モデルを変えてみた

app.py

@@ -1,23 +1,36 @@
 import gradio as gr
+"""
 import pathlib
 from huggingface_hub import hf_hub_download
 from llama_cpp import Llama
+"""
 from sentence_transformers import SentenceTransformer
 import faiss
 import numpy as np
 import pandas as pd
 
+## LLMの読み込み（Qwen2.5-3Bをsafetensorsで読み込み）
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+model_name = "Qwen/Qwen2.5-3B"
+tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
+llm = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, torch_dtype=torch.float16)
+llm.eval()
+
+"""
 ## LLMの読み込み
 models_dir = pathlib.Path(__file__).parent / "models"
 models_dir.mkdir(exist_ok=True)
 
 model_path = hf_hub_download(
-    repo_id="meta-llama/Llama-3.2-3B-Instruct", # Mori-kamiyama/sarashina2-13b-r1
+    repo_id="Mori-kamiyama/sarashina2-13b-r1",
     filename="model.gguf",
     local_dir=models_dir
 )
 
 llm = Llama(model_path=model_path)
+"""
 
 ## 埋め込みモデルの読み込み
 model = SentenceTransformer("BAAI/bge-m3")

requirements.txt

@@ -5,4 +5,6 @@ pathlib
 sentence-transformers
 faiss-cpu
 numpy
-pandas
+pandas
+pytorch
+transformers