Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,108 +1,107 @@
|
|
1 |
-
import gradio as gr
|
2 |
-
import torch
|
3 |
-
import torch.nn.functional as F
|
4 |
-
from transformers import AutoTokenizer
|
5 |
-
import pickle
|
6 |
-
|
7 |
-
from models.rnn import RNNClassifier
|
8 |
-
from models.lstm import LSTMClassifier
|
9 |
-
from models.transformer import TransformerClassifier
|
10 |
-
from utility import simple_tokenizer
|
11 |
-
|
12 |
-
# =========================
|
13 |
-
# Load models and vocab
|
14 |
-
# =========================
|
15 |
-
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
16 |
-
model_name = "prajjwal1/bert-tiny"
|
17 |
-
|
18 |
-
def load_vocab():
|
19 |
-
with open("pretrained_models/vocab.pkl", "rb") as f:
|
20 |
-
return pickle.load(f)
|
21 |
-
|
22 |
-
def load_models(vocab_size, output_dim=6, padding_idx=0):
|
23 |
-
rnn_model = RNNClassifier(vocab_size, 128, 128, output_dim, padding_idx)
|
24 |
-
rnn_model.load_state_dict(torch.load("pretrained_models/best_rnn.pt"))
|
25 |
-
rnn_model = rnn_model.to(device)
|
26 |
-
rnn_model.eval()
|
27 |
-
|
28 |
-
lstm_model = LSTMClassifier(vocab_size, 128, 128, output_dim, padding_idx)
|
29 |
-
lstm_model.load_state_dict(torch.load("pretrained_models/best_lstm.pt"))
|
30 |
-
lstm_model = lstm_model.to(device)
|
31 |
-
lstm_model.eval()
|
32 |
-
|
33 |
-
transformer_model = TransformerClassifier(model_name, output_dim)
|
34 |
-
transformer_model.load_state_dict(torch.load("pretrained_models/best_transformer.pt", map_location=device))
|
35 |
-
transformer_model = transformer_model.to(device)
|
36 |
-
transformer_model.eval()
|
37 |
-
|
38 |
-
return rnn_model, lstm_model, transformer_model
|
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 |
-
|
69 |
-
|
70 |
-
)
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
probs
|
76 |
-
|
77 |
-
|
78 |
-
#
|
79 |
-
#
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
demo.launch()
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
import torch.nn.functional as F
|
4 |
+
from transformers import AutoTokenizer
|
5 |
+
import pickle
|
6 |
+
|
7 |
+
from models.rnn import RNNClassifier
|
8 |
+
from models.lstm import LSTMClassifier
|
9 |
+
from models.transformer import TransformerClassifier
|
10 |
+
from utility import simple_tokenizer
|
11 |
+
|
12 |
+
# =========================
|
13 |
+
# Load models and vocab
|
14 |
+
# =========================
|
15 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
16 |
+
model_name = "prajjwal1/bert-tiny"
|
17 |
+
|
18 |
+
def load_vocab():
|
19 |
+
with open("pretrained_models/vocab.pkl", "rb") as f:
|
20 |
+
return pickle.load(f)
|
21 |
+
|
22 |
+
def load_models(vocab_size, output_dim=6, padding_idx=0):
|
23 |
+
rnn_model = RNNClassifier(vocab_size, 128, 128, output_dim, padding_idx)
|
24 |
+
rnn_model.load_state_dict(torch.load("pretrained_models/best_rnn.pt", map_location=device))
|
25 |
+
rnn_model = rnn_model.to(device)
|
26 |
+
rnn_model.eval()
|
27 |
+
|
28 |
+
lstm_model = LSTMClassifier(vocab_size, 128, 128, output_dim, padding_idx)
|
29 |
+
lstm_model.load_state_dict(torch.load("pretrained_models/best_lstm.pt", map_location=device))
|
30 |
+
lstm_model = lstm_model.to(device)
|
31 |
