app.py

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

model_id = "rasyosef/llama-3.2-amharic-64k-1024"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_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:
    inputs = tokenizer(prompt, return_tensors="pt")
    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=model.generate,
      kwargs={
        "inputs": inputs["input_ids"],
        "attention_mask": inputs["attention_mask"],
        "max_new_tokens": max_new_tokens,
        "temperature": 0.3,
        "do_sample": True,
        "top_k": 8,
        "top_p": 0.8,
        "repetition_penalty": 1.25,
        "streamer": streamer,
        "pad_token_id": tokenizer.pad_token_id,
        "eos_token_id": tokenizer.eos_token_id
      })
    thread.start()

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

with gr.Blocks(css="#prompt_textbox textarea {color: blue}") as demo:
  gr.Markdown("""
  # Llama 3.2 Amharic
  This is a demo for [llama-3.2-amharic](https://huggingface.co/rasyosef/llama-3.2-amharic-64k-1024), a smaller version of Meta's [Llama-3.2-1B](https://huggingface.co/meta-llama/Llama-3.2-1B) decoder transformer model pretrained for 3 days on `210 million` tokens of **Amharic** text. This model has `179 million` parameters and a context size of `1024` tokens. This is a base model and hasn't undergone any supervised finetuing yet.
  
  Please **enter a prompt** and click the **Generate** button to generate completions for the prompt.
  #### Text generation parameters:
  - `temperature` : **0.3**
  - `do_sample` : **True**
  - `top_k` : **8**
  - `top_p` : **0.8**
  - `repetition_penalty` : **1.25**
  """)

  prompt = gr.Textbox(label="Prompt", placeholder="Enter prompt here", lines=4, interactive=True, elem_id="prompt_textbox")
  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=[
            "አዲስ አበባ",
            "በእንግሊዙ ፕሬሚየር ሊግ",
            "ፕሬዚዳንት ዶናልድ ትራምፕ",
            "በመስቀል አደባባይ",
          ],
        inputs=[prompt],
    )
demo.queue().launch(debug=True)