File size: 1,230 Bytes
22a2ad6 5e937d1 22a2ad6 |
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 |
import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
# Load base and adapter model
base_model = "microsoft/phi-2"
adapter_model = "Sabbir772/phi2_sylhet"
tokenizer = AutoTokenizer.from_pretrained(base_model)
base = AutoModelForCausalLM.from_pretrained(base_model)
model = PeftModel.from_pretrained(base, adapter_model)
model.eval()
def translate(model, tokenizer, input_text, direction=0, max_new_tokens=256):
if direction == 0:
prompt = f"Translate Bangla to Sylheti: {input_text}\nOutput:"
else:
prompt = f"Translate Sylheti to Bangla: {input_text}\nOutput:"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
output_ids = model.generate(**inputs, max_new_tokens=max_new_tokens)
return tokenizer.decode(output_ids[0], skip_special_tokens=True)
# Gradio function wrapper
def infer(text, direction):
return translate(model, tokenizer, text, direction=direction)
demo = gr.Interface(
fn=infer,
inputs=[gr.Textbox(label="Input Text"), gr.Radio(["Bangla to Sylheti", "Sylheti to Bangla"], type="index", label="Translation Direction")],
outputs="text",
title="Phi-2 Sylheti Translator"
)
demo.launch()
|