base_model: huggyllama/llama-7b
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
load_in_8bit: false
datasets:
  - path: openaccess-ai-collective/jeopardy
    type: jeopardy
dataset_prepared_path:
val_set_size: 0.02
adapter:
lora_model_dir:
sequence_len: 512
max_packed_sequence_len:
lora_r:
lora_alpha:
lora_dropout:
lora_target_modules:
lora_fan_in_fan_out: false
wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:
output_dir: ./jeopardy-bot-7b
gradient_accumulation_steps: 1
micro_batch_size: 1
num_epochs: 4
optimizer: adamw_bnb_8bit
torchdistx_path:
lr_scheduler: cosine
learning_rate: 0.00003
train_on_inputs: false
group_by_length: false
bf16: auto
tf32: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 5
xformers_attention: true
flash_attention:
gptq_groupsize:
gptq_model_v1:
warmup_steps: 20
evals_per_epoch: 4
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.1
fsdp:
fsdp_config:
tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

base_model: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer

load_in_8bit: true
load_in_4bit: false
strict: false

datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
dataset_prepared_path:
val_set_size: 0.05
output_dir: ./lora-out

sequence_len: 4096
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

adapter: lora
lora_model_dir:
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 4
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 10
evals_per_epoch: 4
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

def load_tokenizer(cfg):
    model_config = load_model_config(cfg)
    tokenizer_kwargs = {}
    use_fast = True  # this is the default

    if cfg.tokenizer_use_fast is not None:
        use_fast = cfg.tokenizer_use_fast
    if cfg.tokenizer_legacy is not None:
        # True is the default w/ https://github.com/huggingface/transformers/pull/25224
        tokenizer_kwargs["legacy"] = cfg.tokenizer_legacy

    tokenizer_cls = AutoTokenizer
    if cfg.tokenizer_type:
        tokenizer_cls = getattr(transformers, cfg.tokenizer_type)

    tokenizer = tokenizer_cls.from_pretrained(
        cfg.tokenizer_config,
        trust_remote_code=cfg.trust_remote_code or False,
        use_fast=use_fast,
        **tokenizer_kwargs,
    )

    if (
        tokenizer.__class__.__name__
        in [
            "LlamaTokenizer",
            "LlamaTokenizerFast",
            "CodeLlamaTokenizer",
            "CodeLlamaTokenizerFast",
        ]
        and hasattr(tokenizer, "pad_token")
        and not tokenizer.pad_token
    ):
        # set a pad_token, but use eos_token so we don't add a new token
        tokenizer.pad_token = LLAMA_DEFAULT_EOS_TOKEN

    if tokenizer.__class__.__name__ == "GPTNeoXTokenizerFast":
        tokenizer.add_special_tokens({"pad_token": "[PAD]"})
        os.environ["TOKENIZERS_PARALLELISM"] = "false"

    # Mistral's official FA implementation requires left padding
    if cfg.is_mistral_derived_model and cfg.flash_attention and not cfg.sample_packing:
        tokenizer.padding_side = "left"

    # Qwen base only has single token, so we need to set the special tokens
    if cfg.is_qwen_derived_model:
        token_ids = ["bos_token_id", "eos_token_id", "pad_token_id", "unk_token_id"]
        for attr_name in token_ids:
            if getattr(tokenizer, attr_name) is None:
                setattr(tokenizer, attr_name, tokenizer.eod_id)

        token_names = ["bos_token", "eos_token", "pad_token", "unk_token"]
        for attr_name in token_names:
            if getattr(tokenizer, attr_name) is None:
                setattr(tokenizer, attr_name, "<|endoftext|>")

    additional_special_tokens = None
    if cfg.special_tokens:
        special_tokens = cfg.special_tokens.to_dict()
        additional_special_tokens = special_tokens.pop(
            "additional_special_tokens", None
        )
        lora_modules_to_save = get_linear_embedding_layers(model_config.model_type)
        for k, val in special_tokens.items():
            # check if new special token is not already in tokenizer and
            # is adapter training to make sure lora_modules_to_save is set
            # pylint: disable=too-many-boolean-expressions
            if (
                (getattr(tokenizer, k) is None or getattr(tokenizer, k) != val)
                and (len(tokenizer.encode(val, add_special_tokens=False)) > 2)
                and cfg.adapter
                and (
                    not cfg.lora_modules_to_save
                    or not all(
                        x in cfg.lora_modules_to_save for x in lora_modules_to_save
                    )
                )
            ):
                lora_modules_to_save = ", ".join(
                    [f"`{x}`" for x in lora_modules_to_save]
                )
                raise ValueError(
                    f"Please set lora_modules_to_save to [{lora_modules_to_save}] when using an adapter and changing the special tokens."
                )

            tokenizer.add_special_tokens(
                {k: AddedToken(val, rstrip=False, lstrip=False, normalized=False)}
            )

        # If we add bos_token and eos_token, we need to update the post processor to
        # handle them correctly.
        # https://github.com/huggingface/transformers/pull/24132
        bos_or_eos_in_special_tokens = (
            "bos_token" in cfg.special_tokens and "eos_token" in cfg.special_tokens
        )
        if (
            tokenizer.__class__.__name__
            in (
                "LlamaTokenizerFast",
                "CodeLlamaTokenizerFast",
            )
            and bos_or_eos_in_special_tokens
        ):
            tokenizer.update_post_processor()

    if cfg.tokens:
        tokenizer.add_tokens(
            [
                AddedToken(token, rstrip=False, lstrip=False, normalized=False)
                for token in cfg.tokens
            ]
        )

    # Additional special tokens are a List, and need to be treated differently than regular special
    # tokens. We add them after we have called `add_tokens` in case these additional special tokens
    # are new tokens.
    #
    # Usage:
    #
    # ```py
    # special_tokens:
    #   additional_special_tokens: ["<|im_start|>", "<|im_end|>"]
    # ```
    if additional_special_tokens is not None:
        tokenizer.add_special_tokens(
            {"additional_special_tokens": additional_special_tokens}
        )

    with zero_only():
        LOG.debug(f"EOS: {tokenizer.eos_token_id} / {tokenizer.eos_token}")
        LOG.debug(f"BOS: {tokenizer.bos_token_id} / {tokenizer.bos_token}")
        LOG.debug(f"PAD: {tokenizer.pad_token_id} / {tokenizer.pad_token}")
        LOG.debug(f"UNK: {tokenizer.unk_token_id} / {tokenizer.unk_token}")

    if cfg.chat_template:
        chat_template_string = chat_templates(cfg.chat_template)
        if cfg.default_system_message and cfg.chat_template == "chatml":
            chat_template_string = chat_template_string.replace(
                "You are a helpful assistant.", cfg.default_system_message
            )

        tokenizer.chat_template = chat_template_string
    else:
        LOG.info(
            "No Chat template selected. Consider adding a chat template for easier inference."
        )
    return tokenizer

base_model: codellama/CodeLlama-34b-hf
model_type: LlamaForCausalLM
tokenizer_type: CodeLlamaTokenizer

load_in_8bit: true
load_in_4bit: false
strict: false

datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
dataset_prepared_path:
val_set_size: 0.05
output_dir: ./lora-out

