Spaces:
Running
Running
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,93 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# app.py
|
2 |
+
|
3 |
+
import streamlit as st
|
4 |
+
import torch
|
5 |
+
from src.model import TransformerModel
|
6 |
+
from src.utils import load_vocab, tokenize
|
7 |
+
import time
|
8 |
+
import random
|
9 |
+
import os
|
10 |
+
|
11 |
+
# Configuration
|
12 |
+
MODEL_PATH = 'models/3ed0k4_model_epoch10.pth' # Update this path based on the latest model
|
13 |
+
VOCAB_PATH = 'vocab.json'
|
14 |
+
EMBED_SIZE = 256
|
15 |
+
NUM_HEADS = 8
|
16 |
+
HIDDEN_DIM = 512
|
17 |
+
NUM_LAYERS = 4
|
18 |
+
DROPOUT = 0.1
|
19 |
+
MAX_LENGTH = 100 # Maximum tokens to generate
|
20 |
+
|
21 |
+
# Title and Description
|
22 |
+
st.title("3ed0k4 NLP Text Generation Model 🚀")
|
23 |
+
st.write("Enter a prompt, and the model will generate text based on your input. It will take 1 to 10 seconds to respond to simulate 'thinking'.")
|
24 |
+
|
25 |
+
# Load vocabulary
|
26 |
+
@st.cache_resource
|
27 |
+
def load_resources():
|
28 |
+
vocab = load_vocab(VOCAB_PATH)
|
29 |
+
return vocab
|
30 |
+
|
31 |
+
vocab = load_resources()
|
32 |
+
vocab_size = len(vocab)
|
33 |
+
|
34 |
+
# Initialize model
|
35 |
+
@st.cache_resource
|
36 |
+
def load_model():
|
37 |
+
model = TransformerModel(
|
38 |
+
vocab_size=vocab_size,
|
39 |
+
embed_size=EMBED_SIZE,
|
40 |
+
num_heads=NUM_HEADS,
|
41 |
+
hidden_dim=HIDDEN_DIM,
|
42 |
+
num_layers=NUM_LAYERS,
|
43 |
+
dropout=DROPOUT
|
44 |
+
)
|
45 |
+
if not os.path.exists(MODEL_PATH):
|
46 |
+
st.error(f"Model file not found at {MODEL_PATH}. Please ensure the model is trained and the path is correct.")
|
47 |
+
return None
|
48 |
+
model.load_state_dict(torch.load(MODEL_PATH, map_location=torch.device('cpu')))
|
49 |
+
model.eval()
|
50 |
+
return model
|
51 |
+
|
52 |
+
model = load_model()
|
53 |
+
|
54 |
+
def generate_text(prompt, max_length=MAX_LENGTH):
|
55 |
+
tokens = tokenize(prompt)
|
56 |
+
numericalized = [vocab.get(token, vocab['<UNK>']) for token in tokens]
|
57 |
+
input_seq = torch.tensor(numericalized, dtype=torch.long).unsqueeze(0) # Batch size 1
|
58 |
+
|
59 |
+
generated = numericalized.copy()
|
60 |
+
|
61 |
+
with torch.no_grad():
|
62 |
+
for _ in range(max_length):
|
63 |
+
src_mask = model.generate_square_subsequent_mask(input_seq.size(1)).to(input_seq.device)
|
64 |
+
outputs = model(input_seq, src_mask)
|
65 |
+
next_token_logits = outputs[0, -1, :]
|
66 |
+
next_token = torch.argmax(next_token_logits).item()
|
67 |
+
|
68 |
+
if next_token == vocab.get('<PAD>', 0):
|
69 |
+
break
|
70 |
+
|
71 |
+
generated.append(next_token)
|
72 |
+
input_seq = torch.tensor(generated, dtype=torch.long).unsqueeze(0)
|
73 |
+
|
74 |
+
# Convert numerical tokens back to words
|
75 |
+
inv_vocab = {idx: word for word, idx in vocab.items()}
|
76 |
+
generated_tokens = [inv_vocab.get(tok, '<UNK>') for tok in generated]
|
77 |
+
return ' '.join(generated_tokens)
|
78 |
+
|
79 |
+
# User Inputs
|
80 |
+
prompt = st.text_input("Enter your prompt:", "")
|
81 |
+
delay = st.slider("Select thinking delay (seconds):", min_value=1, max_value=10, value=3)
|
82 |
+
|
83 |
+
if st.button("Generate"):
|
84 |
+
if not model:
|
85 |
+
st.error("Model is not loaded. Please check the model path.")
|
86 |
+
elif prompt.strip() == "":
|
87 |
+
st.warning("Please enter a prompt to generate text.")
|
88 |
+
else:
|
89 |
+
with st.spinner("Thinking..."):
|
90 |
+
time.sleep(delay)
|
91 |
+
response = generate_text(prompt)
|
92 |
+
st.success("Here's the generated text:")
|
93 |
+
st.write(response)
|