Upload folder using huggingface_hub

README.md

@@ -1,12 +1,6 @@
 ---
-title: Langchain Try Kki
-emoji: 🏆
-colorFrom: yellow
-colorTo: green
-sdk: gradio
-sdk_version: 4.44.0
+title: langchain_try_kki
 app_file: app.py
-pinned: false
+sdk: gradio
+sdk_version: 4.39.0
 ---
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,146 @@
+import weaviate # vector DB
+from openai import OpenAI # LLM
+import PyPDF2 # pdf -> text
+import numpy as np
+from transformers import AutoModel, AutoTokenizer # Model, Tokenzier Load
+import gradio as gr # front-end(ui & ux)
+from sentence_transformers import SentenceTransformer # embedding
+
+# Weaviate에 사용할 클래스 스키마 정의
+def create_schema():
+    class_obj = {
+        "class": "PdfSentence",  # 클래스명
+        "properties": [
+            {
+                "name": "sentence",
+                "dataType": ["text"]
+            },
+            {
+                "name": "embedding",
+                "dataType": ["number[]"]  # 벡터 타입
+            }
+        ]
+    }
+    # 스키마 생성
+    db_client.schema.create_class(class_obj)
+
+# 스키마 확인 및 생성 / 기존에 선언한 스키마가 있는 경우 -> pass(넘어간다.)
+def ensure_schema():
+    schema = db_client.schema.get()
+    classes = [cls["class"] for cls in schema["classes"]]
+    print(classes)
+    if "PdfSentence" not in classes:
+        create_schema()
+
+
+# 스키마가 존재하지 않을 경우 생성
+# if not client.schema.contains({"class": "PdfSentence"}):
+#     create_schema()
+
+# PDF 텍스트 추출 함수
+def extract_text_from_pdf(pdf):
+    pdf_reader = PyPDF2.PdfReader(pdf)
+    text = ""
+    for page in pdf_reader.pages:
+        text += page.extract_text()
+    return text
+
+# 텍스트 임베딩 생성 함수
+def create_embeddings(text, model):
+    result = model.encode(text)
+    return result.astype(np.float64).tolist()
+
+
+# Weaviate에 데이터 저장
+def store_vectors_in_weaviate(sentences, embed_model):
+    with db_client.batch as batch:
+        for sentence in sentences:
+            try:
+                embedding = create_embeddings(sentence, embed_model)
+                # print("embedding", embedding)
+
+                # Weaviate에 문장과 벡터 저장
+                data_object = {
+                    "sentence": sentence,
+                    "embedding": embedding
+                }
+                batch.add_data_object(data_object, "PdfSentence")
+                print("success")
+            except Exception as e:
+                print(e)
+
+# 질문에 가장 유사한 문장 찾기
+def find_similar_sentence_in_weaviate(question_embedding):
+    near_vector = {
+        "vector": question_embedding
+    }
+
+    result = db_client.query.get("PdfSentence", ["sentence", "embedding"]) \
+        .with_near_vector(near_vector) \
+        .do()
+    #         .with_limit(1) \
+    
+    return result # 가장 가까이 있는 문장이 반환
+    # return result['data']['Get']['PdfSentence'][0]['sentence']
+
+def generate_answer(pdf, question):
+    global embed_model
+    # PDF 텍스트 추출
+    text = extract_text_from_pdf(pdf)
+    
+    # 텍스트를 문장별로 나누기
+    sentences = text.split('. ')
+    
+    # 문장들을 Weaviate에 저장
+    store_vectors_in_weaviate(sentences, embed_model)
+    
+    # 질문에 대한 임베딩 생성
+    question_embedding = create_embeddings(question, embed_model)
+    
+    # Weaviate에서 가장 유사한 문장 찾기
+    most_similar_sentence = find_similar_sentence_in_weaviate(question_embedding)
+    print("debug03")
+    print(most_similar_sentence)
+
+    # OpenAI API로 응답 생성
+    ai_client = OpenAI(api_key='sk-TWonV6ldIlpQzTtp5WDW3IiE1mJtQ5eP2p3arsIkDQT3BlbkFJ87T5N5D4WQFHo-QitD7sFOBL6360GxdKTNYpuPbV8A')
+    response = ai_client.chat.completions.create(
+        model="gpt-4o-mini",
+        messages=[
+            {"role": "system", "content": f"You are a helpful assistant. Answer based on context: {most_similar_sentence}"},
+            {"role": "user", "content": question}
+        ]
+    )
+    result = response.choices[0].message.content
+
+    return result
+
+# Gradio 인터페이스 구성
+def interface(pdf, question):
+    return generate_answer(pdf, question)
+
+if __name__=="__main__":
+    # 임베딩 모델 로드
+    embed_model = SentenceTransformer('xlm-r-100langs-bert-base-nli-stsb-mean-tokens')
+
+    # Weaviate 클라이언트 설정
+    db_client = weaviate.Client(
+        url="https://ildmdarvrfcox58ff2tipw.c0.us-west3.gcp.weaviate.cloud",  # 클러스터 URL
+        auth_client_secret=weaviate.AuthApiKey(api_key="SPmVOW99EWg8LkstmLlsKUSuSiHfoefcLQwS"),  # API 키 설정
+        # timeout_config=(5, 150)  # 타임아웃 설정 (선택 사항)
+    )
+
+    ensure_schema()
+
+    # Gradio UI 생성
+    with gr.Blocks() as demo:
+        pdf_input = gr.File(label="Upload PDF", type="filepath")
+        question_input = gr.Textbox(label="Ask a question", placeholder="What do you want to know?")
+        output = gr.Textbox(label="Answer")
+
+        submit_btn = gr.Button("Submit")
+        submit_btn.click(fn=interface, inputs=[pdf_input, question_input], outputs=output)
+
+    # 앱 실행
+    demo.launch()
+

requirements.txt

@@ -0,0 +1,6 @@
+weaviate-client==4.7.1
+openai==1.32.0
+PyPDF2==3.0.1
+transformers==4.43.1
+gradio==4.39.0
+sentence-transformers