File size: 4,583 Bytes
cd8911f
 
 
 
 
 
526fc48
 
 
cd8911f
 
 
 
4a54e93
cd8911f
 
 
 
 
39ec79e
 
 
 
 
 
 
cd8911f
39ec79e
 
 
 
cd8911f
 
39ec79e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cd8911f
39ec79e
cd8911f
 
 
 
 
 
 
 
 
 
 
39ec79e
 
cd8911f
39ec79e
 
cd8911f
39ec79e
cd8911f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
75a7e4b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cd8911f
75a7e4b
 
 
cd8911f
75a7e4b
 
 
cd8911f
75a7e4b
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
import os
import torch
from huggingface_hub import InferenceClient
import streamlit as st
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# Streamlit page configuration 
st.set_page_config(page_title="Insight Snap & Summarizer")

# Load HF_TOKEN securely 
hf_token = os.getenv("HF_TOKEN") 

# Set up the Hugging Face Inference Client with the Bearer token
client = InferenceClient(api_key=hf_token)

# Model paths and IDs
model_id = "mistralai/Mistral-7B-Instruct-v0.3"
bart_model_path = "ChijoTheDatascientist/summarization-model"

# Cache the BART model and tokenizer
@st.cache_resource
def load_summarization_model():
    device = torch.device('cpu')  
    tokenizer = AutoTokenizer.from_pretrained(bart_model_path)
    model = AutoModelForSeq2SeqLM.from_pretrained(bart_model_path).to(device)
    return tokenizer, model

# Load the model and tokenizer 
bart_tokenizer, bart_model = load_summarization_model()

# Summarize reviews
@st.cache_data
def summarize_review(review_text):
    try:
        if len(review_text) > 1000:
            return "The review is too long for summarization. Please limit your text to about 1,000 characters, thank you!."
        inputs = bart_tokenizer(review_text, max_length=1024, truncation=True, return_tensors="pt")
        summary_ids = bart_model.generate(
            inputs["input_ids"], 
            max_length=40, 
            min_length=10, 
            length_penalty=2.0, 
            num_beams=8, 
            early_stopping=True
        )
        summary = bart_tokenizer.decode(summary_ids[0], skip_special_tokens=True)
        return f"Your review has been successfully summarized! Check the result below:\n\n{summary}"
    except Exception as e:
        return f"Something went wrong during the summarization process. Please try again. Error: {e}"


# Generate response 
def generate_response(system_message, user_input, chat_history, max_new_tokens=128):
    try:
        # Prepare the messages for the Hugging Face Inference API
        messages = [{"role": "user", "content": user_input}]
        completion = client.chat.completions.create(
            model=model_id, 
            messages=messages, 
            max_tokens=max_new_tokens,
        )
        response = completion.choices[0].message["content"]
        return response
    except ConnectionError:
        return "we're having trouble connecting to the server. Please try again later."
    except Exception as e:
        return f"Oops! Something went wrong: {e}"


# App configuration
st.title("Insight Snap & Summarizer")

st.markdown("""
- Use specific keywords in your queries to get targeted responses:
    - **"summarize"**: To summarize customer reviews.
    - **"Feedback or insights"**: Get actionable business insights based on feedback.
""")

# Initialize session state for chat history
if "chat_history" not in st.session_state:
    st.session_state.chat_history = []

# Chat interface
user_input = st.text_area("Enter customer reviews or a question:")

if st.button("Submit"):
    if user_input:
        # Summarize if the query is feedback-related
        if "summarize" in user_input.lower():
            summary = summarize_review(user_input)
            st.markdown(f"**Summary:** \n{summary}")
        elif "insight" in user_input.lower() or "feedback" in user_input.lower():
            system_message = (
                "You are a helpful assistant providing actionable insights "
                "from customer feedback to help businesses improve their services."
            )
            # Use the last summarized text if available
            last_summary = st.session_state.get("last_summary", "")
            query_input = last_summary if last_summary else user_input
            response = generate_response(system_message, query_input, st.session_state.chat_history)
            
            if response:
                # Update chat history
                st.session_state.chat_history.append({"role": "user", "content": user_input})
                st.session_state.chat_history.append({"role": "assistant", "content": response})
                st.markdown(f"**Insight:** \n{response}")
            else:
                st.warning("No response generated. Please try again later.")
        else:
            st.warning("Please specify if you want to 'summarize' or get 'insights'.")

        # Store the last summary for insights
        if "summarize" in user_input.lower():
            st.session_state["last_summary"] = summary
    else:
        st.warning("Please enter customer reviews or ask for insights.")