Update app.py

app.py

@@ -1,18 +1,3 @@
-# !pip install -U pymupdf
-# !pip install llama-index-embeddings-openai
-# !pip install llama-index-llms-openai
-# !pip install chromadb
-# !pip install llama-index-vector-stores-chroma
-# !pip install pydantic==1.10.11
-# !pip install llama-index-retrievers-bm25
-# !pip install sentence-transformers
-# !pip install llmlingua
-# !pip install accelerate
-# !pip install rouge 
-# !pip install semantic-text-similarity 
-# !pip install evaluate
-# !pip install streamlit
-
 import os
 import streamlit as st
 import streamlit.components.v1 as components
@@ -30,7 +15,7 @@ from llama_index.core.retrievers import BaseRetriever
 from llama_index.core.node_parser import SentenceSplitter
 from llama_index.embeddings.openai import OpenAIEmbedding
 from llmlingua import PromptCompressor
-from rouge import Rouge
+from rouge_score import rouge_scorer
 from semantic_text_similarity.models import WebBertSimilarity
 import nest_asyncio
 
@@ -98,7 +83,7 @@ if uploaded_files:
                 model=model,
                 messages=[
                     {"role":"system",
-                     "content":"You are a helpful assistant who answers from the following context. If the answer can't be found in context, just say that I don't know, don't try to make up an answer"
+                     "content":"You are a helpful assistant who answers from the following context. If the answer can't be found in context, politely refuse"
                     },
                     {"role": "user",
                      "content": prompt,
@@ -168,8 +153,8 @@ if uploaded_files:
                 st.markdown(compressed_res[3])
 
             # Save token summary and evaluation details to session state
-            rouge = Rouge()
-            scores = rouge.get_scores(compressed_res[3], orig_res[3])
+            scorer = rouge_scorer.RougeScorer(['rougeL'], use_stemmer=True)
+            scores = scorer.score(compressed_res[3],orig_res[3])
             webert_model = WebBertSimilarity(device='cpu')
             similarity_score = webert_model.predict([(compressed_res[3], orig_res[3])])[0] / 5 * 100
          
@@ -184,8 +169,8 @@ if uploaded_files:
 
             st.session_state.messages.append({"role": "assistant", "content": "Comparing Original and Optimized Prompt Response..."})
             st.success("Comparing Original and Optimized Prompt Response...")
-            st.session_state.messages.append({"role": "assistant", "content": f"Rouge Score : {scores[0]['rouge-l']['f'] * 100}"})
-            st.write(f"Rouge Score : {scores[0]['rouge-l']['f'] * 100}")
+            st.session_state.messages.append({"role": "assistant", "content": f"Rouge Score : {scores['rougeL'].fmeasure * 100}"})
+            st.write(f"Rouge Score : {scores['rougeL'].fmeasure * 100}")
             st.session_state.messages.append({"role": "assistant", "content": f"Semantic Text Similarity Score : {similarity_score}"})
             st.write(f"Semantic Text Similarity Score : {similarity_score}")
 
@@ -198,15 +183,3 @@ if uploaded_files:
             st.session_state.messages.append({"role": "assistant", "content": f"The optimized prompt has ${saving:.4f} saved in GPT-4."})
             st.success(f"The optimized prompt has ${saving:.4f} saved in GPT-4.")
 
-        
-### Future scope -
-
-# 1. Make this runnig in JPMC system.
-# 2. Scale it read multiple files at once.
-# 3. Cache the llm lingua roberta model to save time in downloading model every time.
-# 4. Play around with the llm lingua hyperparameters and observe changes in output and dollar value.
-
-### Refereces -
-
-# 1. https://docs.llamaindex.ai/en/stable/understanding/
-# 2. https://github.com/microsoft/LLMLingua/blob/main/examples/LLMLingua2.ipynb