File size: 2,284 Bytes
8b582d7
921072e
6c4a2c9
 
 
 
8e2a2a1
921072e
6c4a2c9
 
 
86d4425
6c4a2c9
 
86d4425
6c4a2c9
 
 
86d4425
6c4a2c9
 
 
 
 
 
921072e
6c4a2c9
 
 
 
 
 
 
 
 
 
 
921072e
 
8e2a2a1
921072e
6c4a2c9
 
 
 
 
 
 
 
921072e
6c4a2c9
 
 
921072e
 
 
 
 
 
6c4a2c9
921072e
 
 
5cb1a88
 
921072e
 
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
import gradio as gr
import pandas as pd
import numpy as np
from transformers import pipeline, BertTokenizer, BertModel
import faiss
import torch
import spaces
# Load CSV data
data = pd.read_csv('RB10kstats.csv')

# Convert embedding column from string to numpy array
data['embeddings'] = data['embeddings'].apply(lambda x: np.fromstring(x[1:-1], sep=', '))

# Initialize FAISS index
dimension = len(data['embeddings'][0])
res = faiss.StandardGpuResources()  # use a single GPU
index = faiss.IndexFlatL2(dimension)
gpu_index = faiss.index_cpu_to_gpu(res, 0, index)  # move to GPU
gpu_index.add(np.stack(data['embeddings'].values))

# Check if GPU is available
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# Load QA model
qa_model = pipeline("question-answering", model="distilbert-base-uncased-distilled-squad", device=0 if torch.cuda.is_available() else -1)

# Load BERT model and tokenizer
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertModel.from_pretrained('bert-base-uncased').to(device)

# Function to embed the question using BERT
@spaces.GPU(duration=120)
def embed_question(question, model, tokenizer):
    inputs = tokenizer(question, return_tensors='pt').to(device)
    with torch.no_grad():
        outputs = model(**inputs)
    return outputs.last_hidden_state.mean(dim=1).cpu().numpy()

# Function to retrieve the relevant document and generate a response
@spaces.GPU(duration=120)
def retrieve_and_generate(question):
    # Embed the question
    question_embedding = embed_question(question, model, tokenizer)
    
    # Search in FAISS index
    _, indices = gpu_index.search(question_embedding, k=1)
    
    # Retrieve the most relevant document
    relevant_doc = data.iloc[indices[0][0]]
    
    # Use the QA model to generate the answer
    context = relevant_doc['Abstract']
    response = qa_model(question=question, context=context)
    
    return response['answer']

# Create a Gradio interface
interface = gr.Interface(
    fn=retrieve_and_generate,
    inputs=gr.inputs.Textbox(lines=2, placeholder="Ask a question about the documents..."),
    outputs="text",
    title="RAG Chatbot",
    description="Ask questions about the documents in the CSV file."
)

# Launch the Gradio app
interface.launch()