+
lstm_model.eval()
|
32 |
+
|
33 |
+
transformer_model = TransformerClassifier(model_name, output_dim)
|
34 |
+
transformer_model.load_state_dict(torch.load("pretrained_models/best_transformer.pt", map_location=device))
|
35 |
+
transformer_model = transformer_model.to(device)
|
36 |
+
transformer_model.eval()
|
37 |
+
|
38 |
+
return rnn_model, lstm_model, transformer_model
|
39 |
+
|
40 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
41 |
+
vocab = load_vocab()
|
42 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
43 |
+
rnn_model, lstm_model, transformer_model = load_models(len(vocab))
|
44 |
+
|
45 |
+
emotions = ["anger", "fear", "joy", "love", "sadness", "surprise"]
|
46 |
+
|
47 |
+
def predict(model, text, model_type, vocab, tokenizer=None, max_length=32):
|
48 |
+
if model_type in ["rnn", "lstm"]:
|
49 |
+
# Match collate_fn_rnn but with no random truncation
|
50 |
+
tokens = simple_tokenizer(text)
|
51 |
+
ids = [vocab.get(token, vocab["<UNK>"]) for token in tokens]
|
52 |
+
|
53 |
+
if len(ids) < max_length:
|
54 |
+
ids += [vocab["<PAD>"]] * (max_length - len(ids))
|
55 |
+
else:
|
56 |
+
ids = ids[:max_length]
|
57 |
+
|
58 |
+
input_ids = torch.tensor([ids], dtype=torch.long).to(device)
|
59 |
+
outputs = model(input_ids)
|
60 |
+
|
61 |
+
else:
|
62 |
+
# Match collate_fn_transformer but with no partial_prob
|
63 |
+
encoding = tokenizer(
|
64 |
+
text,
|
65 |
+
padding="max_length",
|
66 |
+
truncation=True,
|
67 |
+
max_length=128,
|
68 |
+
return_tensors="pt"
|
69 |
+
)
|
70 |
+
input_ids = encoding["input_ids"].to(device)
|
71 |
+
attention_mask = encoding["attention_mask"].to(device)
|
72 |
+
outputs = model(input_ids=input_ids, attention_mask=attention_mask)
|
73 |
+
|
74 |
+
probs = F.softmax(outputs, dim=-1)
|
75 |
+
return probs.squeeze().detach().cpu().numpy()
|
76 |
+
|
77 |
+
# =========================
|
78 |
+
# Gradio App
|
79 |
+
# =========================
|
80 |
+
|
81 |
+
def emotion_typeahead(text):
|
82 |
+
if len(text.strip()) <= 2:
|
83 |
+
return {}, {}, {}
|
84 |
+
|
85 |
+
rnn_probs = predict(rnn_model, text.strip(), "rnn", vocab)
|
86 |
+
lstm_probs = predict(lstm_model, text.strip(), "lstm", vocab)
|
87 |
+
transformer_probs = predict(transformer_model, text.strip(), "transformer", vocab, tokenizer)
|
88 |
+
|
89 |
+
rnn_dict = {emo: float(prob) for emo, prob in zip(emotions, rnn_probs)}
|
90 |
+
lstm_dict = {emo: float(prob) for emo, prob in zip(emotions, lstm_probs)}
|
91 |
+
transformer_dict = {emo: float(prob) for emo, prob in zip(emotions, transformer_probs)}
|
92 |
+
|
93 |
+
return rnn_dict, lstm_dict, transformer_dict
|
94 |
+
|
95 |
+
with gr.Blocks() as demo:
|
96 |
+
gr.Markdown("## 🎯 Emotion Typeahead Predictor (RNN, LSTM, Transformer)")
|
97 |
+
|
98 |
+
text_input = gr.Textbox(label="Type your sentence here...")
|
99 |
+
|
100 |
+
with gr.Row():
|
101 |
+
rnn_output = gr.Label(label="🧠 RNN Prediction")
|
102 |
+
lstm_output = gr.Label(label="🧠 LSTM Prediction")
|
103 |
+
transformer_output = gr.Label(label="🧠 Transformer Prediction")
|
104 |
+
|
105 |
+
text_input.change(emotion_typeahead, inputs=text_input, outputs=[rnn_output, lstm_output, transformer_output])
|
106 |
+
|
107 |
+
demo.launch()
|
|