app.py

import spaces
import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, StoppingCriteria, StoppingCriteriaList, TextIteratorStreamer
from threading import Thread

# model_path = 'dreamerdeo/Sailor2-0.8B-Chat'
model_path = 'sail/Sailor-0.5B-Chat'

# Loading the tokenizer and model from Hugging Face's model hub.
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True, torch_dtype=torch.bfloat16)

# using CUDA for an optimal experience
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = model.to(device)

# Defining a custom stopping criteria class for the model's text generation.
class StopOnTokens(StoppingCriteria):
    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
        stop_ids = [151645]  # IDs of tokens where the generation should stop.
        for stop_id in stop_ids:
            if input_ids[0][-1] == stop_id:  # Checking if the last generated token is a stop token.
                return True
        return False


system_role= 'system'
user_role = 'user'
assistant_role = 'assistant'

sft_start_token =  "<|im_start|>"
sft_end_token = "<|im_end|>"
ct_end_token = "<|endoftext|>"

system_prompt= \
'You are an AI assistant named Sailor2, created by Sea AI Lab. \
As an AI assistant, you can answer questions in English, Chinese, and Southeast Asian languages \
such as Burmese, Cebuano, Ilocano, Indonesian, Javanese, Khmer, Lao, Malay, Sundanese, Tagalog, Thai, Vietnamese, and Waray. \
Your responses should be friendly, unbiased, informative, detailed, and faithful.'

system_prompt = f"<|im_start|>{system_role}\n{system_prompt}<|im_end|>"

# Function to generate model predictions.

@spaces.GPU()
def predict(message, history):
    # history = []
    history_transformer_format = history + [[message, ""]]
    stop = StopOnTokens()

    # Formatting the input for the model.
    messages =  system_prompt + sft_end_token.join([sft_end_token.join([f"\n{sft_start_token}{user_role}\n" + item[0], f"\n{sft_start_token}{assistant_role}\n" + item[1]])
                        for item in history_transformer_format])
    model_inputs = tokenizer([messages], return_tensors="pt").to(device)
    streamer = TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)
    generate_kwargs = dict(
        model_inputs,
        streamer=streamer,
        max_new_tokens=1024,
        do_sample=True,
        top_p=0.8,
        top_k=20,
        temperature=0.7,
        num_beams=1,
        stopping_criteria=StoppingCriteriaList([stop]),
        repetition_penalty=1.1,
    )
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()  # Starting the generation in a separate thread.
    partial_message = ""
    for new_token in streamer:
        partial_message += new_token
        if sft_end_token in partial_message:  # Breaking the loop if the stop token is generated.
            break
        yield partial_message


css = """
full-height {
    height: 100%;
}
"""

prompt_examples = [
    'How to cook a fish?',
    'Cara memanggang ikan',
    'วิธีย่างปลา',
    'Cách nướng cá'
]

placeholder = """
<div style="opacity: 0.5;">
    <img src="https://raw.githubusercontent.com/sail-sg/sailor-llm/main/misc/banner.jpg" style="width:30%;">
    <br>Sailor models are designed to understand and generate text across diverse linguistic landscapes of these SEA regions:
    <br>🇮🇩Indonesian, 🇹🇭Thai, 🇻🇳Vietnamese, 🇲🇾Malay, and 🇱🇦Lao.
</div>
"""

chatbot = gr.Chatbot(label='Sailor', placeholder=placeholder) 
with gr.Blocks(theme=gr.themes.Soft(), fill_height=True) as demo:
    # gr.Markdown("""<center><font size=8>Sailor-Chat Bot⚓</center>""")
    gr.Markdown("""<p align="center"><img src="https://github.com/sail-sg/sailor-llm/raw/main/misc/wide_sailor_banner.jpg" style="height: 110px"/><p>""")
    gr.ChatInterface(predict, chatbot=chatbot, fill_height=True, examples=prompt_examples, css=css)

    demo.launch()  # Launching the web interface.