|
from huggingface_hub import InferenceClient |
|
import gradio as gr |
|
import random |
|
import pandas as pd |
|
from io import BytesIO |
|
import csv |
|
import os |
|
import io |
|
import tempfile |
|
import re |
|
import streamlit as st |
|
import torch |
|
from transformers import M2M100Tokenizer, M2M100ForConditionalGeneration |
|
import time |
|
import logging |
|
|
|
if torch.cuda.is_available(): |
|
device = torch.device("cuda:0") |
|
else: |
|
device = torch.device("cpu") |
|
logging.warning("GPU not found, using CPU, translation will be very slow.") |
|
|
|
client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1") |
|
|
|
lang_id = { |
|
"Afrikaans": "af", |
|
"Amharic": "am", |
|
"Arabic": "ar", |
|
"Asturian": "ast", |
|
"Azerbaijani": "az", |
|
"Bashkir": "ba", |
|
"Belarusian": "be", |
|
"Bulgarian": "bg", |
|
"Bengali": "bn", |
|
"Breton": "br", |
|
"Bosnian": "bs", |
|
"Catalan": "ca", |
|
"Cebuano": "ceb", |
|
"Czech": "cs", |
|
"Welsh": "cy", |
|
"Danish": "da", |
|
"German": "de", |
|
"Greeek": "el", |
|
"English": "en", |
|
"Spanish": "es", |
|
"Estonian": "et", |
|
"Persian": "fa", |
|
"Fulah": "ff", |
|
"Finnish": "fi", |
|
"French": "fr", |
|
"Western Frisian": "fy", |
|
"Irish": "ga", |
|
"Gaelic": "gd", |
|
"Galician": "gl", |
|
"Gujarati": "gu", |
|
"Hausa": "ha", |
|
"Hebrew": "he", |
|
"Hindi": "hi", |
|
"Croatian": "hr", |
|
"Haitian": "ht", |
|
"Hungarian": "hu", |
|
"Armenian": "hy", |
|
"Indonesian": "id" |
|
} |
|
|
|
@st.cache(suppress_st_warning=True, allow_output_mutation=True) |
|
def load_model(pretrained_model: str = "facebook/m2m100_1.2B", cache_dir: str = "models/"): |
|
tokenizer = M2M100Tokenizer.from_pretrained(pretrained_model, cache_dir=cache_dir) |
|
model = M2M100ForConditionalGeneration.from_pretrained(pretrained_model, cache_dir=cache_dir).to(device) |
|
model.eval() |
|
return tokenizer, model |
|
|
|
def extract_text_from_excel(file): |
|
df = pd.read_excel(file) |
|
text = ' '.join(df['Unnamed: 1'].astype(str)) |
|
return text |
|
|
|
def save_to_csv(sentence, output, filename="synthetic_data.csv"): |
|
with open(filename, mode='a', newline='', encoding='utf-8') as file: |
|
writer = csv.writer(file) |
|
writer.writerow([sentence, output]) |
|
|
|
def generate(file, temperature, max_new_tokens, top_p, repetition_penalty, num_similar_sentences): |
|
text = extract_text_from_excel(file) |
|
sentences = text.split('.') |
|
random.shuffle(sentences) |
|
|
|
with tempfile.NamedTemporaryFile(mode='w', newline='', delete=False, suffix='.csv') as tmp: |
|
fieldnames = ['Original Sentence', 'Generated Sentence'] |
|
writer = csv.DictWriter(tmp, fieldnames=fieldnames) |
|
writer.writeheader() |
|
|
|
for sentence in sentences: |
|
sentence = sentence.strip() |
|
if not sentence: |
|
continue |
|
|
|
generate_kwargs = { |
|
"temperature": temperature, |
|
"max_new_tokens": max_new_tokens, |
|
"top_p": top_p, |
|
"repetition_penalty": repetition_penalty, |
|
"do_sample": True, |
|
"seed": 42, |
|
} |
|
|
|
try: |
|
stream = client.text_generation(sentence, **generate_kwargs, stream=True, details=True, return_full_text=False) |
|
output = "" |
|
for response in stream: |
|
output += response.token.text |
|
|
|
generated_sentences = re.split(r'(?<=[\.\!\?:])[\s\n]+', output) |
|
generated_sentences = [s.strip() for s in generated_sentences if s.strip() and s != '.'] |
|
|
|
for _ in range(num_similar_sentences): |
|
if not generated_sentences: |
|
break |
|
generated_sentence = generated_sentences.pop(random.randrange(len(generated_sentences))) |
|
|
|
|
|
tokenizer, model = load_model() |
|
src_lang = lang_id[language] |
|
trg_lang = lang_id["English"] |
|
tokenizer.src_lang = src_lang |
|
with torch.no_grad(): |
|
encoded_input = tokenizer(generated_sentence, return_tensors="pt").to(device) |
|
generated_tokens = model.generate(**encoded_input, forced_bos_token_id=tokenizer.get_lang_id(trg_lang)) |
|
translated_sentence = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0] |
|
|
|
|
|
tokenizer, model = load_model() |
|
src_lang = lang_id["English"] |
|
trg_lang = lang_id["Azerbaijani"] |
|
tokenizer.src_lang = src_lang |
|
with torch.no_grad(): |
|
encoded_input = tokenizer(sentence, return_tensors="pt").to(device) |
|
generated_tokens = model.generate(**encoded_input, forced_bos_token_id=tokenizer.get_lang_id(trg_lang)) |
|
translated_sentence_az = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0] |
|
|
|
writer.writerow({'Original Sentence': translated_sentence_az, 'Generated Sentence': translated_sentence}) |
|
|
|
except Exception as e: |
|
print(f"Error generating data for sentence '{sentence}': {e}") |
|
|
|
tmp_path = tmp.name |
|
|
|
return tmp_path |
|
|
|
gr.Interface( |
|
fn=generate, |
|
inputs=[ |
|
gr.File(label="Upload Excel File", file_count="single", file_types=[".xlsx"]), |
|
gr.Slider(label="Temperature", value=0.9, minimum=0.0, maximum=1.0, step=0.05, interactive=True, info="Higher values produce more diverse outputs"), |
|
gr.Slider(label="Max new tokens", value=256, minimum=0, maximum=5120, step=64, interactive=True, info="The maximum numbers of new tokens"), |
|
gr.Slider(label="Top-p (nucleus sampling)", value=0.95, minimum=0.0, maximum=1, step=0.05, interactive=True, info="Higher values sample more low-probability tokens"), |
|
gr.Slider(label="Repetition penalty", value=1.0, minimum=1.0, maximum=2.0, step=0.1, interactive=True, info="Penalize repeated tokens"), |
|
gr.Slider(label="Number of similar sentences", value=10, minimum=1, maximum=20, step=1, interactive=True, info="Number of similar sentences to generate for each original sentence"), |
|
gr.Dropdown(label="Language of the input data", choices=list(lang_id.keys()), value="English") |
|
], |
|
outputs=gr.File(label="Synthetic Data "), |
|
title="SDG", |
|
description="AYE QABIL.", |
|
allow_flagging="never", |
|
).launch() |
