chore: restore lfs

.env.example

@@ -1,3 +1,3 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:707641c7ac139d1f0f8db0320a2368bea63b3d5ebd9d37c7a596ba092ebc2c7f
-size 79
+VSEGPT_KEY=sk-
+LLM_NAME=meta-llama/codellama-34b-instruct
+OPENAI_BASE=https://

.gitattributes

@@ -33,5 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
-.faiss filter=lfs diff=lfs merge=lfs -text
-*.txt filter=lfs diff=lfs merge=lfs -text
+*.faiss filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -1,3 +1,3 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:86b96044c2a4df69cbca929901ba6e653b179ea3a750d5b0733444417bd0ae92
-size 26
+venv/
+**/__pycache__/
+.env

README.md

@@ -1,3 +1,10 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:0a8c466b1ccb220e7443f02771a3404069c51c69c89e33503dbc1f78b0a7ef40
-size 129
+---
+title: Doc Eater
+emoji: 🚀
+colorFrom: red
+colorTo: green
+sdk: gradio
+sdk_version: 4.29.0
+app_file: app.py
+pinned: false
+---

app.py

@@ -1,3 +1,124 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:55ec3fd8747ee304421c487bfdea048b7d1b7765ee788663c4292e9b96082ded
-size 4758
+import gradio as gr
+
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_community.vectorstores import Clickhouse, ClickhouseSettings
+from langchain_community.vectorstores import FAISS
+from langchain_core.runnables import (
+                RunnableLambda,
+                RunnableParallel,
+                RunnablePassthrough,
+            )
+from langchain_core.output_parsers import StrOutputParser
+import warnings
+
+from lib.embedding import build_embedding
+from lib.model_builder import ModelBuilder
+warnings.filterwarnings('ignore')
+from dotenv import load_dotenv
+
+load_dotenv()
+
+def combine_vectors(vectors):
+    result = []
+    vec1_count = len(vectors["vector1"])
+    # vec2_count = len(vectors["vector2"])
+    for i in range(vec1_count):
+        if i < vec1_count:
+            result.append(vectors['vector1'][i])
+        # if i < vec2_count:
+        #     result.append(vectors['vector2'][i])
+    return result
+
+
+def deploy():
+    llm = ModelBuilder.createVseGptModel("openchat/openchat-7b", 0)
+    model_rag = ModelBuilder.createVseGptModel("cohere/command-r", 0)
+    
+    # embedding = build_embedding(model_name="sentence-transformers/LaBSE")
+    rag_emb = build_embedding("intfloat/multilingual-e5-large")
+     
+    # settings_13_04 = ClickhouseSettings(table="car_table_13_04")
+    # clickhouse = Clickhouse(embedding, config=settings_13_04)
+    faiss_db = FAISS.load_local("./data/faiss_nk_17_05", rag_emb, allow_dangerous_deserialization=True)
+    # clickh_retriever = clickhouse.as_retriever()
+    faiss_retriever = faiss_db.as_retriever()
+    
+    retrievers = RunnableParallel(
+        vector1=faiss_retriever,
+        # vector2=clickh_retriever
+    )
+
+    chain_multivec = RunnableParallel({
+        "original": RunnablePassthrough(),
+        "context": retrievers | RunnableLambda(combine_vectors)
+    })
+
+    template = """Answer the question in Russian based only on the following context:
+    {context}
+
+    Question: {original}
+    """
+    prompt = ChatPromptTemplate.from_template(template)
+    output_parser = StrOutputParser()
+
+    chain_answer = prompt | model_rag | output_parser
+    rag_chain = chain_multivec | RunnableParallel({
+        "original": RunnableLambda(lambda ctx: ctx['original']),
+        "sources": RunnableLambda(lambda ctx: ctx['context']),
+        "answer": chain_answer
+    })
+
+
+
+    def print_source_documents(documents):
+        return "\n\n".join([f"Взято из файла: {doc.metadata['file_name']} \n Metadata: {doc.metadata}"  for doc in documents])
+
+    with gr.Blocks(fill_height=True) as demo:
+        with gr.Row():
+            with gr.Column(scale=1):
+                chatbot_rag = gr.Chatbot(label="RAG: cohere/command-r + документы", height=600)
+            with gr.Column(scale=1):
+                chatbot_llm = gr.Chatbot(label="LLM standalone: openchat/openchat-7b", height=600)
+        chat_input = gr.MultimodalTextbox(interactive=True, file_types=None, placeholder="Введите сообщение...", show_label=False)
+        clear = gr.Button("Clear")
+
+        def user_rag(history, message):
+            if message["text"] is not None:
+                history.append((message["text"], None))
+            return history, gr.MultimodalTextbox(value=None, interactive=False)
+        
+        def user_llm(history, message):
+            if message["text"] is not None:
+                history.append((message["text"], None))
+            return history, gr.MultimodalTextbox(value=None, interactive=False)
+
+        def bot_rag(history):
+            result = rag_chain.invoke(history[-1][0])
+            form_answer = result["answer"].strip() 
+            history[-1][1] = form_answer
+            return history
+
+        def bot_llm(history):
+            result = llm.invoke(history[-1][0])
+            history[-1][1] = result.content.strip()
+            return history
+
+        chat_input.submit(user_rag,  [chatbot_rag, chat_input], [chatbot_rag, chat_input], queue=False).then(
+            bot_rag, chatbot_rag, chatbot_rag
+        ).then(lambda: gr.MultimodalTextbox(interactive=True), None, [chat_input])
+
+        chat_input.submit(user_llm, [chatbot_llm, chat_input], [chatbot_llm, chat_input], queue=False).then(
+            bot_llm, chatbot_llm, chatbot_llm
+        ).then(lambda: gr.MultimodalTextbox(interactive=True), None, [chat_input])
+        clear.click(lambda: None, None, chatbot_rag, queue=False)
+        clear.click(lambda: None, None, chatbot_llm, queue=False)
+
+    demo.launch(share=True)
+
+
+if __name__ == "__main__":
+    # parser = argparse.ArgumentParser(description='Deploy llm chat')
+    # parser.add_argument('--model_name', metavar='M', type=str,
+    #                 help='model name as: openai/gpt-3.5-turbo')
+    
+    deploy()

lib/embedding.py

@@ -1,3 +1,10 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:bbed70158a33431a0205b7bd969f652da3f3364a112b4b63f0322d75114d5e88
-size 429
+from langchain_community.embeddings import HuggingFaceEmbeddings
+from dotenv import load_dotenv
+
+load_dotenv()
+def build_embedding(model_name: str):
+    embedding = HuggingFaceEmbeddings(model_name=model_name, \
+                                    #    model_kwargs={"device": "cuda"}, \
+                                        encode_kwargs={"normalize_embeddings": True})
+    embedding.show_progress = True
+    return embedding

lib/model_builder.py

@@ -1,3 +1,12 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:11209717c09df9cbe99f6ffacc57daaab69f99cebb39738e7326fcff865c42f5
-size 380
+import os
+from langchain_openai import ChatOpenAI
+from dotenv import load_dotenv
+
+load_dotenv()
+VSEGPT_KEY = os.getenv('VSEGPT_KEY')
+OPENAI_BASE = os.getenv('OPENAI_BASE')
+
+class ModelBuilder:
+    def createVseGptModel(model, temperature):
+        return ChatOpenAI(temperature=temperature, model_name=model, \
+                          api_key=VSEGPT_KEY, base_url = OPENAI_BASE)