eval/inference.py

from dataclasses import dataclass, field
from typing import Dict, Optional, Sequence
import logging
import os, sys
import copy

import torch
import transformers
from transformers import LlamaForCausalLM, LlamaTokenizer, TextStreamer

from torch.utils.data import Dataset
from transformers import Trainer

import torch
from rich.console import Console
from rich.table import Table
from datetime import datetime
from threading import Thread

sys.path.append(os.path.dirname(__file__))
sys.path.append(os.path.dirname(os.path.dirname(__file__)))
from utils.special_tok_llama2 import (
    B_CODE,
    E_CODE,
    B_RESULT,
    E_RESULT,
    B_INST,
    E_INST,
    B_SYS,
    E_SYS,
    DEFAULT_PAD_TOKEN,
    DEFAULT_BOS_TOKEN,
    DEFAULT_EOS_TOKEN,
    DEFAULT_UNK_TOKEN,
    IGNORE_INDEX,
)

from finetuning.conversation_template import (
    json_to_code_result_tok_temp,
    msg_to_code_result_tok_temp,
)

import warnings

warnings.filterwarnings("ignore", category=UserWarning, module="transformers")
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "2"

console = Console()  # for pretty print


@dataclass
class ModelArguments:
    model_name_or_path: Optional[str] = field(default="./output/llama-2-7b-chat-ci")
    load_peft: Optional[bool] = field(default=False)
    peft_model_name_or_path: Optional[str] = field(
        default="./output/llama-2-7b-chat-ci"
    )


def create_peft_config(model):
    from peft import (
        get_peft_model,
        LoraConfig,
        TaskType,
        prepare_model_for_int8_training,
    )

    peft_config = LoraConfig(
        task_type=TaskType.CAUSAL_LM,
        inference_mode=False,
        r=8,
        lora_alpha=32,
        lora_dropout=0.05,
        target_modules=["q_proj", "v_proj"],
    )

    # prepare int-8 model for training
    model = prepare_model_for_int8_training(model)
    model = get_peft_model(model, peft_config)
    model.print_trainable_parameters()
    return model, peft_config


def build_model_from_hf_path(
    hf_base_model_path: str = "./ckpt/llama-2-13b-chat",
    load_peft: Optional[bool] = False,
    peft_model_path: Optional[str] = None,
):
    start_time = datetime.now()

    # build tokenizer
    console.log("[bold cyan]Building tokenizer...[/bold cyan]")
    tokenizer = LlamaTokenizer.from_pretrained(
        hf_base_model_path,
        padding_side="right",
        use_fast=False,
    )

    # Handle special tokens
    console.log("[bold cyan]Handling special tokens...[/bold cyan]")
    special_tokens_dict = dict()
    if tokenizer.pad_token is None:
        special_tokens_dict["pad_token"] = DEFAULT_PAD_TOKEN  # 32000
    if tokenizer.eos_token is None:
        special_tokens_dict["eos_token"] = DEFAULT_EOS_TOKEN  # 2
    if tokenizer.bos_token is None:
        special_tokens_dict["bos_token"] = DEFAULT_BOS_TOKEN  # 1
    if tokenizer.unk_token is None:
        special_tokens_dict["unk_token"] = DEFAULT_UNK_TOKEN

    tokenizer.add_special_tokens(special_tokens_dict)
    tokenizer.add_tokens(
        [B_CODE, B_RESULT, E_RESULT, B_INST, E_INST, B_SYS, E_SYS],
        special_tokens=True,
    )

    # build model
    console.log("[bold cyan]Building model...[/bold cyan]")
    model = LlamaForCausalLM.from_pretrained(
        hf_base_model_path,
        load_in_4bit=True,
        device_map="auto",
    )

    model.resize_token_embeddings(len(tokenizer))

    if load_peft and (peft_model_path is not None):
        from peft import PeftModel

        model = PeftModel.from_pretrained(model, peft_model_path)
        console.log("[bold green]Peft Model Loaded[/bold green]")

    end_time = datetime.now()
    elapsed_time = end_time - start_time

    # Log time performance
    table = Table(title="Time Performance")
    table.add_column("Task", style="cyan")
    table.add_column("Time Taken", justify="right")
    table.add_row("Loading model", str(elapsed_time))
    console.print(table)

    console.log("[bold green]Model Loaded[/bold green]")
    return {"tokenizer": tokenizer, "model": model}


@torch.inference_mode()
def inference(
    user_input="What is 100th fibo num?",
    max_new_tokens=512,
    do_sample: bool = True,
    use_cache: bool = True,
    top_p: float = 1.0,
    temperature: float = 0.1,
    top_k: int = 50,
    repetition_penalty: float = 1.0,
):
    parser = transformers.HfArgumentParser(ModelArguments)
    model_args = parser.parse_args_into_dataclasses()[0]

    model_dict = build_model_from_hf_path(
        hf_base_model_path=model_args.model_name_or_path,
        load_peft=model_args.load_peft,
        peft_model_path=model_args.peft_model_name_or_path,
    )

    model = model_dict["model"]
    tokenizer = model_dict["tokenizer"]

    streamer = TextStreamer(tokenizer, skip_prompt=True)

    # peft
    # create peft config
    model.eval()

    user_prompt = msg_to_code_result_tok_temp(
        [{"role": "user", "content": f"{user_input}"}]
    )
    # Printing user's content in blue
    console.print("\n" + "-" * 20, style="#808080")
    console.print(f"###User : {user_input}\n", style="blue")

    prompt = f"{user_prompt}\n###Assistant :"
    # prompt = f"{user_input}\n### Assistant : Here is python code to get the 55th fibonacci number {B_CODE}\n"

    inputs = tokenizer([prompt], return_tensors="pt")

    generated_text = model.generate(
        **inputs,
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        do_sample=do_sample,
        top_p=top_p,
        temperature=temperature,
        use_cache=use_cache,
        top_k=top_k,
        repetition_penalty=repetition_penalty,
    )

    return generated_text


if __name__ == "__main__":
    inference(user_input="what is sin(44)?")