Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,45 +1,69 @@
|
|
1 |
# app.py
|
2 |
-
|
3 |
import gradio as gr
|
4 |
-
from
|
5 |
-
import
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
MODEL_NAME = "bert-base-uncased"
|
11 |
-
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
|
12 |
-
model = AutoModel.from_pretrained(MODEL_NAME, output_attentions=True)
|
13 |
-
|
14 |
-
def visualize_attention(text):
|
15 |
-
inputs = tokenizer(text, return_tensors="pt")
|
16 |
-
outputs = model(**inputs)
|
17 |
-
|
18 |
-
# Grab attentions from output
|
19 |
-
attentions = outputs.attentions # List of (num_layers, batch, num_heads, seq_len, seq_len)
|
20 |
-
tokens = tokenizer.convert_ids_to_tokens(inputs['input_ids'][0])
|
21 |
-
|
22 |
-
fig, ax = plt.subplots(figsize=(8, 6))
|
23 |
-
# Just visualize attention from last layer, first head
|
24 |
-
attn_matrix = attentions[-1][0][0].detach().numpy()
|
25 |
-
|
26 |
-
cax = ax.matshow(attn_matrix, cmap='viridis')
|
27 |
-
fig.colorbar(cax)
|
28 |
-
|
29 |
-
ax.set_xticks(range(len(tokens)))
|
30 |
-
ax.set_yticks(range(len(tokens)))
|
31 |
-
ax.set_xticklabels(tokens, rotation=90)
|
32 |
-
ax.set_yticklabels(tokens)
|
33 |
-
ax.set_title("Attention Map - Last Layer, Head 1")
|
34 |
-
|
35 |
-
return fig
|
36 |
-
|
37 |
-
iface = gr.Interface(
|
38 |
-
fn=visualize_attention,
|
39 |
-
inputs=gr.Textbox(lines=2, placeholder="Enter your text here..."),
|
40 |
-
outputs=gr.Plot(),
|
41 |
-
title="🧠 Transformer Attention Visualizer",
|
42 |
-
description="Visualizes the self-attention of the BERT model's last layer."
|
43 |
)
|
44 |
|
45 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
# app.py
|
|
|
2 |
import gradio as gr
|
3 |
+
from model_utils import load_model_info, get_model_stats
|
4 |
+
from visualize import (
|
5 |
+
visualize_attention,
|
6 |
+
visualize_token_embeddings,
|
7 |
+
plot_tokenization,
|
8 |
+
compare_model_sizes
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
)
|
10 |
|
11 |
+
MODEL_CHOICES = {
|
12 |
+
"BERT (base)": "bert-base-uncased",
|
13 |
+
"DistilBERT": "distilbert-base-uncased",
|
14 |
+
"RoBERTa": "roberta-base",
|
15 |
+
"GPT-2": "gpt2",
|
16 |
+
"Electra": "google/electra-base-discriminator",
|
17 |
+
"ALBERT": "albert-base-v2",
|
18 |
+
"XLNet": "xlnet-base-cased"
|
19 |
+
}
|
20 |
+
|
21 |
+
def run_visualizer(model_name, text, layer, head):
|
22 |
+
model_info = load_model_info(model_name)
|
23 |
+
attention_plot = visualize_attention(model_info, text, layer, head)
|
24 |
+
token_heatmap = visualize_token_embeddings(model_info, text)
|
25 |
+
token_plot = plot_tokenization(model_info, text)
|
26 |
+
model_stats = get_model_stats(model_info)
|
27 |
+
|
28 |
+
return attention_plot, token_heatmap, token_plot, model_stats
|
29 |
+
|
30 |
+
def run_comparison_chart():
|
31 |
+
return compare_model_sizes(MODEL_CHOICES.values())
|
32 |
+
|
33 |
+
with gr.Blocks() as demo:
|
34 |
+
gr.Markdown("""
|
35 |
+
# 🤖 Transformer Model Visualizer
|
36 |
+
Explore attention heads, token embeddings, and tokenizer behavior across popular transformer models.
|
37 |
+
""")
|
38 |
+
|
39 |
+
with gr.Row():
|
40 |
+
model_selector = gr.Dropdown(label="Choose Model", choices=list(MODEL_CHOICES.keys()), value="BERT (base)")
|
41 |
+
input_text = gr.Textbox(label="Input Text", placeholder="Enter text to analyze")
|
42 |
+
|
43 |
+
with gr.Row():
|
44 |
+
layer_slider = gr.Slider(minimum=0, maximum=11, step=1, value=0, label="Layer")
|
45 |
+
head_slider = gr.Slider(minimum=0, maximum=11, step=1, value=0, label="Attention Head")
|
46 |
+
|
47 |
+
run_btn = gr.Button("Run Analysis")
|
48 |
+
|
49 |
+
with gr.Row():
|
50 |
+
attention_output = gr.Plot(label="Self-Attention Visualization")
|
51 |
+
embedding_output = gr.Plot(label="Token Embedding Heatmap")
|
52 |
+
|
53 |
+
with gr.Row():
|
54 |
+
token_output = gr.Plot(label="Tokenization Overview")
|
55 |
+
model_output = gr.JSON(label="Model Details")
|
56 |
+
|
57 |
+
run_btn.click(
|
58 |
+
fn=run_visualizer,
|
59 |
+
inputs=[model_selector, input_text, layer_slider, head_slider],
|
60 |
+
outputs=[attention_output, embedding_output, token_output, model_output]
|
61 |
+
)
|
62 |
+
|
63 |
+
with gr.Accordion("📊 Compare Model Sizes", open=False):
|
64 |
+
compare_btn = gr.Button("Generate Comparison Chart")
|
65 |
+
comparison_output = gr.Plot()
|
66 |
+
compare_btn.click(fn=run_comparison_chart, outputs=comparison_output)
|
67 |
+
|
68 |
+
demo.launch()
|
69 |
+
|