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

adapter: lora
lora_model_dir:
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 4
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
s2_attention:

warmup_steps: 10
evals_per_epoch: 4
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

base_model: openlm-research/open_llama_3b_v2
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
load_in_8bit: false
load_in_4bit: false
strict: false
push_dataset_to_hub:
datasets:
  - path: teknium/GPT4-LLM-Cleaned
    type: alpaca
dataset_prepared_path:
val_set_size: 0.02
adapter:
lora_model_dir:
sequence_len: 1024
sample_packing: true
lora_r:
lora_alpha:
lora_dropout:
lora_target_modules:
lora_target_linear:
lora_fan_in_fan_out:
wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:
output_dir: ./openllama-out
gradient_accumulation_steps: 1
micro_batch_size: 1
num_epochs: 4
optimizer: adamw_bnb_8bit
torchdistx_path:
lr_scheduler: cosine
learning_rate: 0.000003
train_on_inputs: false
group_by_length: false
float16: true
bf16: false
fp16: false
tf32: false
gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
gptq_groupsize:
gptq_model_v1:
warmup_steps: 20
evals_per_epoch: 4
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.1
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

base_model: codellama/CodeLlama-7b-hf
model_type: LlamaForCausalLM
tokenizer_type: CodeLlamaTokenizer

load_in_8bit: true
load_in_4bit: false
strict: false

datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
dataset_prepared_path:
val_set_size: 0.05
output_dir: ./lora-out

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

adapter: lora
lora_model_dir:
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 4
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
s2_attention:

warmup_steps: 10
evals_per_epoch: 4
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

base_model: codellama/CodeLlama-13b-hf
model_type: LlamaForCausalLM
tokenizer_type: CodeLlamaTokenizer

load_in_8bit: true
load_in_4bit: false
strict: false

datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
dataset_prepared_path:
val_set_size: 0.05
output_dir: ./lora-out

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

adapter: lora
lora_model_dir:
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 4
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
s2_attention:

warmup_steps: 10
evals_per_epoch: 4
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

import os

os.environ["CUDA_VISIBLE_DEVICES"] = "1" from peft import PeftConfig, PeftModel from peft import PeftModel, PeftConfig from transformers import AutoModelForCausalLM, AutoTokenizer from datasets import load_dataset import torch import random

peft_model_id = "smangrul/tinyllama_lora_norobots" device = "cuda" config = PeftConfig.from_pretrained(peft_model_id) model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path, load_in_4bit=True, device_map="auto") tokenizer = AutoTokenizer.from_pretrained(peft_model_id) model.resize_token_embeddings(len(tokenizer)) model = PeftModel.from_pretrained(model, peft_model_id, adapter_name="norobots") _ = model.load_adapter("smangrul/tinyllama_lora_sql", adapter_name="sql") _ = model.load_adapter("smangrul/tinyllama_lora_adcopy", adapter_name="adcopy")

%%time

[0.8, 0.1, 0.1] linear #[1.0, 0.2] 0.7 density dare_linear #[1.5, 0.3] 0.5 density ties #[0.8, 0.5] cat

adapters = ["norobots", "adcopy", "sql"] weights = [2.0, 0.3, 0.7] adapter_name = "merge" density = 0.2 combination_type = "ties" if adapter_name in model.peft_config: model.delete_adapter(adapter_name) model.add_weighted_adapter(adapters, weights, adapter_name, combination_type=combination_type, density=density)

model.eval() model.set_adapter("merge")

messages = [ {"role": "user", "content": "Write an essay about Generative AI."}, ] text = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False) inputs = tokenizer(text, return_tensors="pt") # , add_special_tokens=False) inputs = {k: v.to("cuda") for k, v in inputs.items()} outputs = model.generate( **inputs, max_new_tokens=256, do_sample=True, top_p=0.95, temperature=0.2, repetition_penalty=1.2, eos_token_id=tokenizer.eos_token_id, ) print(tokenizer.decode(outputs[0]))

messages = [ {"role": "system", "content": "Create a text ad given the following product and description."}, { "role": "user", "content": "Product: Sony PS5 PlayStation Console\nDescription: The PS5™ console unleashes new gaming possibilities that you never anticipated.", }, ] text = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False) inputs = tokenizer(text, return_tensors="pt") # , add_special_tokens=False) inputs = {k: v.to("cuda") for k, v in inputs.items()} outputs = model.generate( **inputs, max_new_tokens=128, do_sample=True, top_p=0.95, temperature=0.2, repetition_penalty=1.2, eos_token_id=tokenizer.eos_token_id, ) print(tokenizer.decode(outputs[0]))

text = """Table: 2-11365528-2 Columns: ['Team', 'Head Coach', 'President', 'Home Ground', 'Location'] Natural Query: Who is the Head Coach of the team whose President is Mario Volarevic? SQL Query:"""

inputs = tokenizer(text, return_tensors="pt") # , add_special_tokens=False) inputs = {k: v.to("cuda") for k, v in inputs.items()} outputs = model.generate( **inputs, max_new_tokens=64, repetition_penalty=1.1, eos_token_id=tokenizer("</s>").input_ids[-1] ) print(tokenizer.decode(outputs[0]))

