Update app.py

app.py

@@ -1,12 +1,30 @@
-
 import gradio as gr
-import spaces
-import transformers_gradio
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+# Base modelとadapterの設定
+base_model_id = "unsloth/Llama-3.2-3B"  # または適切なbase model
+adapter_model_id = "MakiAi/Llama-3.2-3B-Instruct-bnb-4bit-OKU_wiki_llama3.1_8b_inst_Reflexive_chunk200_overlap700-10epochs"
+
+# モデルとトークナイザーの読み込み
+model = AutoModelForCausalLM.from_pretrained(
+    adapter_model_id,
+    torch_dtype=torch.float16,
+    device_map="auto"
+)
+tokenizer = AutoTokenizer.from_pretrained(adapter_model_id)
 
-#demo = gr.load(name="MakiAi/Llama-3.2-3B-Instruct-bnb-4bit-OKU_wiki_llama3.1_8b_inst_Reflexive_chunk200_overlap700-10epochs", src=transformers_gradio.registry)
-demo = gr.load(name="unsloth/Llama-3.2-1B", src=transformers_gradio.registry)
+# Gradioインターフェースの作成
+def generate_text(prompt):
+    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+    outputs = model.generate(**inputs, max_new_tokens=100)
+    return tokenizer.decode(outputs[0], skip_special_tokens=True)
 
-demo.fn = spaces.GPU()(demo.fn)
+demo = gr.Interface(
+    fn=generate_text,
+    inputs="text",
+    outputs="text"
+)
 
 if __name__ == "__main__":
     demo.launch()