Update app.py

app.py

@@ -2,22 +2,18 @@ import os
 import glob
 import textwrap
 import time
-
 import langchain
 import locale
-
-
 import gradio as gr
+
 locale.getpreferredencoding = lambda: "UTF-8"
-### loaders
-from langchain.document_loaders import PyPDFLoader, DirectoryLoader
 
-### splits
+from langchain.document_loaders import PyPDFLoader, DirectoryLoader
 from langchain.text_splitter import RecursiveCharacterTextSplitter
-
-### prompts
 from langchain import PromptTemplate, LLMChain
 
+from langchain.embeddings.huggingface import HuggingFaceEmbeddings
+
 ### vector stores
 from langchain.vectorstores import FAISS
 
@@ -38,27 +34,52 @@ from transformers import (
 
 sorted(glob.glob('/content/anatomy_vol_*'))
 
-class CFG:
-    # LLMs
-    model_name = 'llama2-13b-chat' # wizardlm, llama2-7b-chat, llama2-13b-chat, mistral-7B
-    temperature = 0
-    top_p = 0.95
-    repetition_penalty = 1.15
 
-    # splitting
-    split_chunk_size = 800
-    split_overlap = 0
+def wrap_text_preserve_newlines(text, width=700):
+    # Split the input text into lines based on newline characters
+    lines = text.split('\n')
 
-    # embeddings
-    embeddings_model_repo = 'sentence-transformers/all-MiniLM-L6-v2'
+    # Wrap each line individually
+    wrapped_lines = [textwrap.fill(line, width=width) for line in lines]
 
-    # similar passages
-    k = 6
+    # Join the wrapped lines back together using newline characters
+    wrapped_text = '\n'.join(wrapped_lines)
+
+    return wrapped_text
 
-    # paths
-    PDFs_path = '/content/'
-    Embeddings_path =  '/content/faiss-hp-sentence-transformers'
-    Output_folder = './rag-vectordb'
+
+def process_llm_response(llm_response):
+    ans = wrap_text_preserve_newlines(llm_response['result'])
+
+    sources_used = ' \n'.join(
+        [
+            source.metadata['source'].split('/')[-1][:-4]
+            + ' - page: '
+            + str(source.metadata['page'])
+            for source in llm_response['source_documents']
+        ]
+    )
+
+    ans = ans + '\n\nSources: \n' + sources_used
+    return ans
+
+def llm_ans(query):
+    start = time.time()
+
+    llm_response = qa_chain.invoke(query)
+    ans = process_llm_response(llm_response)
+
+    end = time.time()
+
+    time_elapsed = int(round(end - start, 0))
+    time_elapsed_str = f'\n\nTime elapsed: {time_elapsed} s'
+    return ans + time_elapsed_str
+
+
+
+ def predict(message, history):
+     output = str(llm_ans(message)).replace("\n", "<br/>")
+     return output
 
 def get_model(model = CFG.model_name):
 
@@ -251,6 +272,30 @@ def get_model(model = CFG.model_name):
 
     return tokenizer, model, max_len
 
+
+class CFG:
+    # LLMs
+    model_name = 'llama2-13b-chat' # wizardlm, llama2-7b-chat, llama2-13b-chat, mistral-7B
+    temperature = 0
+    top_p = 0.95
+    repetition_penalty = 1.15
+
+    # splitting
+    split_chunk_size = 800
+    split_overlap = 0
+
+    # embeddings
+    embeddings_model_repo = 'sentence-transformers/all-MiniLM-L6-v2'
+
+    # similar passages
+    k = 6
+
+    # paths
+    PDFs_path = '/content/'
+    Embeddings_path =  '/content/faiss-hp-sentence-transformers'
+    Output_folder = './rag-vectordb'
+
+
 tokenizer, model, max_len = get_model(model = CFG.model_name)
 
 pipe = pipeline(
@@ -285,8 +330,6 @@ text_splitter = RecursiveCharacterTextSplitter(
 
 texts = text_splitter.split_documents(documents)
 
-from langchain.embeddings.huggingface import HuggingFaceEmbeddings
-
 vectordb = FAISS.from_documents(
     texts,
     HuggingFaceEmbeddings(model_name='sentence-transformers/all-mpnet-base-v2')
@@ -323,56 +366,10 @@ qa_chain = RetrievalQA.from_chain_type(
     verbose = False
 )
 
-def wrap_text_preserve_newlines(text, width=700):
-    # Split the input text into lines based on newline characters
-    lines = text.split('\n')
-
-    # Wrap each line individually
-    wrapped_lines = [textwrap.fill(line, width=width) for line in lines]
-
-    # Join the wrapped lines back together using newline characters
-    wrapped_text = '\n'.join(wrapped_lines)
-
-    return wrapped_text
-
-
-def process_llm_response(llm_response):
-    ans = wrap_text_preserve_newlines(llm_response['result'])
-
-    sources_used = ' \n'.join(
-        [
-            source.metadata['source'].split('/')[-1][:-4]
-            + ' - page: '
-            + str(source.metadata['page'])
-            for source in llm_response['source_documents']
-        ]
-    )
-
-    ans = ans + '\n\nSources: \n' + sources_used
-    return ans
-
-def llm_ans(query):
-    start = time.time()
-
-    llm_response = qa_chain.invoke(query)
-    ans = process_llm_response(llm_response)
-
-    end = time.time()
-
-    time_elapsed = int(round(end - start, 0))
-    time_elapsed_str = f'\n\nTime elapsed: {time_elapsed} s'
-    return ans + time_elapsed_str
-
-
-
- def predict(message, history):
-     output = str(llm_ans(message)).replace("\n", "<br/>")
-     return output
-
  demo = gr.ChatInterface(
      predict,
      title = f' Open-Source LLM ({CFG.model_name}) Question Answering'
  )
 
  demo.queue()
- demo.launch() correct the code
+ demo.launch()