def parse_args():
    parser = argparse.ArgumentParser(description="Finetune a transformers model on a causal language modeling task")
    parser.add_argument(
        "--dataset_name",
        type=str,
        default=None,
        help="The name of the dataset to use (via the datasets library).",
    )
    parser.add_argument(
        "--dataset_config_name",
        type=str,
        default=None,
        help="The configuration name of the dataset to use (via the datasets library).",
    )
    parser.add_argument(
        "--train_file", type=str, default=None, help="A csv, txt or a json file containing the training data."
    )
    parser.add_argument(
        "--validation_file", type=str, default=None, help="A csv, txt or a json file containing the validation data."
    )
    parser.add_argument(
        "--validation_split_percentage",
        default=5,
        help="The percentage of the train set used as validation set in case there's no validation split",
    )
    parser.add_argument(
        "--model_name_or_path",
        type=str,
        help="Path to pretrained model or model identifier from huggingface.co/models.",
        required=False,
    )
    parser.add_argument(
        "--config_name",
        type=str,
        default=None,
        help="Pretrained config name or path if not the same as model_name",
    )
    parser.add_argument(
        "--tokenizer_name",
        type=str,
        default=None,
        help="Pretrained tokenizer name or path if not the same as model_name",
    )
    parser.add_argument(
        "--use_slow_tokenizer",
        action="store_true",
        help="If passed, will use a slow tokenizer (not backed by the 🤗 Tokenizers library).",
    )
    parser.add_argument(
        "--per_device_train_batch_size",
        type=int,
        default=8,
        help="Batch size (per device) for the training dataloader.",
    )
    parser.add_argument(
        "--per_device_eval_batch_size",
        type=int,
        default=8,
        help="Batch size (per device) for the evaluation dataloader.",
    )
    parser.add_argument(
        "--learning_rate",
        type=float,
        default=5e-5,
        help="Initial learning rate (after the potential warmup period) to use.",
    )
    parser.add_argument("--weight_decay", type=float, default=0.0, help="Weight decay to use.")
    parser.add_argument("--num_train_epochs", type=int, default=3, help="Total number of training epochs to perform.")
    parser.add_argument(
        "--max_train_steps",
        type=int,
        default=None,
        help="Total number of training steps to perform. If provided, overrides num_train_epochs.",
    )
    parser.add_argument(
        "--gradient_accumulation_steps",
        type=int,
        default=1,
        help="Number of updates steps to accumulate before performing a backward/update pass.",
    )
    parser.add_argument(
        "--lr_scheduler_type",
        type=SchedulerType,
        default="linear",
        help="The scheduler type to use.",
        choices=["linear", "cosine", "cosine_with_restarts", "polynomial", "constant", "constant_with_warmup"],
    )
    parser.add_argument(
        "--num_warmup_steps", type=int, default=0, help="Number of steps for the warmup in the lr scheduler."
    )
    parser.add_argument("--output_dir", type=str, default=None, help="Where to store the final model.")
    parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
    parser.add_argument(
        "--model_type",
        type=str,
        default=None,
        help="Model type to use if training from scratch.",
        choices=MODEL_TYPES,
    )
    parser.add_argument(
        "--ignore_pad_token_for_loss",
        type=bool,
        default=True,
        help="Whether to ignore the tokens corresponding to padded labels in the loss computation or not.",
    )
    parser.add_argument(
        "--max_source_length",
        type=int,
        default=128,
        help=(
            "The maximum total input sequence length after "
            "tokenization.Sequences longer than this will be truncated, sequences shorter will be padded."
        ),
    )
    parser.add_argument(
        "--max_target_length",
        type=int,
        default=128,
        help=(
            "The maximum total sequence length for target text after "
            "tokenization. Sequences longer than this will be truncated, sequences shorter will be padded."
            "during ``evaluate`` and ``predict``."
        ),
    )
    parser.add_argument(
        "--pad_to_max_length",
        action="store_true",
        help="If passed, pad all samples to `max_length`. Otherwise, dynamic padding is used.",
    )
    parser.add_argument(
        "--preprocessing_num_workers",
        type=int,
        default=None,
        help="The number of processes to use for the preprocessing.",
    )
    parser.add_argument(
        "--overwrite_cache", action="store_true", help="Overwrite the cached training and evaluation sets"
    )
    parser.add_argument(
        "--no_keep_linebreaks", action="store_true", help="Do not keep line breaks when using TXT files."
    )
    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
    parser.add_argument(
        "--hub_model_id", type=str, help="The name of the repository to keep in sync with the local `output_dir`."
    )
    parser.add_argument("--hub_token", type=str, help="The token to use to push to the Model Hub.")
    parser.add_argument(
        "--trust_remote_code",
        type=bool,
        default=False,
        help=(
            "Whether or not to allow for custom models defined on the Hub in their own modeling files. This option"
            "should only be set to `True` for repositories you trust and in which you have read the code, as it will"
            "execute code present on the Hub on your local machine."
        ),
    )
    parser.add_argument(
        "--checkpointing_steps",
        type=str,
        default=None,
        help="Whether the various states should be saved at the end of every n steps, or 'epoch' for each epoch.",
    )
    parser.add_argument(
        "--resume_from_checkpoint",
        type=str,
        default=None,
        help="If the training should continue from a checkpoint folder.",
    )
    parser.add_argument(
        "--with_tracking",
        action="store_true",
        help="Whether to enable experiment trackers for logging.",
    )
    parser.add_argument(
        "--report_to",
        type=str,
        default="tensorboard",
        help=(
            'The integration to report the results and logs to. Supported platforms are `"tensorboard"`,'
            ' `"wandb"`, `"comet_ml"` and `"clearml"`. Use `"all"` (default) to report to all integrations.'
            "Only applicable when `--with_tracking` is passed."
        ),
    )
    parser.add_argument(
        "--low_cpu_mem_usage",
        action="store_true",
        help=(
            "It is an option to create the model as an empty shell, then only materialize its parameters when the pretrained weights are loaded."
            "If passed, LLM loading time and RAM consumption will be benefited."
        ),
    )
    ##########################
    #   Generation Config    #
    ##########################
    parser.add_argument(
        "--temperature",
        type=float,
        default=0.8,
        help="temperature of 1.0 has no effect, lower tend toward greedy sampling",
    )
    parser.add_argument("--k", type=int, default=40, help="Choose k candidate words")
    parser.add_argument("--p", type=float, default=0.95, help="The sum of probability of candidate words is 0.9 ")

    ##########################
    #        Exp Args        #
    ##########################
    parser.add_argument(
        "--adapter_name_or_path",
        type=str,
        default=None,
        help=(
            "The LoRA adapter checkpoint. Set None if you want to fine-tune from LoftQ."
            "Specify a path if you want to evaluate."
        ),
    )

    args = parser.parse_args()

    # Sanity checks
    if args.dataset_name is None and args.train_file is None and args.validation_file is None:
        raise ValueError("Need either a dataset name or a training/validation file.")
    else:
        if args.train_file is not None:
            extension = args.train_file.split(".")[-1]
            assert extension in ["csv", "json", "txt"], "`train_file` should be a csv, json or txt file."
        if args.validation_file is not None:
            extension = args.validation_file.split(".")[-1]
            assert extension in ["csv", "json", "txt"], "`validation_file` should be a csv, json or txt file."

    if args.push_to_hub:
        assert args.output_dir is not None, "Need an `output_dir` to create a repo when `--push_to_hub` is passed."

    return args

# Training PEFT models with new tokens being added to the embedding layers and tokenizer

In this example, we will learn how to train a LoRA model when adding new tokens to the tokenizer and model. 
This is a common usecase when doing the following:
1. Instruction finetuning with new tokens beind added such as `<|user|>`, `<|assistant|>`, `<|system|>`, `</s>`, `<s>` to properly format the conversations
2. Finetuning on a specific language wherein language spoecific tokens are added, e.g., korean tokens being added to vocabulary for finetuning LLM on Korean datasets.
3. Instruction finetuning to return outputs in certain format to enable agent behaviour new tokens such as `<|FUNCTIONS|>`, `<|BROWSE|>`, `<|TEXT2IMAGE|>`, `<|ASR|>`, `<|TTS|>`, `<|GENERATECODE|>`, `<|RAG|>`.

