Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -5,44 +5,54 @@ import faiss
|
|
5 |
import numpy as np
|
6 |
from transformers import pipeline
|
7 |
|
8 |
-
|
9 |
dataset = load_dataset("lex_glue", "scotus")
|
10 |
-
corpus = [doc['text'] for doc in dataset['train'].select(range(200))]
|
11 |
-
|
12 |
|
|
|
13 |
embedder = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
|
14 |
corpus_embeddings = embedder.encode(corpus, convert_to_numpy=True)
|
15 |
|
16 |
-
|
17 |
dimension = corpus_embeddings.shape[1]
|
18 |
index = faiss.IndexFlatL2(dimension)
|
19 |
index.add(corpus_embeddings)
|
20 |
|
21 |
-
|
22 |
gen_pipeline = pipeline("text2text-generation", model="facebook/bart-large-cnn")
|
23 |
|
24 |
-
|
25 |
def rag_query(user_question):
|
|
|
26 |
question_embedding = embedder.encode([user_question])
|
27 |
-
|
|
|
28 |
if index.ntotal < k:
|
29 |
-
k = index.ntotal
|
|
|
|
|
30 |
_, indices = index.search(np.array(question_embedding), k=k)
|
31 |
|
32 |
-
|
|
|
|
|
|
|
33 |
return "Sorry, no relevant documents were found."
|
|
|
|
|
|
|
34 |
|
35 |
-
|
36 |
-
|
37 |
prompt = f"Question: {user_question}\nContext: {context}\nAnswer:"
|
38 |
result = gen_pipeline(prompt, max_length=250, do_sample=False)[0]['generated_text']
|
|
|
39 |
return result
|
40 |
|
41 |
-
|
42 |
def chatbot_interface(query):
|
43 |
return rag_query(query)
|
44 |
|
45 |
-
|
46 |
css = """
|
47 |
.gradio-container {
|
48 |
background-color: #f0f4f8;
|
@@ -88,7 +98,7 @@ css = """
|
|
88 |
}
|
89 |
"""
|
90 |
|
91 |
-
|
92 |
iface = gr.Interface(
|
93 |
fn=chatbot_interface,
|
94 |
inputs="text",
|
@@ -99,5 +109,5 @@ iface = gr.Interface(
|
|
99 |
css=css
|
100 |
)
|
101 |
|
102 |
-
|
103 |
-
iface.launch()
|
|
|
5 |
import numpy as np
|
6 |
from transformers import pipeline
|
7 |
|
8 |
+
# Load dataset
|
9 |
dataset = load_dataset("lex_glue", "scotus")
|
10 |
+
corpus = [doc['text'] for doc in dataset['train'].select(range(200))] # just 200 to keep it light
|
|
|
11 |
|
12 |
+
# Embedding model
|
13 |
embedder = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
|
14 |
corpus_embeddings = embedder.encode(corpus, convert_to_numpy=True)
|
15 |
|
16 |
+
# Build FAISS index
|
17 |
dimension = corpus_embeddings.shape[1]
|
18 |
index = faiss.IndexFlatL2(dimension)
|
19 |
index.add(corpus_embeddings)
|
20 |
|
21 |
+
# Text generation model
|
22 |
gen_pipeline = pipeline("text2text-generation", model="facebook/bart-large-cnn")
|
23 |
|
24 |
+
# RAG-like query function
|
25 |
def rag_query(user_question):
|
26 |
+
# Encode the user question
|
27 |
question_embedding = embedder.encode([user_question])
|
28 |
+
|
29 |
+
k = 3 # top 3 documents
|
30 |
if index.ntotal < k:
|
31 |
+
k = index.ntotal # Adjust if there are fewer documents than requested
|
32 |
+
|
33 |
+
# Perform the search in the FAISS index
|
34 |
_, indices = index.search(np.array(question_embedding), k=k)
|
35 |
|
36 |
+
# Ensure indices are valid (within range of the corpus)
|
37 |
+
valid_indices = [i for i in indices[0] if i < len(corpus)]
|
38 |
+
|
39 |
+
if len(valid_indices) == 0:
|
40 |
return "Sorry, no relevant documents were found."
|
41 |
+
|
42 |
+
# Extract relevant context from the corpus based on valid indices
|
43 |
+
context = " ".join([corpus[i] for i in valid_indices])
|
44 |
|
45 |
+
# Prepare the prompt and generate the response
|
|
|
46 |
prompt = f"Question: {user_question}\nContext: {context}\nAnswer:"
|
47 |
result = gen_pipeline(prompt, max_length=250, do_sample=False)[0]['generated_text']
|
48 |
+
|
49 |
return result
|
50 |
|
51 |
+
# Gradio UI
|
52 |
def chatbot_interface(query):
|
53 |
return rag_query(query)
|
54 |
|
55 |
+
# Styling for the interface
|
56 |
css = """
|
57 |
.gradio-container {
|
58 |
background-color: #f0f4f8;
|
|
|
98 |
}
|
99 |
"""
|
100 |
|
101 |
+
# Create the Gradio interface
|
102 |
iface = gr.Interface(
|
103 |
fn=chatbot_interface,
|
104 |
inputs="text",
|
|
|
109 |
css=css
|
110 |
)
|
111 |
|
112 |
+
# Launch the Gradio interface
|
113 |
+
iface.launch()
|