Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from huggingface_hub import InferenceClient
|
3 |
+
from sentence_transformers import SentenceTransformer
|
4 |
+
import faiss
|
5 |
+
import numpy as np
|
6 |
+
import pdfplumber
|
7 |
+
|
8 |
+
# Initialize the InferenceClient
|
9 |
+
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
|
10 |
+
|
11 |
+
# Function to extract text from PDFs
|
12 |
+
def extract_text_from_pdf(pdf_path):
|
13 |
+
text = ""
|
14 |
+
with pdfplumber.open(pdf_path) as pdf:
|
15 |
+
for page in pdf.pages:
|
16 |
+
page_text = page.extract_text()
|
17 |
+
if page_text:
|
18 |
+
text += page_text
|
19 |
+
return text
|
20 |
+
|
21 |
+
# Load and preprocess book PDFs
|
22 |
+
pdf_files = ["Diagnostic and statistical manual of mental disorders _ DSM-5 ( PDFDrive.com ).pdf"]
|
23 |
+
all_texts = [extract_text_from_pdf(pdf) for pdf in pdf_files]
|
24 |
+
|
25 |
+
# Split text into chunks
|
26 |
+
def chunk_text(text, chunk_size=300):
|
27 |
+
sentences = text.split('. ')
|
28 |
+
chunks, current_chunk = [], ""
|
29 |
+
for sentence in sentences:
|
30 |
+
if len(current_chunk) + len(sentence) <= chunk_size:
|
31 |
+
current_chunk += sentence + ". "
|
32 |
+
else:
|
33 |
+
chunks.append(current_chunk.strip())
|
34 |
+
current_chunk = sentence + ". "
|
35 |
+
if current_chunk:
|
36 |
+
chunks.append(current_chunk.strip())
|
37 |
+
return chunks
|
38 |
+
|
39 |
+
# Prepare embeddings for each book
|
40 |
+
model = SentenceTransformer("all-MiniLM-L6-v2")
|
41 |
+
index = faiss.IndexFlatL2(model.get_sentence_embedding_dimension())
|
42 |
+
chunked_texts = [chunk_text(text) for text in all_texts]
|
43 |
+
all_chunks = [chunk for chunks in chunked_texts for chunk in chunks]
|
44 |
+
embeddings = model.encode(all_chunks, convert_to_tensor=True).detach().cpu().numpy()
|
45 |
+
index.add(embeddings)
|
46 |
+
|
47 |
+
# Function to generate response
|
48 |
+
def respond(message, history, system_message, max_tokens, temperature, top_p):
|
49 |
+
# Step 1: Retrieve relevant chunks based on user message
|
50 |
+
query_embedding = model.encode([message], convert_to_tensor=True).detach().cpu().numpy()
|
51 |
+
k = 5
|
52 |
+
_, indices = index.search(query_embedding, k)
|
53 |
+
relevant_chunks = " ".join([all_chunks[idx] for idx in indices[0]])
|
54 |
+
|
55 |
+
# Step 2: Create prompt for the model
|
56 |
+
prompt = f"{system_message}\n\nUser Query: {message}\n\nRelevant Information: {relevant_chunks}"
|
57 |
+
response = ""
|
58 |
+
|
59 |
+
# Step 3: Generate response
|
60 |
+
for message in client.chat_completion(
|
61 |
+
[{"role": "system", "content": system_message}, {"role": "user", "content": message}],
|
62 |
+
max_tokens=max_tokens,
|
63 |
+
stream=True,
|
64 |
+
temperature=temperature,
|
65 |
+
top_p=top_p,
|
66 |
+
):
|
67 |
+
token = message.choices[0].delta.content
|
68 |
+
response += token
|
69 |
+
yield response
|
70 |
+
|
71 |
+
# Gradio ChatInterface with additional inputs
|
72 |
+
demo = gr.ChatInterface(
|
73 |
+
respond,
|
74 |
+
additional_inputs=[
|
75 |
+
gr.Textbox(value="You are a helpful and empathetic mental health assistant.", label="System message"),
|
76 |
+
gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
|
77 |
+
gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
|
78 |
+
gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
|
79 |
+
],
|
80 |
+
)
|
81 |
+
|
82 |
+
# Launch the Gradio interface
|
83 |
+
if __name__ == "__main__":
|
84 |
+
demo.launch()
|