Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -3,26 +3,24 @@ import re
|
|
| 3 |
import torch
|
| 4 |
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
|
| 5 |
|
| 6 |
-
# Load
|
| 7 |
-
model_name = "
|
| 8 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 9 |
-
model = AutoModelForSeq2SeqLM.from_pretrained(model_name, torch_dtype=torch.float16) # Use
|
| 10 |
|
| 11 |
-
# Move model to CPU (
|
| 12 |
-
|
| 13 |
-
model.to(device)
|
| 14 |
|
| 15 |
-
# Initialize paraphrase pipeline
|
| 16 |
paraphrase_pipeline = pipeline(
|
| 17 |
"text2text-generation",
|
| 18 |
model=model,
|
| 19 |
tokenizer=tokenizer,
|
| 20 |
-
truncation=True
|
| 21 |
-
device=0 if device == "cuda" else -1 # Use GPU if available
|
| 22 |
)
|
| 23 |
|
| 24 |
def split_sentences(text):
|
| 25 |
-
"""Split text into sentences using regex."""
|
| 26 |
return re.split(r'(?<=[.!?])\s+', text.strip())
|
| 27 |
|
| 28 |
def paraphrase_text(text):
|
|
@@ -32,10 +30,10 @@ def paraphrase_text(text):
|
|
| 32 |
|
| 33 |
sentences = split_sentences(text)
|
| 34 |
|
| 35 |
-
# Apply
|
| 36 |
paraphrased_results = paraphrase_pipeline(
|
| 37 |
-
[f"
|
| 38 |
-
max_length=50, do_sample=True, batch_size=8, num_return_sequences=1
|
| 39 |
)
|
| 40 |
|
| 41 |
paraphrased_sentences = [result['generated_text'] for result in paraphrased_results]
|
|
@@ -46,8 +44,8 @@ demo = gr.Interface(
|
|
| 46 |
fn=paraphrase_text,
|
| 47 |
inputs=gr.Textbox(label="Enter text", placeholder="Type your text to paraphrase...", lines=10),
|
| 48 |
outputs=gr.Textbox(label="Paraphrased Text", lines=10),
|
| 49 |
-
title="🚀
|
| 50 |
-
description="Enter text and let AI generate a paraphrased version using
|
| 51 |
theme="huggingface"
|
| 52 |
)
|
| 53 |
|
|
|
|
| 3 |
import torch
|
| 4 |
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
|
| 5 |
|
| 6 |
+
# Load T5 paraphrase model (faster than PEGASUS)
|
| 7 |
+
model_name = "Vamsi/T5_Paraphrase_Paws"
|
| 8 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 9 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(model_name, torch_dtype=torch.float16) # Use fp16 for speed
|
| 10 |
|
| 11 |
+
# Move model to CPU (remove if using GPU)
|
| 12 |
+
model.to("cpu")
|
|
|
|
| 13 |
|
| 14 |
+
# Initialize paraphrase pipeline with optimized settings
|
| 15 |
paraphrase_pipeline = pipeline(
|
| 16 |
"text2text-generation",
|
| 17 |
model=model,
|
| 18 |
tokenizer=tokenizer,
|
| 19 |
+
truncation=True
|
|
|
|
| 20 |
)
|
| 21 |
|
| 22 |
def split_sentences(text):
|
| 23 |
+
"""Split text into sentences using regex (faster than nltk)."""
|
| 24 |
return re.split(r'(?<=[.!?])\s+', text.strip())
|
| 25 |
|
| 26 |
def paraphrase_text(text):
|
|
|
|
| 30 |
|
| 31 |
sentences = split_sentences(text)
|
| 32 |
|
| 33 |
+
# Apply T5 paraphrasing to each sentence
|
| 34 |
paraphrased_results = paraphrase_pipeline(
|
| 35 |
+
[f"paraphrase: {sentence} </s>" for sentence in sentences if sentence],
|
| 36 |
+
max_length=50, do_sample=True, batch_size=8, num_return_sequences=1 # Faster settings
|
| 37 |
)
|
| 38 |
|
| 39 |
paraphrased_sentences = [result['generated_text'] for result in paraphrased_results]
|
|
|
|
| 44 |
fn=paraphrase_text,
|
| 45 |
inputs=gr.Textbox(label="Enter text", placeholder="Type your text to paraphrase...", lines=10),
|
| 46 |
outputs=gr.Textbox(label="Paraphrased Text", lines=10),
|
| 47 |
+
title="🚀 Fast & Clean T5 Paraphraser",
|
| 48 |
+
description="Enter text and let AI generate a paraphrased version using an optimized T5 model!",
|
| 49 |
theme="huggingface"
|
| 50 |
)
|
| 51 |
|