In such cases, you add the Embedding modules to the LORA `target_modules`. PEFT will take care of saving the embedding layers with the new added tokens along with the adapter weights that were trained on the specific initialization of the embeddings weights of the added tokens.
Let's import the necessary libraries

import os

os.environ["CUDA_VISIBLE_DEVICES"] = "3" os.environ["WANDB_PROJECT"] = "PeftExamples" import transformers from peft import ( LoraConfig, PeftConfig, PeftModel, get_peft_model, prepare_model_for_kbit_training, ) from transformers import ( AutoModelForCausalLM, AutoTokenizer, HfArgumentParser, TrainingArguments, Trainer, default_data_collator, ) import torch from dataclasses import dataclass, field from typing import Optional from dataclass_csv import DataclassReader from torch.utils.data import Dataset, DataLoader

from enum import Enum

## Prepare Model and Tokenizer
Now, we will be adding 27 new tokens as well as replace the existing pad, bos and eos tokens of the model.

class SpecialTokens(str, Enum): begin_target = "<|begintarget|>" end_target = "<|endtarget|>" begin_context = "<|begincontext|>" end_context = "<|endcontext|>" system = "<|system|>" user = "<|user|>" begin_last_user_utterance = "<|beginlastuserutterance|>" end_last_user_utterance = "<|endlastuserutterance|>" begin_dsts = "<|begindsts|>" end_dsts = "<|enddsts|>" begin_dst = "<|begindst|>" end_dst = "<|enddst|>" begin_belief = "<|beginbelief|>" end_belief = "<|endbelief|>" begin_response = "<|beginresponse|>" end_response = "<|endresponse|>" begin_action = "<|beginaction|>" end_action = "<|endaction|>" begin_user_action = "<|beginuseraction|>" end_user_action = "<|enduseraction|>" sys_actions = "<|sysactions|>" begin_intent = "<|beginintent|>" end_intent = "<|endintent|>" begin_requested_slots = "<|beginrequestedslots|>" end_requested_slots = "<|endrequestedslots|>" pad_token = "<|pad|>" bos_token = "<|startoftext|>"

@classmethod
def list(cls):
    return [c.value for c in cls]

We will be finetuning Mistral-7B model. Let's load the tokenizer and add the special tokens followed by loading the base model and resizzing the embedding layers to accomodate the newly added tokens.

model_name = "mistralai/Mistral-7B-v0.1" tokenizer = AutoTokenizer.from_pretrained( model_name, pad_token=SpecialTokens.pad_token.value, bos_token=SpecialTokens.bos_token.value, eos_token=SpecialTokens.end_target.value, additional_special_tokens=SpecialTokens.list(), ) model = AutoModelForCausalLM.from_pretrained( model_name, low_cpu_mem_usage=True # use_flash_attention_2=True, # leading to an error ) model.resize_token_embeddings(len(tokenizer))

## Apply LoRA

config = LoraConfig( r=64, lora_alpha=128, lora_dropout=0.0, target_modules=["embed_tokens", "lm_head", "q_proj", "v_proj"] ) model = get_peft_model(model, config) print(model.print_trainable_parameters()) print(model)

## Preapre Dataset

from datasets import load_dataset

dataset = load_dataset("smangrul/assistant_chatbot_dataset") dataset = dataset["train"].train_test_split(0.2)

text_column = "context" label_column = "target" max_length = 512

def preprocess_function(examples): batch_size = len(examples[text_column]) targets = [str(x) for x in examples[label_column]] model_inputs = tokenizer(examples[text_column]) labels = tokenizer(targets, add_special_tokens=False) # don't add bos token because we concatenate with inputs for i in range(batch_size): sample_input_ids = model_inputs["input_ids"][i] label_input_ids = labels["input_ids"][i] + [tokenizer.eos_token_id] # print(i, sample_input_ids, label_input_ids) model_inputs["input_ids"][i] = sample_input_ids + label_input_ids labels["input_ids"][i] = [-100] * len(sample_input_ids) + label_input_ids model_inputs["attention_mask"][i] = [1] * len(model_inputs["input_ids"][i]) # print(model_inputs) for i in range(batch_size): sample_input_ids = model_inputs["input_ids"][i] label_input_ids = labels["input_ids"][i] model_inputs["input_ids"][i] = [tokenizer.pad_token_id] * ( max_length - len(sample_input_ids) ) + sample_input_ids model_inputs["attention_mask"][i] = [0] * (max_length - len(sample_input_ids)) + model_inputs[ "attention_mask" ][i] labels["input_ids"][i] = [-100] * (max_length - len(sample_input_ids)) + label_input_ids model_inputs["input_ids"][i] = model_inputs["input_ids"][i][:max_length] model_inputs["attention_mask"][i] = model_inputs["attention_mask"][i][:max_length] labels["input_ids"][i] = labels["input_ids"][i][:max_length] model_inputs["labels"] = labels["input_ids"] return model_inputs

processed_datasets = dataset.map( preprocess_function, batched=True, num_proc=1, remove_columns=dataset["train"].column_names, load_from_cache_file=False, desc="Running tokenizer on dataset", )

train_dataset = processed_datasets["train"]

train_dataset

train_dataloader = DataLoader( train_dataset, shuffle=True, collate_fn=default_data_collator, batch_size=8, pin_memory=True )

next(iter(train_dataloader))

tokenizer.decode(train_dataset[0]["input_ids"])

# Train the model

training_args = TrainingArguments( output_dir="mistral_lora_clm_with_added_tokens", num_train_epochs=2, save_total_limit=5, per_device_train_batch_size=8, warmup_steps=10, weight_decay=0.0001, dataloader_drop_last=True, bf16=True, logging_steps=10, learning_rate=1e-5, gradient_checkpointing=True, gradient_checkpointing_kwargs={"use_reentrant": False}, remove_unused_columns=False, hub_model_id="smangrul/mistral_lora_clm_with_added_tokens", push_to_hub=True, hub_private_repo=True, ) trainer = Trainer( model=model, args=training_args, train_dataset=train_dataset, data_collator=default_data_collator, )

model.config.use_cache = False

trainer.train()

# Check the model output on a sample from evaluation dataset

import random

i = random.randint(0, len(dataset["test"])) context = dataset["test"][i]["context"]

batch = tokenizer(context, return_tensors="pt") batch = {k: v.to("cuda") for k, v in batch.items()} model.eval() output_tokens = model.generate( **batch, max_new_tokens=256, do_sample=True, temperature=0.2, top_p=0.95, top_k=50, eos_token_id=tokenizer.eos_token_id, pad_token_id=tokenizer.pad_token_id, ) target_predicted = tokenizer.decode(output_tokens[0], skip_special_tokens=False).split("<|endcontext|>")[1] target = dataset["test"][i]["target"] print(f"{context=} \n\n {target_predicted=} \n\n {target=}")

# Save the Adapter model 
When the lora layers are applied to embedding layers, the corresponding base model embedding layers are also saved. 

trainer.push_to_hub() trainer.model.push_to_hub(training_args.output_dir)

# Check the model loading is working as expected and generating plausible outputs.

from peft import PeftModel

inference_model = AutoModelForCausalLM.from_pretrained( model_name, low_cpu_mem_usage=True, # use_flash_attention_2=True, ) inference_model.resize_token_embeddings(len(tokenizer))

inference_model = PeftModel.from_pretrained(inference_model, "smangrul/mistral_lora_clm_with_added_tokens") inference_model.to("cuda") inference_model.eval()

output_tokens = inference_model.generate( **batch, max_new_tokens=256, do_sample=True, temperature=0.2, top_p=0.95, top_k=50, eos_token_id=tokenizer.eos_token_id, pad_token_id=tokenizer.pad_token_id, )

target_predicted = tokenizer.decode(output_tokens[0], skip_special_tokens=False).split("<|endcontext|>")[1] print(f"{context=} \n\n {target_predicted=} \n\n {target=}")

logger = get_logger(__name__)

require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")

