input_ids = x_token.input_ids.to(device)
+        token_type_ids = x_token.token_type_ids.to(device)
+        gen_token = model.generate(input_ids, token_type_ids=token_type_ids, max_new_tokens=10)
+        repeated_text += tokenizer.decode(gen_token[0])
+    return repeated_text
+
+
+
 input_text = gr.inputs.Textbox(lines=5, label="Input Text")
-repetition_count = gr.inputs.Number(default=1, label="Repetition Count")
+num_repeats = gr.inputs.Number(default=1, label="Number of Repeats")
 output_text = gr.outputs.Textbox(label="Generated Text")
 
 interface = gr.Interface(
     fn=generate_text,
-    inputs=[input_text, repetition_count],
+    inputs=[input_text, num_repeats],
     outputs=output_text,
     title=get_memory_usage(),
     description="Enter a prompt in Japanese to generate text."
 )
-interface.launch()
-