Spaces:
Build error
Build error
Update app.py
Browse files
app.py
CHANGED
|
@@ -10,48 +10,46 @@ from langchain.chains import ConversationalRetrievalChain
|
|
| 10 |
from htmlTemplates import css, bot_template, user_template
|
| 11 |
from langchain.llms import HuggingFaceHub, LlamaCpp
|
| 12 |
from huggingface_hub import snapshot_download, hf_hub_download
|
| 13 |
-
|
|
|
|
| 14 |
|
| 15 |
repo_name = "IlyaGusev/saiga2_13b_gguf"
|
| 16 |
model_name = "model-q4_K.gguf"
|
| 17 |
-
|
| 18 |
snapshot_download(repo_id=repo_name, local_dir=".", allow_patterns=model_name)
|
| 19 |
|
|
|
|
| 20 |
def get_pdf_text(pdf_docs):
|
| 21 |
-
|
| 22 |
text = ""
|
| 23 |
for pdf in pdf_docs:
|
| 24 |
pdf_reader = PdfReader(pdf)
|
| 25 |
for page in pdf_reader.pages:
|
| 26 |
text += page.extract_text()
|
| 27 |
-
|
| 28 |
return text
|
| 29 |
|
| 30 |
|
| 31 |
def get_text_chunks(text):
|
| 32 |
-
|
| 33 |
text_splitter = CharacterTextSplitter(separator="\n",
|
| 34 |
-
chunk_size=500,
|
| 35 |
-
chunk_overlap=30,
|
| 36 |
length_function=len
|
| 37 |
-
|
| 38 |
chunks = text_splitter.split_text(text)
|
| 39 |
-
|
| 40 |
return chunks
|
| 41 |
|
| 42 |
|
| 43 |
def get_vectorstore(text_chunks):
|
| 44 |
-
|
| 45 |
-
#embeddings = OpenAIEmbeddings()
|
| 46 |
embeddings = HuggingFaceInstructEmbeddings(model_name="hkunlp/instructor-xl")
|
| 47 |
-
#embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/paraphrase-multilingual-mpnet-base-v2")
|
| 48 |
vectorstore = FAISS.from_texts(texts=text_chunks, embedding=embeddings)
|
| 49 |
-
|
| 50 |
return vectorstore
|
| 51 |
|
| 52 |
|
| 53 |
def get_conversation_chain(vectorstore, model_name):
|
| 54 |
-
|
| 55 |
llm = LlamaCpp(model_path=model_name,
|
| 56 |
temperature=0.1,
|
| 57 |
top_k=30,
|
|
@@ -60,26 +58,27 @@ def get_conversation_chain(vectorstore, model_name):
|
|
| 60 |
n_ctx=2048,
|
| 61 |
n_parts=1,
|
| 62 |
echo=True
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
#llm = ChatOpenAI()
|
| 66 |
|
|
|
|
|
|
|
| 67 |
memory = ConversationBufferMemory(memory_key='chat_history', return_messages=True)
|
| 68 |
|
| 69 |
conversation_chain = ConversationalRetrievalChain.from_llm(llm=llm,
|
| 70 |
-
#condense_question_prompt=CONDENSE_QUESTION_PROMPT,
|
| 71 |
retriever=vectorstore.as_retriever(),
|
| 72 |
memory=memory,
|
| 73 |
return_source_documents=True
|
| 74 |
-
|
| 75 |
|
| 76 |
-
|
|
|
|
|
|
|
| 77 |
|
| 78 |
|
| 79 |
def handle_userinput(user_question):
|
| 80 |
-
|
| 81 |
response = st.session_state.conversation({'question': user_question})
|
| 82 |
-
|
| 83 |
st.session_state.chat_history = response['chat_history']
|
| 84 |
|
| 85 |
for i, message in enumerate(st.session_state.chat_history):
|
|
@@ -90,6 +89,7 @@ def handle_userinput(user_question):
|
|
| 90 |
st.write(bot_template.replace(
|
| 91 |
"{{MSG}}", message.content), unsafe_allow_html=True)
|
| 92 |
|
|
|
|
| 93 |
# main code
|
| 94 |
load_dotenv()
|
| 95 |
|
|
@@ -126,3 +126,4 @@ with st.sidebar:
|
|
| 126 |
# create conversation chain
|
| 127 |