MODEL_CONFIG_CLASSES = list(MODEL_MAPPING.keys())
MODEL_TYPES = tuple(conf.model_type for conf in MODEL_CONFIG_CLASSES)

        # the entire model is fine-tuned.
        tokens += [sep_token]
        token_boxes += [sep_token_box]
        actual_bboxes += [[0, 0, width, height]]
        label_ids += [pad_token_label_id]
        if sep_token_extra:
            # roberta uses an extra separator b/w pairs of sentences
            tokens += [sep_token]
            token_boxes += [sep_token_box]
            actual_bboxes += [[0, 0, width, height]]
            label_ids += [pad_token_label_id]
        segment_ids = [sequence_a_segment_id] * len(tokens)

        if cls_token_at_end:
            tokens += [cls_token]
            token_boxes += [cls_token_box]
            actual_bboxes += [[0, 0, width, height]]
            label_ids += [pad_token_label_id]
            segment_ids += [cls_token_segment_id]
        else:
            tokens = [cls_token] + tokens
            token_boxes = [cls_token_box] + token_boxes
            actual_bboxes = [[0, 0, width, height]] + actual_bboxes
            label_ids = [pad_token_label_id] + label_ids
            segment_ids = [cls_token_segment_id] + segment_ids

        input_ids = tokenizer.convert_tokens_to_ids(tokens)

        # The mask has 1 for real tokens and 0 for padding tokens. Only real
        # tokens are attended to.
        input_mask = [1 if mask_padding_with_zero else 0] * len(input_ids)

        # Zero-pad up to the sequence length.
        padding_length = max_seq_length - len(input_ids)
        if pad_on_left:
            input_ids = ([pad_token] * padding_length) + input_ids
            input_mask = ([0 if mask_padding_with_zero else 1] * padding_length) + input_mask
            segment_ids = ([pad_token_segment_id] * padding_length) + segment_ids
            label_ids = ([pad_token_label_id] * padding_length) + label_ids
            token_boxes = ([pad_token_box] * padding_length) + token_boxes
        else:
            input_ids += [pad_token] * padding_length
            input_mask += [0 if mask_padding_with_zero else 1] * padding_length
            segment_ids += [pad_token_segment_id] * padding_length
            label_ids += [pad_token_label_id] * padding_length
            token_boxes += [pad_token_box] * padding_length

        assert len(input_ids) == max_seq_length
        assert len(input_mask) == max_seq_length
        assert len(segment_ids) == max_seq_length
        assert len(label_ids) == max_seq_length
        assert len(token_boxes) == max_seq_length

        if ex_index < 5:
            logger.info("*** Example ***")
            logger.info("guid: %s", example.guid)
            logger.info("tokens: %s", " ".join([str(x) for x in tokens]))
            logger.info("input_ids: %s", " ".join([str(x) for x in input_ids]))
            logger.info("input_mask: %s", " ".join([str(x) for x in input_mask]))
            logger.info("segment_ids: %s", " ".join([str(x) for x in segment_ids]))
            logger.info("label_ids: %s", " ".join([str(x) for x in label_ids]))
            logger.info("boxes: %s", " ".join([str(x) for x in token_boxes]))
            logger.info("actual_bboxes: %s", " ".join([str(x) for x in actual_bboxes]))

        features.append(
            InputFeatures(
                input_ids=input_ids,
                input_mask=input_mask,
                segment_ids=segment_ids,
                label_ids=label_ids,
                boxes=token_boxes,
                actual_bboxes=actual_bboxes,
                file_name=file_name,
                page_size=page_size,
            )
        )
    return features

from transformers import LayoutLMTokenizer

# from .unilm.layoutlm.data.funsd import FunsdDataset, InputFeatures
from torch.utils.data import DataLoader, RandomSampler, SequentialSampler

batch_size = 16
args = {
    "local_rank": -1,
    "overwrite_cache": True,
    "data_dir": "/home/sourab/temp/data/",
    "model_name_or_path": "microsoft/layoutlm-base-uncased",
    "max_seq_length": 512,
    "model_type": "layoutlm",
}


# class to turn the keys of a dict into attributes (thanks Stackoverflow)
class AttrDict(dict):
    def __init__(self, *args, **kwargs):
        super(AttrDict, self).__init__(*args, **kwargs)
        self.__dict__ = self


args = AttrDict(args)

tokenizer = LayoutLMTokenizer.from_pretrained("microsoft/layoutlm-base-uncased")

# the LayoutLM authors already defined a specific FunsdDataset, so we are going to use this here
train_dataset = FunsdDataset(args, tokenizer, labels, pad_token_label_id, mode="train")
train_sampler = RandomSampler(train_dataset)
train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=batch_size)

eval_dataset = FunsdDataset(args, tokenizer, labels, pad_token_label_id, mode="test")
eval_sampler = SequentialSampler(eval_dataset)
eval_dataloader = DataLoader(eval_dataset, sampler=eval_sampler, batch_size=batch_size)

len(train_dataloader)

len(eval_dataloader)

batch = next(iter(train_dataloader))
input_ids = batch[0][0]
tokenizer.decode(input_ids)

Define and fine-tune the model

As this is a sequence labeling task, we are going to load LayoutLMForTokenClassification (the base sized model) from the hub. We are going to fine-tune it on a downstream task, namely FUNSD.

from peft import get_peft_config, PeftModel, get_peft_model, LoraConfig, TaskType

peft_config = LoraConfig(
    task_type=TaskType.TOKEN_CLS, inference_mode=False, r=16, lora_alpha=16, lora_dropout=0.1, bias="all"
)
peft_config

from transformers import LayoutLMForTokenClassification
import torch
from transformers import set_seed

seed = 100
set_seed(seed)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

model = LayoutLMForTokenClassification.from_pretrained("microsoft/layoutlm-base-uncased", num_labels=num_labels)
model = get_peft_model(model, peft_config)
model.to(device)

print(model.model.layoutlm.encoder.layer[0].attention.self.query.weight)
print(model.model.layoutlm.encoder.layer[0].attention.self.query.lora_A.weight)
print(model.model.classifier.weight)

Now we can start training:

from transformers import AdamW, get_linear_schedule_with_warmup
from tqdm import tqdm

num_train_epochs = 100

optimizer = torch.optim.AdamW(model.parameters(), lr=3e-3)
lr_scheduler = get_linear_schedule_with_warmup(
    optimizer=optimizer,
    num_warmup_steps=0.06 * (len(train_dataloader) * num_train_epochs),
    num_training_steps=(len(train_dataloader) * num_train_epochs),
)


global_step = 0

t_total = len(train_dataloader) * num_train_epochs  # total number of training steps

# put the model in training mode
model.train()
for epoch in range(num_train_epochs):
    for batch in tqdm(train_dataloader, desc="Training"):
        input_ids = batch[0].to(device)
        bbox = batch[4].to(device)
        attention_mask = batch[1].to(device)
        token_type_ids = batch[2].to(device)
        labels = batch[3].to(device)

