sarashinaをr2に変更

.idea/.gitignore

@@ -0,0 +1,10 @@
+# デフォルトの無視対象ファイル
+/shelf/
+/workspace.xml
+# エディターベースの HTTP クライアントリクエスト
+/httpRequests/
+# Datasource local storage ignored files
+/dataSources/
+/dataSources.local.xml
+
+.idea/

.idea/inspectionProfiles/profiles_settings.xml

@@ -0,0 +1,6 @@
+<component name="InspectionProjectProfileManager">
+  <settings>
+    <option name="USE_PROJECT_PROFILE" value="false" />
+    <version value="1.0" />
+  </settings>
+</component>

.idea/misc.xml

@@ -0,0 +1,7 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="Black">
+    <option name="sdkName" value="Python 3.9 (Clang-to-japanese)" />
+  </component>
+  <component name="ProjectRootManager" version="2" project-jdk-name="Python 3.9 (Clang-to-japanese)" project-jdk-type="Python SDK" />
+</project>

.idea/modules.xml

@@ -0,0 +1,8 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="ProjectModuleManager">
+    <modules>
+      <module fileurl="file://$PROJECT_DIR$/.idea/processing_test.iml" filepath="$PROJECT_DIR$/.idea/processing_test.iml" />
+    </modules>
+  </component>
+</project>

.idea/processing_test.iml

@@ -0,0 +1,8 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<module type="PYTHON_MODULE" version="4">
+  <component name="NewModuleRootManager">
+    <content url="file://$MODULE_DIR$" />
+    <orderEntry type="inheritedJdk" />
+    <orderEntry type="sourceFolder" forTests="false" />
+  </component>
+</module>

.idea/vcs.xml

@@ -0,0 +1,6 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="VcsDirectoryMappings">
+    <mapping directory="" vcs="Git" />
+  </component>
+</project>

app.py

@@ -1,15 +1,14 @@
 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
 
@@ -17,20 +16,19 @@ 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="Mori-kamiyama/sarashina2-13b-r1",
+    repo_id="Mori-kamiyama/sarashina2-13b-r2",
     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-small-en-v1.5")
@@ -50,12 +48,8 @@ index.add(np.array(doc_embeddings))
 
 def generate_text(prompt):
     full_prompt = search(prompt)
-    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(**inputs, max_new_tokens=256)
-    result_text = tokenizer.decode(output_ids[0], skip_special_tokens=True)
-    return result_text
+    output = llm(full_prompt, max_tokens=256)
+    return output["choices"][0]["text"]
 
 def search(query):
     query_embedding = model.encode([query], normalize_embeddings=True)
@@ -72,7 +66,9 @@ def search(query):
         print(f"→ {doc_text}")
 
     # RAG用のプロンプトを作成
-    prompt = "以下の文書を参照して質問に答えてください。\n\n文書:\n"
+    prompt = "以下の文書を参照して質問に答えてください。"
+    prompt += "生成は以下のフォーマットで回答しなさい\n<reasoning>\n...\n</reasoning>\n\n<answer>\n...\n</answer>"
+    prompt += "\n\n文書:\n"
     prompt += "\n".join(retrieved_docs)
     prompt += f"\n\n質問: {query}"
 

requirements.txt

@@ -1,7 +1,7 @@
 gradio
-#llama-cpp-python
-#huggingface_hub
-#pathlib
+llama-cpp-python
+huggingface_hub
+pathlib
 sentence-transformers
 faiss-cpu
 numpy