app.py

import gradio as gr
from threading import Thread
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, TextIteratorStreamer

model_id = "rasyosef/gpt2-small-amharic-128-v3"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

gpt2_am = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    pad_token_id=tokenizer.pad_token_id,
    eos_token_id=tokenizer.eos_token_id
  )

def generate(prompt):
  prompt_length = len(tokenizer.tokenize(prompt))
  if prompt_length >= 128:
    yield prompt + "\n\nPrompt is too long. It needs to be less than 128 tokens."
  else:
    max_new_tokens = max(0, 128 - prompt_length)
    streamer = TextIteratorStreamer(tokenizer=tokenizer, skip_prompt=False, skip_special_tokens=True, timeout=300.0)
    thread = Thread(
      target=gpt2_am,
      kwargs={
        "text_inputs": prompt,
        "max_new_tokens": max_new_tokens,
        "temperature": 0.8,
        "do_sample": True,
        "top_k": 8,
        "top_p": 0.8,
        "repetition_penalty": 1.25,
        "streamer": streamer
      })
    thread.start()

    generated_text = ""
    for word in streamer:
      generated_text += word
      response = generated_text.strip()
      yield response

with gr.Blocks() as demo:
  gr.Markdown("""
  # GPT2 Amharic
  This is a demo for a smaller version of the gpt2 decoder transformer model pretrained for 1.5 days on `290 million` tokens of **Amharic** text. The context size of `gpt2-small-amharic` is 128 tokens.
  """)

  prompt = gr.Textbox(label="Prompt", placeholder="Enter prompt here", lines=4, interactive=True)
  with gr.Row():
    with gr.Column():
      gen = gr.Button("Generate")
    with gr.Column():
      btn = gr.ClearButton([prompt])
  gen.click(generate, inputs=[prompt], outputs=[prompt])
  examples = gr.Examples(
        examples=[
            "የ አዲስ አበባ",
            "በ ኢንግሊዝ ፕሪምየር ሊግ",
            "ፕሬዚዳንት ዶናልድ ትራምፕ"
          ],
        inputs=[prompt],
    )
demo.queue().launch(debug=True)