        # forward pass
        outputs = model(
            input_ids=input_ids, bbox=bbox, attention_mask=attention_mask, token_type_ids=token_type_ids, labels=labels
        )
        loss = outputs.loss

        # print loss every 100 steps
        if global_step % 10 == 0:
            print(f"Loss after {global_step} steps: {loss.item()}")

        # backward pass to get the gradients
        loss.backward()

        # print("Gradients on classification head:")
        # print(model.classifier.weight.grad[6,:].sum())

        # update
        optimizer.step()
        lr_scheduler.step()
        optimizer.zero_grad()
        global_step += 1

import numpy as np
from seqeval.metrics import (
    classification_report,
    f1_score,
    precision_score,
    recall_score,
)

eval_loss = 0.0
nb_eval_steps = 0
preds = None
out_label_ids = None

# put model in evaluation mode
model.eval()
for batch in tqdm(eval_dataloader, desc="Evaluating"):
    with torch.no_grad():
        input_ids = batch[0].to(device)
        bbox = batch[4].to(device)
        attention_mask = batch[1].to(device)
        token_type_ids = batch[2].to(device)
        labels = batch[3].to(device)

        # forward pass
        outputs = model(
            input_ids=input_ids, bbox=bbox, attention_mask=attention_mask, token_type_ids=token_type_ids, labels=labels
        )
        # get the loss and logits
        tmp_eval_loss = outputs.loss
        logits = outputs.logits

        eval_loss += tmp_eval_loss.item()
        nb_eval_steps += 1

        # compute the predictions
        if preds is None:
            preds = logits.detach().cpu().numpy()
            out_label_ids = labels.detach().cpu().numpy()
        else:
            preds = np.append(preds, logits.detach().cpu().numpy(), axis=0)
            out_label_ids = np.append(out_label_ids, labels.detach().cpu().numpy(), axis=0)

# compute average evaluation loss
eval_loss = eval_loss / nb_eval_steps
preds = np.argmax(preds, axis=2)

out_label_list = [[] for _ in range(out_label_ids.shape[0])]
preds_list = [[] for _ in range(out_label_ids.shape[0])]

for i in range(out_label_ids.shape[0]):
    for j in range(out_label_ids.shape[1]):
        if out_label_ids[i, j] != pad_token_label_id:
            out_label_list[i].append(label_map[out_label_ids[i][j]])
            preds_list[i].append(label_map[preds[i][j]])

results = {
    "loss": eval_loss,
    "precision": precision_score(out_label_list, preds_list),
    "recall": recall_score(out_label_list, preds_list),
    "f1": f1_score(out_label_list, preds_list),
}
print(results)

model.print_trainable_parameters()

model.save_pretrained("peft_layoutlm")

!du -h "peft_layoutlm/adapter_model.bin"

def write_model(model_path, input_base_path, tokenizer_path=None, safe_serialization=True, fix_eos_token_id=True):
    os.makedirs(model_path, exist_ok=True)
    tmp_model_path = os.path.join(model_path, "tmp")
    os.makedirs(tmp_model_path, exist_ok=True)

    config_path = Path(input_base_path) / "config.yaml"
    olmo_config = yaml.safe_load(config_path.read_text())["model"]

    n_layers = olmo_config["n_layers"]
    n_heads = olmo_config["n_heads"]
    dim = olmo_config["d_model"]
    dims_per_head = dim // n_heads
    base = 10000.0
    inv_freq = 1.0 / (base ** (torch.arange(0, dims_per_head, 2).float() / dims_per_head))
    max_position_embeddings = olmo_config["max_sequence_length"]

    vocab_size = olmo_config.get("embedding_size", olmo_config["vocab_size"])

    if olmo_config.get("n_kv_heads", None) is not None:
        num_key_value_heads = olmo_config["n_kv_heads"]  # for GQA / MQA
    elif olmo_config["multi_query_attention"]:  # compatibility with other checkpoints
        num_key_value_heads = 1
    else:
        num_key_value_heads = n_heads

    print(f"Fetching all parameters from the checkpoint at {input_base_path}.")

    # Not sharded
    # (The sharded implementation would also work, but this is simpler.)
    loaded = torch.load(os.path.join(input_base_path, "model.pt"), map_location="cpu")

    param_count = 0
    index_dict = {"weight_map": {}}
    for layer_i in range(n_layers):
        filename = f"pytorch_model-{layer_i + 1}-of-{n_layers + 1}.bin"
        # Unsharded
        # TODO: Layernorm stuff
        # TODO: multi query attention
        fused_dims = [dim, dims_per_head * num_key_value_heads, dims_per_head * num_key_value_heads]
        q_proj_weight, k_proj_weight, v_proj_weight = torch.split(
            loaded[f"transformer.blocks.{layer_i}.att_proj.weight"], fused_dims, dim=0
        )
        up_proj_weight, gate_proj_weight = torch.chunk(
            loaded[f"transformer.blocks.{layer_i}.ff_proj.weight"], 2, dim=0
        )
        state_dict = {
            f"model.layers.{layer_i}.self_attn.q_proj.weight": q_proj_weight,
            f"model.layers.{layer_i}.self_attn.k_proj.weight": k_proj_weight,
            f"model.layers.{layer_i}.self_attn.v_proj.weight": v_proj_weight,
            f"model.layers.{layer_i}.self_attn.o_proj.weight": loaded[f"transformer.blocks.{layer_i}.attn_out.weight"],
            f"model.layers.{layer_i}.mlp.gate_proj.weight": gate_proj_weight,
            f"model.layers.{layer_i}.mlp.down_proj.weight": loaded[f"transformer.blocks.{layer_i}.ff_out.weight"],
            f"model.layers.{layer_i}.mlp.up_proj.weight": up_proj_weight,
        }

        state_dict[f"model.layers.{layer_i}.self_attn.rotary_emb.inv_freq"] = inv_freq

        for k, v in state_dict.items():
            index_dict["weight_map"][k] = filename
            param_count += v.numel()
        torch.save(state_dict, os.path.join(tmp_model_path, filename))

    filename = f"pytorch_model-{n_layers + 1}-of-{n_layers + 1}.bin"

    # Unsharded
    # TODO: Deal with weight-tying
    state_dict = {
        "model.embed_tokens.weight": loaded["transformer.wte.weight"],
        "lm_head.weight": loaded["transformer.ff_out.weight"]
        if "transformer.ff_out.weight" in loaded
        else loaded["transformer.wte.weight"],
    }

    for k, v in state_dict.items():
        index_dict["weight_map"][k] = filename
        param_count += v.numel()
    torch.save(state_dict, os.path.join(tmp_model_path, filename))

    # Write configs
    index_dict["metadata"] = {"total_size": param_count * 2}
    write_json(index_dict, os.path.join(tmp_model_path, "pytorch_model.bin.index.json"))

    if olmo_config.get("mlp_hidden_size", None) is not None:
        intermediate_size = olmo_config["mlp_hidden_size"] // 2
    else:
        intermediate_size = (dim * olmo_config["mlp_ratio"]) // 2

    config = OlmoConfig(
        vocab_size=vocab_size,
        hidden_size=dim,
        intermediate_size=intermediate_size,
        num_hidden_layers=n_layers,
        num_attention_heads=n_heads,
        num_key_value_heads=num_key_value_heads,
        max_position_embeddings=max_position_embeddings,
        pad_token_id=olmo_config["pad_token_id"],
        bos_token_id=None,
        eos_token_id=olmo_config["eos_token_id"],
        tie_word_embeddings=olmo_config["weight_tying"],
        rope_theta=base,
        clip_qkv=olmo_config.get("clip_qkv"),
    )
    config.save_pretrained(tmp_model_path)

