Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import pandas as pd
|
3 |
+
from datasets import load_dataset
|
4 |
+
from sentence_transformers import SentenceTransformer
|
5 |
+
import faiss
|
6 |
+
import numpy as np
|
7 |
+
from groq import Groq
|
8 |
+
|
9 |
+
# Load dataset
|
10 |
+
@st.cache_data
|
11 |
+
def load_data():
|
12 |
+
dataset = load_dataset("FreedomIntelligence/RAG-Instruct", split="train")
|
13 |
+
df = pd.DataFrame(dataset)
|
14 |
+
return df[["instruction", "response"]]
|
15 |
+
|
16 |
+
# Generate embeddings and index
|
17 |
+
@st.cache_resource
|
18 |
+
def setup_faiss(data):
|
19 |
+
model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
|
20 |
+
embeddings = model.encode(data["instruction"].tolist())
|
21 |
+
index = faiss.IndexFlatL2(embeddings.shape[1])
|
22 |
+
index.add(np.array(embeddings))
|
23 |
+
return model, index, embeddings
|
24 |
+
|
25 |
+
# Retrieve relevant context
|
26 |
+
def retrieve_context(query, model, index, data, top_k=1):
|
27 |
+
query_vec = model.encode([query])
|
28 |
+
distances, indices = index.search(np.array(query_vec), top_k)
|
29 |
+
results = [data.iloc[i]["instruction"] + "\n\n" + data.iloc[i]["response"] for i in indices[0]]
|
30 |
+
return "\n\n".join(results)
|
31 |
+
|
32 |
+
# Call Groq LLM
|
33 |
+
def query_groq(context, query):
|
34 |
+
prompt = f"Context:\n{context}\n\nQuestion: {query}\n\nAnswer:"
|
35 |
+
client = Groq(api_key=st.secrets["GROQ_API_KEY"])
|
36 |
+
response = client.chat.completions.create(
|
37 |
+
messages=[{"role": "user", "content": prompt}],
|
38 |
+
model="llama-3-70b-8192"
|
39 |
+
)
|
40 |
+
return response.choices[0].message.content
|
41 |
+
|
42 |
+
# Streamlit UI
|
43 |
+
st.set_page_config(page_title="RAG Demo with Groq", layout="wide")
|
44 |
+
st.title("🧠 RAG App using Groq API + RAG-Instruct Dataset")
|
45 |
+
|
46 |
+
data = load_data()
|
47 |
+
model, index, _ = setup_faiss(data)
|
48 |
+
|
49 |
+
st.markdown("Ask a question based on the instruction-response knowledge base.")
|
50 |
+
|
51 |
+
# Optional queries
|
52 |
+
optional_queries = [
|
53 |
+
"How to use a specific API function?",
|
54 |
+
"Explain how to fine-tune a model.",
|
55 |
+
"What is the difference between pretraining and finetuning?",
|
56 |
+
"How does retrieval-augmented generation work?",
|
57 |
+
"Explain self-supervised learning."
|
58 |
+
]
|
59 |
+
|
60 |
+
query = st.text_input("Enter your question:", value=optional_queries[0])
|
61 |
+
if st.button("Ask"):
|
62 |
+
with st.spinner("Retrieving and generating response..."):
|
63 |
+
context = retrieve_context(query, model, index, data)
|
64 |
+
answer = query_groq(context, query)
|
65 |
+
st.subheader("📄 Retrieved Context")
|
66 |
+
st.write(context)
|
67 |
+
st.subheader("💬 Answer from Groq LLM")
|
68 |
+
st.write(answer)
|
69 |
+
|
70 |
+
st.markdown("### Optional Queries to Try:")
|
71 |
+
st.write(", ".join(optional_queries))
|