app.py

import torch
import transformers
import gradio as gr
import sys

from train import smart_tokenizer_and_embedding_resize, \
	DEFAULT_PAD_TOKEN, DEFAULT_EOS_TOKEN, DEFAULT_BOS_TOKEN, DEFAULT_UNK_TOKEN, \
	PROMPT_DICT, \
    DataArguments
    
BASE_MODEL = sys.argv[1]

if torch.cuda.is_available():
    device = "cuda"
else:
    device = "cpu"

try:
    if torch.backends.mps.is_available():
        device = "mps"
except:
    pass

if device == "cuda":
    model = transformers.AutoModelForCausalLM.from_pretrained(
        BASE_MODEL,
        load_in_8bit=False,
        torch_dtype=torch.bfloat16,
        device_map="auto",
    )
    
elif device == "mps":
    model = transformers.AutoModelForCausalLM.from_pretrained(
        BASE_MODEL,
        device_map={"": device},
        torch_dtype=torch.bfloat16,
    )
else:
    model = transformers.AutoModelForCausalLM.from_pretrained(
        BASE_MODEL, device_map={"": device}, low_cpu_mem_usage=True
    )

tokenizer = transformers.AutoTokenizer.from_pretrained(
    BASE_MODEL,
    use_fast=False,
)

special_tokens_dict = dict()
if tokenizer.pad_token is None:
    special_tokens_dict["pad_token"] = DEFAULT_PAD_TOKEN
if tokenizer.eos_token is None:
    special_tokens_dict["eos_token"] = DEFAULT_EOS_TOKEN
if tokenizer.bos_token is None:
    special_tokens_dict["bos_token"] = DEFAULT_BOS_TOKEN
if tokenizer.unk_token is None:
    special_tokens_dict["unk_token"] = DEFAULT_UNK_TOKEN

if tokenizer.pad_token is None:
    smart_tokenizer_and_embedding_resize(
        special_tokens_dict=special_tokens_dict,
        tokenizer=tokenizer,
        model=model,
    )
tokenizer.padding_side = "left"

def generate_prompt(instruction, input=None, template="alpaca"):
    if template == "alpaca":
        if input:
            return PROMPT_DICT["prompt_input"].format(instruction=instruction, input=input)
        else:
            return PROMPT_DICT["prompt_no_input"].format(instruction=instruction)
    elif template == "raw":
        if input:
            return f"{instruction}\n\n{input}"
        else:
            return f"{instruction}"
    else:
        raise NotImplementedError

if device != "cpu":
    model.half()
model.eval()
if torch.__version__ >= "2":
    model = torch.compile(model)


def evaluate(
    instruction,
    input=None,
    template="alpaca",
    do_sample=False,
    temperature=0.1,
    top_p=0.75,
    top_k=40,
    num_beams=4,
    max_new_tokens=128,
    **kwargs,
):
    prompt = generate_prompt(instruction, input, template)
    inputs = tokenizer(prompt, return_tensors="pt")
    input_ids = inputs["input_ids"].to(device)
    generation_config = transformers.GenerationConfig(
        temperature=temperature,
        top_p=top_p,
        top_k=top_k,
        num_beams=num_beams,
        **kwargs,
    )
    with torch.no_grad():
        generation_output = model.generate(
            input_ids=input_ids,
            generation_config=generation_config,
            return_dict_in_generate=True,
            output_scores=True,
            max_new_tokens=max_new_tokens,
        )
    s = generation_output.sequences[0]
    output = tokenizer.decode(s, skip_special_tokens=True)
    return output[len(prompt):].strip()


g = gr.Interface(
    fn=evaluate,
    inputs=[
        gr.components.Textbox(
            lines=2, label="Instruction", placeholder="Tell me about alpacas."
        ),
        gr.components.Textbox(lines=2, label="Input", placeholder="none"),
        gr.components.Dropdown(["raw", "alpaca"], value="alpaca", label="Template Format"),
        gr.components.Checkbox(label="Do Sample"),
        gr.components.Slider(minimum=0, maximum=1, value=0.7, label="Temperature"),
        gr.components.Slider(minimum=0, maximum=1, value=0.75, label="Top p"),
        gr.components.Slider(minimum=0, maximum=100, step=1, value=40, label="Top k"),
        gr.components.Slider(minimum=1, maximum=4, step=1, value=1, label="Beams"),
        gr.components.Slider(
            minimum=1, maximum=512, step=1, value=512, label="Max tokens"
        ),
    ],
    outputs=[
        gr.inputs.Textbox(
            lines=5,
            label="Output",
        )
    ],
    title="🦙Alpaca Web Demo",
    description="Made with 💖",
    flagging_options=["👍🏼", "👎🏼"]
)
g.queue(concurrency_count=1)
g.launch()

# Old testing code follows.

"""
if __name__ == "__main__":
    # testing code for readme
    for instruction in [
        "Tell me about alpacas.",
        "Tell me about the president of Mexico in 2019.",
        "Tell me about the king of France in 2019.",
        "List all Canadian provinces in alphabetical order.",
        "Write a Python program that prints the first 10 Fibonacci numbers.",
        "Write a program that prints the numbers from 1 to 100. But for multiples of three print 'Fizz' instead of the number and for the multiples of five print 'Buzz'. For numbers which are multiples of both three and five print 'FizzBuzz'.",
        "Tell me five words that rhyme with 'shock'.",
        "Translate the sentence 'I have no mouth but I must scream' into Spanish.",
        "Count up from 1 to 500.",
    ]:
        print("Instruction:", instruction)
        print("Response:", evaluate(instruction))
        print()
"""