File size: 4,588 Bytes
2196aef |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 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 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 |
# app_gradio.py
import gradio as gr
from deep_translator import GoogleTranslator
from langdetect import detect
from transformers import GPT2LMHeadModel, GPT2Tokenizer
import torch
import re # Import regex for post-processing
MODEL_DIR = "./fine_tuned_model"
def load_model():
tokenizer = GPT2Tokenizer.from_pretrained(MODEL_DIR, local_files_only=True)
model = GPT2LMHeadModel.from_pretrained(MODEL_DIR, local_files_only=True)
if tokenizer.pad_token is None:
tokenizer.pad_token = tokenizer.eos_token
model.config.pad_token_id = tokenizer.pad_token_id
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device).eval()
return tokenizer, model, device
tokenizer, model, device = load_model()
def to_en(text):
try:
lang = detect(text)
except Exception:
lang = "en"
if lang == "en":
return text, "en"
translated_text = GoogleTranslator(source=lang, target="en").translate(text)
# Handle potential None return from translator
return translated_text if translated_text is not None else text, lang
def from_en(text, tgt):
if tgt == "en":
return text
translated_text = GoogleTranslator(source="en", target=tgt).translate(text)
# Handle potential None return from translator
return translated_text if translated_text is not None else text
def generate(prompt, max_new_tokens=120, temperature=0.8):
inputs = tokenizer(prompt, return_tensors="pt").to(device)
with torch.no_grad():
out = model.generate(
**inputs,
max_new_tokens=max_new_tokens,
do_sample=True,
top_k=50,
top_p=0.95,
temperature=temperature,
eos_token_id=tokenizer.eos_token_id,
pad_token_id=tokenizer.pad_token_id,
)
return tokenizer.decode(out[0], skip_special_tokens=True)
def post_process_generated_text(text, prompt):
# Simple post-processing to clean up potential repetitions or unwanted tokens
cleaned_text = text.replace(prompt, "").strip() # Remove the prompt from the output
# Remove consecutive repeated words - improved
words = cleaned_text.split()
if not words:
return ""
cleaned_words = [words[0]]
for i in range(1, len(words)):
if words[i].lower() != words[i-1].lower(): # Case-insensitive comparison
cleaned_words.append(words[i])
return " ".join(cleaned_words)
def recommend_course(t):
t = t.lower()
if "python" in t: return "🐍 Python Programming – Beginner to Advanced"
if "data science" in t: return "📊 Data Science Master Program"
if "ai" in t or "machine learning" in t or "ml" in t: return "🤖 AI & Machine Learning with Real Projects"
if "web" in t or "full stack" in t or "javascript" in t or "react" in t: return "🌐 Full Stack Web Development"
if "java" in t: return "☕ Java Programming Essentials"
return None
def chat(user_input, history):
en, lang = to_en(user_input)
course = recommend_course(en)
if course:
en_resp = f"I recommend you check out: {course}"
else:
# Modify prompt to encourage structured output based on keywords
prompt = f"User: {en}\nAssistant:"
if any(keyword in en.lower() for keyword in ["what is", "tell me about"]):
prompt = f"User: {en}\nAssistant: Here is information about {en.lower().replace('what is', '').replace('tell me about', '').strip()}:\n"
elif "recommend" in en.lower():
prompt = f"User: {en}\nAssistant: Based on your request, here is a recommendation:\n"
en_resp = generate(prompt)
# Apply post-processing to clean the generated text
en_resp = post_process_generated_text(en_resp, prompt)
if en_resp.startswith(prompt):
en_resp = en_resp[len(prompt):].strip()
final = from_en(en_resp, lang)
history = history + [(user_input, final)]
return history, history
with gr.Blocks(theme=gr.themes.Soft()) as demo:
gr.Markdown("# 🌐 Multilingual GPT-2 Chatbot")
chatbot = gr.Chatbot(height=420)
msg = gr.Textbox(label="Your Message", placeholder="Type here...")
clear = gr.Button("🗑️ Clear")
state = gr.State([])
msg.submit(chat, [msg, state], [chatbot, state])
clear.click(lambda: ([], []), None, [chatbot, state], queue=False)
# You can run this in a separate cell using !python app_gradio.py if needed,
# but running it directly in the notebook cell is also possible.
# if __name__ == "__main__":
# demo.launch()
|