st.session_state.conversation, retrieved_docs = get_conversation_chain(vectorstore, model_name)
|
| 128 |
st.text_area(retrieved_docs)
|
|
|
|
|
|
| 10 |
from htmlTemplates import css, bot_template, user_template
|
| 11 |
from langchain.llms import HuggingFaceHub, LlamaCpp
|
| 12 |
from huggingface_hub import snapshot_download, hf_hub_download
|
| 13 |
+
|
| 14 |
+
# from prompts import CONDENSE_QUESTION_PROMPT
|
| 15 |
|
| 16 |
repo_name = "IlyaGusev/saiga2_13b_gguf"
|
| 17 |
model_name = "model-q4_K.gguf"
|
| 18 |
+
|
| 19 |
snapshot_download(repo_id=repo_name, local_dir=".", allow_patterns=model_name)
|
| 20 |
|
| 21 |
+
|
| 22 |
def get_pdf_text(pdf_docs):
|
|
|
|
| 23 |
text = ""
|
| 24 |
for pdf in pdf_docs:
|
| 25 |
pdf_reader = PdfReader(pdf)
|
| 26 |
for page in pdf_reader.pages:
|
| 27 |
text += page.extract_text()
|
| 28 |
+
|
| 29 |
return text
|
| 30 |
|
| 31 |
|
| 32 |
def get_text_chunks(text):
|
|
|
|
| 33 |
text_splitter = CharacterTextSplitter(separator="\n",
|
| 34 |
+
chunk_size=500, # 1000
|
| 35 |
+
chunk_overlap=30, # 200
|
| 36 |
length_function=len
|
| 37 |
+
)
|
| 38 |
chunks = text_splitter.split_text(text)
|
| 39 |
+
|
| 40 |
return chunks
|
| 41 |
|
| 42 |
|
| 43 |
def get_vectorstore(text_chunks):
|
| 44 |
+
# embeddings = OpenAIEmbeddings()
|
|
|
|
| 45 |
embeddings = HuggingFaceInstructEmbeddings(model_name="hkunlp/instructor-xl")
|
| 46 |
+
# embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/paraphrase-multilingual-mpnet-base-v2")
|
| 47 |
vectorstore = FAISS.from_texts(texts=text_chunks, embedding=embeddings)
|
| 48 |
+
|
| 49 |
return vectorstore
|
| 50 |
|
| 51 |
|
| 52 |
def get_conversation_chain(vectorstore, model_name):
|
|
|
|
| 53 |
llm = LlamaCpp(model_path=model_name,
|
| 54 |
temperature=0.1,
|
| 55 |
top_k=30,
|
|
|
|
| 58 |
n_ctx=2048,
|
| 59 |
n_parts=1,
|
| 60 |
echo=True
|
| 61 |
+
)
|
|
|
|
|
|
|
| 62 |
|
| 63 |
+
# llm = ChatOpenAI()
|
| 64 |
+
|
| 65 |
memory = ConversationBufferMemory(memory_key='chat_history', return_messages=True)
|
| 66 |
|
| 67 |
conversation_chain = ConversationalRetrievalChain.from_llm(llm=llm,
|
| 68 |
+
# condense_question_prompt=CONDENSE_QUESTION_PROMPT,
|
| 69 |
retriever=vectorstore.as_retriever(),
|
| 70 |
memory=memory,
|
| 71 |
return_source_documents=True
|
| 72 |
+
)
|
| 73 |
|
| 74 |
+
result = conversation_chain
|
| 75 |
+
|
| 76 |
+
return result, result['source_documents'][0]
|
| 77 |
|
| 78 |
|
| 79 |
def handle_userinput(user_question):
|
|
|
|
| 80 |
response = st.session_state.conversation({'question': user_question})
|
| 81 |
+
|
| 82 |
st.session_state.chat_history = response['chat_history']
|
| 83 |
|
| 84 |
for i, message in enumerate(st.session_state.chat_history):
|
|
|
|
| 89 |
st.write(bot_template.replace(
|
| 90 |
"{{MSG}}", message.content), unsafe_allow_html=True)
|
| 91 |
|
| 92 |
+
|
| 93 |
# main code
|
| 94 |
load_dotenv()
|
| 95 |
|
|
|
|
| 126 |
# create conversation chain
|
| 127 |
st.session_state.conversation, retrieved_docs = get_conversation_chain(vectorstore, model_name)
|
| 128 |
st.text_area(retrieved_docs)
|
| 129 |
+
|