モデルを変えてみた

app.py

@@ -50,9 +50,10 @@ index.add(np.array(doc_embeddings))
 
 def generate_text(prompt):
     full_prompt = search(prompt)
-    input_ids = tokenizer(full_prompt, return_tensors="pt").to("cpu")  # または .to("cuda") / .to("mps") / .to("cpu") に変更
+    inputs = tokenizer(full_prompt, return_tensors="pt")
+    inputs = {k: v.to("cpu") for k, v in inputs.items()}
     with torch.no_grad():
-        output_ids = llm.generate(input_ids, max_new_tokens=256)
+        output_ids = llm.generate(**inputs, max_new_tokens=256)
     result_text = tokenizer.decode(output_ids[0], skip_special_tokens=True)
     return result_text