app.py

def predict(message, history, system_prompt, temperature, max_new_tokens, top_k, repetition_penalty, top_p):
    if CHAT_TEMPLATE == "Auto":
        stop_tokens = [tokenizer.eos_token_id]
        instruction = system_prompt + "\n\n"
        for user, assistant in history:
            instruction += f"User: {user}\nAssistant: {assistant}\n"
        instruction += f"User: {message}\nAssistant:"
    elif CHAT_TEMPLATE == "ChatML":
        stop_tokens = ["<|endoftext|>", "<|im_end|>"]
        instruction = '<|im_start|>system\n' + system_prompt + '\n<|im_end|>\n'
        for user, assistant in history:
            instruction += f'<|im_start|>user\n{user}\n<|im_end|>\n<|im_start|>assistant\n{assistant}\n<|im_end|>\n'
        instruction += f'<|im_start|>user\n{message}\n<|im_end|>\n<|im_start|>assistant\n'
    elif CHAT_TEMPLATE == "Mistral Instruct":
        stop_tokens = ["</s>", "[INST]", "[INST] ", "<s>", "[/INST]", "[/INST] "]
        instruction = f'<s>[INST] {system_prompt}\n'
        for user, assistant in history:
            instruction += f'{user} [/INST] {assistant}</s>[INST]'
        instruction += f' {message} [/INST]'
    else:
        raise Exception("Incorrect chat template, select 'Auto', 'ChatML' or 'Mistral Instruct'")
    print(instruction)
    
    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
    enc = tokenizer(instruction, return_tensors="pt", padding=True, truncation=True)
    input_ids, attention_mask = enc.input_ids, enc.attention_mask

    if input_ids.shape[1] > CONTEXT_LENGTH:
        input_ids = input_ids[:, -CONTEXT_LENGTH:]
        attention_mask = attention_mask[:, -CONTEXT_LENGTH:]

    generate_kwargs = dict(
        input_ids=input_ids.to(device),
        attention_mask=attention_mask.to(device),
        streamer=streamer,
        do_sample=True,
        temperature=temperature,
        max_new_tokens=max_new_tokens,
        top_k=top_k,
        repetition_penalty=repetition_penalty,
        top_p=top_p
    )
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()
    outputs = []
    for new_token in streamer:
        outputs.append(new_token)
        if new_token in stop_tokens:
            break
        result = "".join(outputs)
        # Wrapping result in Markdown for LaTeX rendering
        yield f"$$ {result} $$"


# Create Gradio interface
gr.ChatInterface(
    predict,
    title=EMOJI + " " + MODEL_NAME,
    description=DESCRIPTION,
    additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False),
    additional_inputs=[
        gr.Textbox("You are a code assistant.", label="System prompt"),
        gr.Slider(0, 1, 0.3, label="Temperature"),
        gr.Slider(128, 4096, 1024, label="Max new tokens"),
        gr.Slider(1, 80, 40, label="Top K sampling"),
        gr.Slider(0, 2, 1.1, label="Repetition penalty"),
        gr.Slider(0, 1, 0.95, label="Top P sampling"),
    ],
    theme=gr.themes.Soft(primary_hue=COLOR),
).queue().launch()