# Code based on https://github.com/tloen/alpaca-lora import os import sys from typing import List import fire import torch import transformers from datasets import load_dataset """ Unused imports: import torch.nn as nn import bitsandbytes as bnb """ from peft import ( LoraConfig, get_peft_model, get_peft_model_state_dict, prepare_model_for_kbit_training, set_peft_model_state_dict, ) from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig from transformers.trainer_utils import PREFIX_CHECKPOINT_DIR from transformers.trainer_callback import TrainerCallback from utils.prompter import Prompter class SavePeftModelCallback(TrainerCallback): def on_save(self, args, state, control, **kwargs): checkpoint_folder = os.path.join( args.output_dir, f"{PREFIX_CHECKPOINT_DIR}-{state.global_step}" ) peft_model_path = os.path.join(checkpoint_folder, "adapter_model") kwargs["model"].save_pretrained(peft_model_path) pytorch_model_path = os.path.join(checkpoint_folder, "pytorch_model.bin") if os.path.exists(pytorch_model_path): os.remove(pytorch_model_path) return control def train( # model/data params base_model: str = "", # the only required argument data_path: str = "yahma/alpaca-cleaned", output_dir: str = "./lora-alpaca", trust_remote_code: bool = False, load_in_8bit: bool = False, load_in_4bit: bool = False, # If 8 bit is also specified, 4 bit has priority pad_token_id: int = 0, # training hyperparams batch_size: int = 128, micro_batch_size: int = 4, num_epochs: int = 3, learning_rate: float = 3e-4, cutoff_len: int = 256, val_set_size: int = 2000, # lora hyperparams lora_r: int = 8, lora_alpha: int = 16, lora_dropout: float = 0.05, lora_target_modules: List[str] = [ "q_proj", "v_proj", ], use_gradient_checkpointing=True, # llm hyperparams train_on_inputs: bool = True, # if False, masks out inputs in loss add_eos_token: bool = False, group_by_length: bool = False, # faster, but produces an odd training loss curve # wandb params wandb_project: str = "", wandb_run_name: str = "", wandb_watch: str = "", # options: false | gradients | all wandb_log_model: str = "", # options: false | true resume_from_checkpoint: str = None, # either training checkpoint or final adapter prompt_template_name: str = "alpaca", # The prompt template to use, will default to alpaca. ): if int(os.environ.get("LOCAL_RANK", 0)) == 0: print( f"Training Alpaca-LoRA model with params:\n" f"base_model: {base_model}\n" f"data_path: {data_path}\n" f"output_dir: {output_dir}\n" f"batch_size: {batch_size}\n" f"micro_batch_size: {micro_batch_size}\n" f"num_epochs: {num_epochs}\n" f"learning_rate: {learning_rate}\n" f"cutoff_len: {cutoff_len}\n" f"val_set_size: {val_set_size}\n" f"lora_r: {lora_r}\n" f"lora_alpha: {lora_alpha}\n" f"lora_dropout: {lora_dropout}\n" f"lora_target_modules: {lora_target_modules}\n" f"train_on_inputs: {train_on_inputs}\n" f"add_eos_token: {add_eos_token}\n" f"group_by_length: {group_by_length}\n" f"wandb_project: {wandb_project}\n" f"wandb_run_name: {wandb_run_name}\n" f"wandb_watch: {wandb_watch}\n" f"wandb_log_model: {wandb_log_model}\n" f"resume_from_checkpoint: {resume_from_checkpoint or False}\n" f"prompt template: {prompt_template_name}\n" ) assert ( base_model ), "Please specify a --base_model, e.g. --base_model='decapoda-research/llama-7b-hf'" gradient_accumulation_steps = batch_size // micro_batch_size prompter = Prompter(prompt_template_name) device_map = "auto" world_size = int(os.environ.get("WORLD_SIZE", 1)) print(world_size) ddp = world_size != 1 if ddp: device_map = {"": int(os.environ.get("LOCAL_RANK") or 0)} gradient_accumulation_steps = gradient_accumulation_steps // world_size # Check if parameter passed or if set within environ use_wandb = len(wandb_project) > 0 or ( "WANDB_PROJECT" in os.environ and len(os.environ["WANDB_PROJECT"]) > 0 ) # Only overwrite environ if wandb param passed if len(wandb_project) > 0: os.environ["WANDB_PROJECT"] = wandb_project if len(wandb_watch) > 0: os.environ["WANDB_WATCH"] = wandb_watch if len(wandb_log_model) > 0: os.environ["WANDB_LOG_MODEL"] = wandb_log_model if load_in_4bit: nf4_config = BitsAndBytesConfig( load_in_4bit=True, bnb_4bit_quant_type="nf4", bnb_4bit_use_double_quant=True, bnb_4bit_compute_dtype=torch.bfloat16 ) model = AutoModelForCausalLM.from_pretrained( base_model, quantization_config=nf4_config, device_map=device_map, trust_remote_code=trust_remote_code ) else: model = AutoModelForCausalLM.from_pretrained( base_model, load_in_8bit=load_in_8bit, torch_dtype=torch.float16, device_map=device_map, trust_remote_code=trust_remote_code ) model.model_parallel = False # For MPT patch compatibility tokenizer = AutoTokenizer.from_pretrained(base_model) print(tokenizer.special_tokens_map) print("bos_token :", tokenizer.eos_token, ",", tokenizer.bos_token_id) print("eos_token :", tokenizer.bos_token, ",", tokenizer.eos_token_id) print("unk_token :", tokenizer.unk_token, ",", tokenizer.unk_token_id) print("pad_token :", tokenizer.pad_token, ",", tokenizer.pad_token_id) tokenizer.pad_token_id = pad_token_id # we want this to be different from the eos token tokenizer.padding_side = "left" # Allow batched inference print("pad_token changed to:", tokenizer.pad_token, ",", tokenizer.pad_token_id) assert ( tokenizer.pad_token_id != tokenizer.eos_token_id ), "Please set pad_token_id which is different from eos_token_id " def tokenize(prompt, add_eos_token=True): # there's probably a way to do this with the tokenizer settings # but again, gotta move fast result = tokenizer( prompt, ) if ( result["input_ids"][-1] != tokenizer.eos_token_id and add_eos_token ): result["input_ids"].append(tokenizer.eos_token_id) result["attention_mask"].append(1) return { "input_ids": result["input_ids"], "attention_mask": result["attention_mask"], } def generate_and_tokenize_prompt(data_point): full_prompt = prompter.generate_prompt( data_point["instruction"], data_point["input"], data_point["output"], ) tokenized_full_prompt = tokenize(full_prompt) if not train_on_inputs: user_prompt = prompter.generate_prompt( data_point["instruction"], data_point["input"] ) tokenized_user_prompt = tokenize(user_prompt, add_eos_token=add_eos_token) user_prompt_len = len(tokenized_user_prompt["input_ids"]) if add_eos_token: user_prompt_len -= 1 tokenized_full_prompt["labels"] = [ -100 ] * user_prompt_len + tokenized_full_prompt["labels"][ user_prompt_len: ] # could be sped up, probably return tokenized_full_prompt if load_in_4bit or load_in_8bit: model = prepare_model_for_kbit_training( model, use_gradient_checkpointing=use_gradient_checkpointing ) config = LoraConfig( r=lora_r, lora_alpha=lora_alpha, target_modules=lora_target_modules, lora_dropout=lora_dropout, bias="none", task_type="CAUSAL_LM", ) model = get_peft_model(model, config) if data_path.endswith(".json") or data_path.endswith(".jsonl"): data = load_dataset("json", data_files=data_path) else: data = load_dataset(data_path) if resume_from_checkpoint: # Check the available weights and load them checkpoint_name = os.path.join( resume_from_checkpoint, "pytorch_model.bin" ) # Full checkpoint if not os.path.exists(checkpoint_name): checkpoint_name = os.path.join( resume_from_checkpoint, "adapter_model.bin" ) # only LoRA model - LoRA config above has to fit resume_from_checkpoint = ( False # So the trainer won't try loading its state ) # The two files above have a different name depending on how they were saved, but are actually the same. if os.path.exists(checkpoint_name): print(f"Restarting from {checkpoint_name}") adapters_weights = torch.load(checkpoint_name) model = set_peft_model_state_dict(model, adapters_weights) else: print(f"Checkpoint {checkpoint_name} not found") model.print_trainable_parameters() # Be more transparent about the % of trainable params. if val_set_size > 0: train_val = data["train"].train_test_split( test_size=val_set_size, shuffle=True, seed=42 ) train_data = ( train_val["train"].shuffle().map(generate_and_tokenize_prompt) ) val_data = ( train_val["test"].shuffle().map(generate_and_tokenize_prompt) ) else: train_data = data["train"].shuffle().map(generate_and_tokenize_prompt) val_data = None # Remove rows exceeding cutoff length print("Dataset Size Before Filter: ", train_data.shape, val_data.shape) train_data = train_data.filter(lambda x: len(x['input_ids']) < cutoff_len) val_data = val_data.filter(lambda x: len(x['input_ids']) < cutoff_len) print("Dataset Size After Filter: ", train_data.shape, val_data.shape) if not ddp and torch.cuda.device_count() > 1: # keeps Trainer from trying its own DataParallelism when more than 1 gpu is available model.is_parallelizable = True model.model_parallel = True trainer = transformers.Trainer( model=model, train_dataset=train_data, eval_dataset=val_data, args=transformers.TrainingArguments( auto_find_batch_size=True, # per_device_train_batch_size=micro_batch_size, # gradient_accumulation_steps=gradient_accumulation_steps, warmup_steps=100, num_train_epochs=num_epochs, learning_rate=learning_rate, fp16=True, logging_steps=10, optim="adamw_torch", evaluation_strategy="steps" if val_set_size > 0 else "no", save_strategy="steps", eval_steps=200 if val_set_size > 0 else None, save_steps=200, output_dir=output_dir, save_total_limit=3, load_best_model_at_end=True if val_set_size > 0 else False, ddp_find_unused_parameters=False if ddp else None, group_by_length=group_by_length, report_to="wandb" if use_wandb else None, run_name=wandb_run_name if use_wandb else None, ), data_collator=transformers.DataCollatorForLanguageModeling(tokenizer, mlm=False), callbacks=[SavePeftModelCallback] ) model.config.use_cache = False trainer.train(resume_from_checkpoint=resume_from_checkpoint) model.save_pretrained(output_dir) if __name__ == "__main__": fire.Fire(train)