Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,82 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import os
|
3 |
+
import requests
|
4 |
+
|
5 |
+
# Ensure that the Groq API key is set
|
6 |
+
os.environ["GROQ_API_KEY"] = "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"
|
7 |
+
|
8 |
+
# Groq API endpoint
|
9 |
+
GROQ_API_URL = "https://api.groq.com/v1/inference"
|
10 |
+
|
11 |
+
# Function to perform embedding retrieval using MiniLM via Groq API
|
12 |
+
def retrieve_embedding(user_query):
|
13 |
+
payload = {
|
14 |
+
"model": "microsoft/MiniLM-L6-H384-uncased",
|
15 |
+
"input_text": user_query
|
16 |
+
}
|
17 |
+
headers = {
|
18 |
+
"Authorization": f"Bearer {os.getenv('GROQ_API_KEY')}"
|
19 |
+
}
|
20 |
+
response = requests.post(f"{GROQ_API_URL}/embedding", json=payload, headers=headers)
|
21 |
+
return response.json()["embedding"]
|
22 |
+
|
23 |
+
# Function to perform response generation using FLAN-T5 via Groq API
|
24 |
+
def generate_response(context):
|
25 |
+
payload = {
|
26 |
+
"model": "google/flan-t5-small",
|
27 |
+
"input_text": f"Given the following context, provide a supportive response: {context}"
|
28 |
+
}
|
29 |
+
headers = {
|
30 |
+
"Authorization": f"Bearer {os.getenv('GROQ_API_KEY')}"
|
31 |
+
}
|
32 |
+
response = requests.post(f"{GROQ_API_URL}/generate", json=payload, headers=headers)
|
33 |
+
return response.json()["text"]
|
34 |
+
|
35 |
+
# Load the counseling conversations dataset
|
36 |
+
from datasets import load_dataset
|
37 |
+
dataset = load_dataset("Amod/mental_health_counseling_conversations")["train"]
|
38 |
+
|
39 |
+
# Precompute embeddings for the dataset responses using Groq API
|
40 |
+
@st.cache(allow_output_mutation=True)
|
41 |
+
def embed_dataset(dataset):
|
42 |
+
embeddings = []
|
43 |
+
for entry in dataset:
|
44 |
+
embedding = retrieve_embedding(entry["response"])
|
45 |
+
embeddings.append(embedding)
|
46 |
+
return embeddings
|
47 |
+
|
48 |
+
dataset_embeddings = embed_dataset(dataset)
|
49 |
+
|
50 |
+
# Function to retrieve closest responses from the dataset using cosine similarity
|
51 |
+
from sklearn.metrics.pairwise import cosine_similarity
|
52 |
+
import numpy as np
|
53 |
+
|
54 |
+
def retrieve_response(user_query, dataset, dataset_embeddings, k=5):
|
55 |
+
query_embedding = retrieve_embedding(user_query)
|
56 |
+
cos_scores = cosine_similarity([query_embedding], dataset_embeddings)[0]
|
57 |
+
top_indices = np.argsort(cos_scores)[-k:][::-1]
|
58 |
+
|
59 |
+
retrieved_responses = []
|
60 |
+
for idx in top_indices:
|
61 |
+
retrieved_responses.append(dataset[idx]["response"])
|
62 |
+
return retrieved_responses
|
63 |
+
|
64 |
+
# Streamlit app UI
|
65 |
+
st.title("Emotional Support Buddy")
|
66 |
+
st.write("Enter your thoughts or concerns, and I'll provide some comforting words.")
|
67 |
+
|
68 |
+
# User input
|
69 |
+
user_query = st.text_input("How are you feeling today?")
|
70 |
+
|
71 |
+
if user_query:
|
72 |
+
# Retrieve similar responses from the dataset
|
73 |
+
retrieved_responses = retrieve_response(user_query, dataset, dataset_embeddings)
|
74 |
+
|
75 |
+
# Join retrieved responses to create a supportive context
|
76 |
+
context = " ".join(retrieved_responses)
|
77 |
+
|
78 |
+
# Generate a supportive response using FLAN-T5 via Groq API
|
79 |
+
supportive_response = generate_response(context)
|
80 |
+
|
81 |
+
st.write("Here's some advice or support for you:")
|
82 |
+
st.write(supportive_response)
|