    # Make space so we can load the model properly now.
    del state_dict
    del loaded
    gc.collect()

    if tokenizer_path is not None:
        _write_tokenizer(model_path, config, tokenizer_path, fix_eos_token_id)

    print("Loading the checkpoint in a OLMo model.")
    model = OlmoForCausalLM.from_pretrained(tmp_model_path, torch_dtype=torch.float32, low_cpu_mem_usage=True)
    # Avoid saving this as part of the config.
    del model.config._name_or_path
    print("Saving in the Transformers format.")
    model.save_pretrained(model_path, safe_serialization=safe_serialization)
    shutil.rmtree(tmp_model_path)

def main():
    # See all possible arguments in src/transformers/training_args.py
    # or by passing the --help flag to this script.
    # We now keep distinct sets of args, for a cleaner separation of concerns.

    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
        # If we pass only one argument to the script and it's the path to a json file,
        # let's parse it to get our arguments.
        model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
    else:
        model_args, data_args, training_args = parser.parse_args_into_dataclasses()

    if model_args.use_auth_token is not None:
        warnings.warn(
            "The `use_auth_token` argument is deprecated and will be removed in v4.34. Please use `token` instead.",
            FutureWarning,
        )
        if model_args.token is not None:
            raise ValueError("`token` and `use_auth_token` are both specified. Please set only the argument `token`.")
        model_args.token = model_args.use_auth_token

    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
    # information sent is the one passed as arguments along with your Python/PyTorch versions.
    send_example_telemetry("run_plm", model_args, data_args)

    # Setup logging
    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
        datefmt="%m/%d/%Y %H:%M:%S",
        handlers=[logging.StreamHandler(sys.stdout)],
    )

    if training_args.should_log:
        # The default of training_args.log_level is passive, so we set log level at info here to have that default.
        transformers.utils.logging.set_verbosity_info()

    log_level = training_args.get_process_log_level()
    logger.setLevel(log_level)
    datasets.utils.logging.set_verbosity(log_level)
    transformers.utils.logging.set_verbosity(log_level)
    transformers.utils.logging.enable_default_handler()
    transformers.utils.logging.enable_explicit_format()

    # Log on each process the small summary:
    logger.warning(
        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}, "
        + f"distributed training: {training_args.parallel_mode.value == 'distributed'}, 16-bits training: {training_args.fp16}"
    )
    logger.info(f"Training/evaluation parameters {training_args}")

    # Detecting last checkpoint.
    last_checkpoint = None
    if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
        last_checkpoint = get_last_checkpoint(training_args.output_dir)
        if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
            raise ValueError(
                f"Output directory ({training_args.output_dir}) already exists and is not empty. "
                "Use --overwrite_output_dir to overcome."
            )
        elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
            logger.info(
                f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
                "the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
            )

    # Set seed before initializing model.
    set_seed(training_args.seed)

    # Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
    # or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
    # (the dataset will be downloaded automatically from the datasets Hub).
    #
    # For CSV/JSON files, this script will use the column called 'text' or the first column if no column called
    # 'text' is found. You can easily tweak this behavior (see below).
    #
    # In distributed training, the load_dataset function guarantee that only one local process can concurrently
    # download the dataset.
    if data_args.dataset_name is not None:
        # Downloading and loading a dataset from the hub.
        raw_datasets = load_dataset(
            data_args.dataset_name,
            data_args.dataset_config_name,
            cache_dir=model_args.cache_dir,
            token=model_args.token,
        )
        if "validation" not in raw_datasets.keys():
            raw_datasets["validation"] = load_dataset(
                data_args.dataset_name,
                data_args.dataset_config_name,
                split=f"train[:{data_args.validation_split_percentage}%]",
                cache_dir=model_args.cache_dir,
                token=model_args.token,
            )
            raw_datasets["train"] = load_dataset(
                data_args.dataset_name,
                data_args.dataset_config_name,
                split=f"train[{data_args.validation_split_percentage}%:]",
                cache_dir=model_args.cache_dir,
                token=model_args.token,
            )
    else:
        data_files = {}
        if data_args.train_file is not None:
            data_files["train"] = data_args.train_file
            extension = data_args.train_file.split(".")[-1]
        if data_args.validation_file is not None:
            data_files["validation"] = data_args.validation_file
            extension = data_args.validation_file.split(".")[-1]
        if extension == "txt":
            extension = "text"
        raw_datasets = load_dataset(extension, data_files=data_files, cache_dir=model_args.cache_dir)
        # If no validation data is there, validation_split_percentage will be used to divide the dataset.
        if "validation" not in raw_datasets.keys():
            raw_datasets["validation"] = load_dataset(
                extension,
                data_files=data_files,
                split=f"train[:{data_args.validation_split_percentage}%]",
                cache_dir=model_args.cache_dir,
                token=model_args.token,
            )
            raw_datasets["train"] = load_dataset(
                extension,
                data_files=data_files,
                split=f"train[{data_args.validation_split_percentage}%:]",
                cache_dir=model_args.cache_dir,
                token=model_args.token,
            )

    # See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at
    # https://huggingface.co/docs/datasets/loading_datasets.

    # Load pretrained model and tokenizer
    #
    # Distributed training:
    # The .from_pretrained methods guarantee that only one local process can concurrently
    # download model & vocab.
    config_kwargs = {
        "cache_dir": model_args.cache_dir,
        "revision": model_args.model_revision,
        "token": model_args.token,
    }
    if model_args.config_name:
        config = AutoConfig.from_pretrained(model_args.config_name, **config_kwargs)
    elif model_args.model_name_or_path:
        config = AutoConfig.from_pretrained(model_args.model_name_or_path, **config_kwargs)
    else:
        config = XLNetConfig()
        logger.warning("You are instantiating a new config instance from scratch.")
        if model_args.config_overrides is not None:
            logger.info(f"Overriding config: {model_args.config_overrides}")
            config.update_from_string(model_args.config_overrides)
            logger.info(f"New config: {config}")

    tokenizer_kwargs = {
        "cache_dir": model_args.cache_dir,
        "use_fast": model_args.use_fast_tokenizer,
        "revision": model_args.model_revision,
        "token": model_args.token,
    }
    if model_args.tokenizer_name:
        tokenizer = AutoTokenizer.from_pretrained(model_args.tokenizer_name, **tokenizer_kwargs)
    elif model_args.model_name_or_path:
        tokenizer = AutoTokenizer.from_pretrained(model_args.model_name_or_path, **tokenizer_kwargs)
    else:
        raise ValueError(
            "You are instantiating a new tokenizer from scratch. This is not supported by this script. "
            "You can do it from another script, save it, and load it from here, using --tokenizer_name."
        )

    if model_args.model_name_or_path:
        model = XLNetLMHeadModel.from_pretrained(
            model_args.model_name_or_path,
            from_tf=bool(".ckpt" in model_args.model_name_or_path),
            config=config,
            cache_dir=model_args.cache_dir,
            revision=model_args.model_revision,
            token=model_args.token,
            low_cpu_mem_usage=model_args.low_cpu_mem_usage,
        )
    else:
        logger.info("Training new model from scratch")
        model = XLNetLMHeadModel(config)

