Upload 4 files

README.md

@@ -1,12 +1,3 @@
----
-title: Lawllm
-emoji: 📚
-colorFrom: red
-colorTo: purple
-sdk: streamlit
-sdk_version: 1.31.0
-app_file: app.py
-pinned: false
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# lawllm
+Law LLM Model to work on Indian Judiciary Acts, Orders, Provisions and Citations
+![image](https://github.com/suchinth08/lawllm/assets/21136148/9e47d810-c3b4-487a-9663-07ad9b3186a5)

lawmain.py

@@ -0,0 +1,29 @@
+import streamlit as st
+from PIL import Image
+from lppchain import get_lpphelper_chain,process_llm_response
+
+#st.title( "Lakna Reddy & Associates 🤖")
+col1, mid, col2 = st.columns(3)
+image = Image.open('lawimage2.jpg')
+with col1:
+    st.image(image, width=150)
+with col2:
+    st.markdown("## Lakna Reddy & Associates")    
+
+question = st.text_input("Question: ")
+@st.cache_resource
+def load_qa_chain():
+    chain = get_lpphelper_chain()
+    return chain
+
+if question:
+    chain = load_qa_chain()
+    #response = chain.run(question)
+    #llm_response = process_llm_response(response)
+    with st.spinner('Generating response...'):
+        response = chain.invoke(question)
+        print(response)
+        #answer = response['result']
+        answer = process_llm_response(response)
+        st.header("Answer")        
+        st.write(answer.replace("<pad>",""))

lppchain.py

@@ -0,0 +1,58 @@
+import transformers
+import torch
+import os
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+from transformers import pipeline
+from langchain.llms import HuggingFacePipeline
+from langchain.vectorstores import Chroma
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.chains import RetrievalQA
+from langchain.document_loaders import TextLoader
+from langchain.document_loaders import PyPDFLoader
+from langchain.document_loaders import DirectoryLoader
+from InstructorEmbedding import INSTRUCTOR
+from langchain.embeddings import HuggingFaceInstructEmbeddings
+from langchain_community.vectorstores import Chroma
+import textwrap
+import streamlit as st
+
+persist_directory = 'db'
+instructor_embeddings = HuggingFaceInstructEmbeddings(model_name="hkunlp/instructor-base") 
+embedding = instructor_embeddings
+tokenizer = AutoTokenizer.from_pretrained("lmsys/fastchat-t5-3b-v1.0")
+model = AutoModelForSeq2SeqLM.from_pretrained("lmsys/fastchat-t5-3b-v1.0")
+pipe = pipeline("text2text-generation",model=model, tokenizer=tokenizer, max_length=256)
+local_llm = HuggingFacePipeline(pipeline=pipe)
+vectordb = Chroma(persist_directory=persist_directory,embedding_function=embedding)
+retriever = vectordb.as_retriever(search_kwargs={"k": 3})
+
+def get_lpphelper_chain():
+    qa_chain = RetrievalQA.from_chain_type(llm=local_llm, 
+                                  chain_type="stuff", 
+                                  retriever=retriever, 
+                                  return_source_documents=True)
+    return qa_chain
+
+def wrap_text_preserve_newlines(text, width=110):
+    # 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):
+    wrap_text = wrap_text_preserve_newlines(llm_response['result'])   
+    sources = '\n\nSources:'
+    print('\n\nSources:')
+    for source in llm_response["source_documents"]:
+        sources.join(source.metadata['source'])  
+    print(wrap_text.join(sources))    
+    return wrap_text.replace("<pad>","")
+
+if __name__=="__main__":
+    get_lpphelper_chain()

lpphelper.py

@@ -0,0 +1,50 @@
+import transformers
+import torch
+import os
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+from transformers import pipeline
+from langchain.llms import HuggingFacePipeline
+from langchain.vectorstores import Chroma
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.chains import RetrievalQA
+from langchain.document_loaders import TextLoader
+from langchain.document_loaders import PyPDFLoader
+from langchain.document_loaders import DirectoryLoader
+from InstructorEmbedding import INSTRUCTOR
+from langchain.embeddings import HuggingFaceInstructEmbeddings
+from langchain_community.vectorstores import Chroma
+import textwrap
+
+def gen_vectordb():
+    tokenizer = AutoTokenizer.from_pretrained("lmsys/fastchat-t5-3b-v1.0")
+    model = AutoModelForSeq2SeqLM.from_pretrained("lmsys/fastchat-t5-3b-v1.0")
+    pipe = pipeline(
+        "text2text-generation",
+        model=model, 
+        tokenizer=tokenizer, 
+        max_length=256
+    )
+
+    local_llm = HuggingFacePipeline(pipeline=pipe)
+    loader = DirectoryLoader('C:/Users/SudheerRChinthala/sivallm/new_papers', glob="./*.pdf", loader_cls=PyPDFLoader)
+    documents = loader.load()
+    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
+    texts = text_splitter.split_documents(documents)
+
+    instructor_embeddings = HuggingFaceInstructEmbeddings(model_name="hkunlp/instructor-base") 
+    persist_directory = 'db'
+    embedding = instructor_embeddings
+    vectordb = Chroma.from_documents(documents=texts, 
+                                    embedding=embedding,
+                                    persist_directory=persist_directory)
+    retriever = vectordb.as_retriever(search_kwargs={"k": 3})
+    qa_chain = RetrievalQA.from_chain_type(llm=local_llm, 
+                                    chain_type="stuff", 
+                                    retriever=retriever, 
+                                    return_source_documents=True)
+    vectordb.persist()
+    vectordb = None 
+
+
+if __name__=="__main__":
+    gen_vectordb()