    # We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch
    # on a small vocab and want a smaller embedding size, remove this test.
    embedding_size = model.get_input_embeddings().weight.shape[0]
    if len(tokenizer) > embedding_size:
        model.resize_token_embeddings(len(tokenizer))

    # Preprocessing the datasets.
    # First we tokenize all the texts.
    if training_args.do_train:
        column_names = raw_datasets["train"].column_names
    else:
        column_names = raw_datasets["validation"].column_names
    text_column_name = "text" if "text" in column_names else column_names[0]

    if data_args.max_seq_length > tokenizer.model_max_length:
        logger.warning(
            f"The max_seq_length passed ({data_args.max_seq_length}) is larger than the maximum length for the "
            f"model ({tokenizer.model_max_length}). Using max_seq_length={tokenizer.model_max_length}."
        )
    max_seq_length = min(data_args.max_seq_length, tokenizer.model_max_length)

    if data_args.line_by_line:
        # When using line_by_line, we just tokenize each nonempty line.
        padding = "max_length" if data_args.pad_to_max_length else False

        def tokenize_function(examples):
            # Remove empty lines
            examples["text"] = [line for line in examples["text"] if len(line) > 0 and not line.isspace()]
            return tokenizer(examples["text"], padding=padding, truncation=True, max_length=max_seq_length)

        with training_args.main_process_first(desc="dataset map tokenization"):
            tokenized_datasets = raw_datasets.map(
                tokenize_function,
                batched=True,
                num_proc=data_args.preprocessing_num_workers,
                remove_columns=[text_column_name],
                load_from_cache_file=not data_args.overwrite_cache,
                desc="Running tokenizer on dataset line_by_line",
            )
    else:
        # Otherwise, we tokenize every text, then concatenate them together before splitting them in smaller parts.
        def tokenize_function(examples):
            return tokenizer(examples[text_column_name])

        with training_args.main_process_first(desc="dataset map tokenization"):
            tokenized_datasets = raw_datasets.map(
                tokenize_function,
                batched=True,
                num_proc=data_args.preprocessing_num_workers,
                remove_columns=column_names,
                load_from_cache_file=not data_args.overwrite_cache,
                desc="Running tokenizer on every text in dataset",
            )

        # Main data processing function that will concatenate all texts from our dataset and generate chunks of
        # max_seq_length.
        def group_texts(examples):
            # Concatenate all texts.
            concatenated_examples = {k: list(chain(*examples[k])) for k in examples.keys()}
            total_length = len(concatenated_examples[list(examples.keys())[0]])
            # We drop the small remainder, and if the total_length < max_seq_length  we exclude this batch and return an empty dict.
            # We could add padding if the model supported it instead of this drop, you can customize this part to your needs.
            total_length = (total_length // max_seq_length) * max_seq_length
            # Split by chunks of max_len.
            result = {
                k: [t[i : i + max_seq_length] for i in range(0, total_length, max_seq_length)]
                for k, t in concatenated_examples.items()
            }
            return result

        # Note that with `batched=True`, this map processes 1,000 texts together, so group_texts throws away a
        # remainder for each of those groups of 1,000 texts. You can adjust that batch_size here but a higher value
        # might be slower to preprocess.
        #
        # To speed up this part, we use multiprocessing. See the documentation of the map method for more information:
        # https://huggingface.co/docs/datasets/process#map

        with training_args.main_process_first(desc="grouping texts together"):
            tokenized_datasets = tokenized_datasets.map(
                group_texts,
                batched=True,
                num_proc=data_args.preprocessing_num_workers,
                load_from_cache_file=not data_args.overwrite_cache,
                desc=f"Grouping texts in chunks of {max_seq_length}",
            )

    if training_args.do_train:
        if "train" not in tokenized_datasets:
            raise ValueError("--do_train requires a train dataset")
        train_dataset = tokenized_datasets["train"]
        if data_args.max_train_samples is not None:
            max_train_samples = min(len(train_dataset), data_args.max_train_samples)
            train_dataset = train_dataset.select(range(max_train_samples))

    if training_args.do_eval:
        if "validation" not in tokenized_datasets:
            raise ValueError("--do_eval requires a validation dataset")
        eval_dataset = tokenized_datasets["validation"]
        if data_args.max_eval_samples is not None:
            max_eval_samples = min(len(eval_dataset), data_args.max_eval_samples)
            eval_dataset = eval_dataset.select(range(max_eval_samples))

    # Data collator
    data_collator = DataCollatorForPermutationLanguageModeling(
        tokenizer=tokenizer,
        plm_probability=data_args.plm_probability,
        max_span_length=data_args.max_span_length,
    )

    # Initialize our Trainer
    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=train_dataset if training_args.do_train else None,
        eval_dataset=eval_dataset if training_args.do_eval else None,
        tokenizer=tokenizer,
        data_collator=data_collator,
    )

    # Training
    if training_args.do_train:
        checkpoint = None
        if training_args.resume_from_checkpoint is not None:
            checkpoint = training_args.resume_from_checkpoint
        elif last_checkpoint is not None:
            checkpoint = last_checkpoint
        train_result = trainer.train(resume_from_checkpoint=checkpoint)
        trainer.save_model()  # Saves the tokenizer too for easy upload
        metrics = train_result.metrics

        max_train_samples = (
            data_args.max_train_samples if data_args.max_train_samples is not None else len(train_dataset)
        )
        metrics["train_samples"] = min(max_train_samples, len(train_dataset))

        trainer.log_metrics("train", metrics)
        trainer.save_metrics("train", metrics)
        trainer.save_state()

    # Evaluation
    if training_args.do_eval:
        logger.info("*** Evaluate ***")

        metrics = trainer.evaluate()

        max_eval_samples = data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset)
        metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset))
        try:
            perplexity = math.exp(metrics["eval_loss"])
        except OverflowError:
            perplexity = float("inf")
        metrics["perplexity"] = perplexity

        trainer.log_metrics("eval", metrics)
        trainer.save_metrics("eval", metrics)

    kwargs = {"finetuned_from": model_args.model_name_or_path, "tasks": "language-modeling"}
    if data_args.dataset_name is not None:
        kwargs["dataset_tags"] = data_args.dataset_name
        if data_args.dataset_config_name is not None:
            kwargs["dataset_args"] = data_args.dataset_config_name
            kwargs["dataset"] = f"{data_args.dataset_name} {data_args.dataset_config_name}"
        else:
            kwargs["dataset"] = data_args.dataset_name

    if training_args.push_to_hub:
        trainer.push_to_hub(**kwargs)
    else:
        trainer.create_model_card(**kwargs)