# Copyright (c) Meta Platforms, Inc. and affiliates. # All rights reserved. # # This source code is licensed under the BSD-style license found in the # LICENSE file in the root directory of this source tree. import sys import time from functools import partial from typing import Any, Dict, List, Optional, Tuple, Union from warnings import warn import torch from omegaconf import DictConfig, ListConfig from torch import nn from torch.distributed import destroy_process_group, init_process_group from torch.optim import Optimizer from torch.utils.data import DataLoader, DistributedSampler from torchtune import config, modules, training, utils from torchtune.data import padded_collate_packed, padded_collate_sft from torchtune.datasets import ConcatDataset from torchtune.modules.peft import ( DoRALinear, get_adapter_params, get_adapter_state_dict, get_lora_module_names, get_merged_lora_ckpt, LoRALinear, set_trainable_params, validate_missing_and_unexpected_for_lora, ) from torchtune.recipe_interfaces import FTRecipeInterface from torchtune.training import DummyProfiler, PROFILER_KEY from tqdm import tqdm log = utils.get_logger("DEBUG") class KDRecipeDistributed(FTRecipeInterface): """ Knowledge distillation recipe for dense transformer-based LLMs such as Llama3. This recipe is optimized for single GPU training. Training on CPU is not supported. Features: - FSDP. Supported using PyTorch's FSDP APIs. CPU offload of parameters, gradients, and optimizer states is supported via ``fsdp_cpu_offload``. Resharding of parameters after the forward pass is done by default (corresponding to FULL_SHARD sharding strategy), but can be disabled by setting the config ``fsdp_reshard_after_forward`` to False (this corresponds to SHARD_GRAD_OP sharding strategy). DDP is currently not supported. Training on CPU is not supported. - Activation Checkpointing. This can be controlled using the ``activation_checkpointing`` flag. Activation checkpointing helps reduce the memory footprint since we no longer keep activations in memory and instead recompute them during the backward pass. This is especially helpful for larger batch sizes when you're memory constrained. But these savings in memory come at the cost of training performance. In most cases training can slow-down quite a bit as a result of this activation recomputation. - Precision. Full fp32 and bf16 training are supported. Precision is controlled using the ``dtype`` flag. When ``dtype=bf16``, all activations, gradients and optimizer states are in bfloat16. In most cases this should halve the memory footprint of full precision (fp32) training, without loss in model quality (will depend on the model, training data and other settings). For GPUs which do not support bfloat16, we fall back to fp32. Mixed precision training and fp16 precision are currently not supported. - Gradient Accumulation. You can simulate larger batch sizes by accumulating gradients. This is controlled using the ``gradient_accumulation_steps`` flag. Total Batch Size = batch_size * gradient accumulation steps. For example: with batch_size=1 and gradient_accumulation_steps=32 we get a total batch size of 32. Gradient accumulation is especially useful when you are memory constrained. In this case, accumulating gradients might give you better training speed than enabling activation checkpointing. - Checkpointing. Model weights are checkpointed both at the end of each epoch and at the end of training. Currently we checkpoint both the adapter weights (trainable params only) and the complete merged weights (adapter weights added back to the base model). For more details please take a look at our LoRA tutorial (https://pytorch.org/torchtune/main/tutorials/lora_finetune.html). Optimizer State and recipe state (seed, total_epochs, number of epochs run etc) are only saved at the end of a given epoch and used in case of resuming training. Resuming training is controlled by the ``resume_from_checkpoint`` flag. Mid-epoch checkpointing is currently not supported. For more details on the checkpointer, please take a look at our checkpointer deepdive (https://pytorch.org/torchtune/main/tutorials/checkpointer.html). - Logging. Terminal, Disk, WandB and TensorBoard are all supported. For a full list of example configs for this recipe, run ``tune ls`` on the command line. Each config has example commands for how to kick-off training. Args: cfg (DictConfig): OmegaConf object parsed from yaml file Raises: ValueError: If ``dtype`` is set to fp16. RuntimeError: If ``dtype`` is set to bf16 and the hardware does not support bf16. """ def __init__(self, cfg: DictConfig) -> None: self._device = utils.get_device(device=cfg.device) # Reduced precision logic self._dtype = training.get_dtype(cfg.dtype, device=self._device) # fp16 precision is explicitly disabled as it is not supported in this # recipe (for example, no gradient scaling). if self._dtype == torch.float16: raise ValueError( "fp16 precision is not supported in this recipe. Please use fp32 or bf16." ) self.world_size, self.rank = utils.get_world_size_and_rank() self._is_rank_zero = self.rank == 0 # logging attributes self._output_dir = cfg.output_dir self._log_every_n_steps = cfg.get("log_every_n_steps", 1) self._log_peak_memory_stats = cfg.get("log_peak_memory_stats", False) # training attributes self._enable_activation_checkpointing = cfg.enable_activation_checkpointing # These are public properties which are updated by the checkpoint loader # when ``resume_from_checkpoint`` is `True` or validated in tests self.seed = training.set_seed( seed=cfg.seed, debug_mode=cfg.get("cudnn_deterministic_mode", None) ) self.epochs_run = 0 self.total_epochs = cfg.epochs self.max_steps_per_epoch = cfg.max_steps_per_epoch self.global_step = 0 self._resume_from_checkpoint = cfg.resume_from_checkpoint self._save_adapter_weights_only = cfg.get("save_adapter_weights_only", False) self._gradient_accumulation_steps = cfg.gradient_accumulation_steps self._kd_ratio = cfg.get("kd_ratio", 0.5) def load_checkpoint(self, cfg_checkpointer: DictConfig) -> Dict[str, Any]: """ Extract the checkpoint state from file and validate. This includes the base model weights. If resume_from_checkpoint is True, this also includes the adapter weights and recipe state """ self._checkpointer = config.instantiate( cfg_checkpointer, should_load_recipe_state=self._resume_from_checkpoint, ) checkpoint_dict = self._checkpointer.load_checkpoint() if self._resume_from_checkpoint: if training.ADAPTER_KEY not in checkpoint_dict: raise ValueError( "Adapter weights not found. Please ensure a valid adapter checkpoint is provided." ) # _update_recipe_state will throw an exception if the recipe state is not corrctly loaded # no need to check here self._update_recipe_state(checkpoint_dict) return checkpoint_dict def load_teacher_checkpoint(self, cfg_checkpointer: DictConfig) -> Dict[str, Any]: """ Extract the teacher checkpoint state from file. """ teacher_checkpointer = config.instantiate( cfg_checkpointer, ) checkpoint_dict = teacher_checkpointer.load_checkpoint() return checkpoint_dict def _update_recipe_state(self, ckpt_dict: Dict[str, Any]) -> None: """ Updates the recipe state from checkpoint. """ try: self.epochs_run = ckpt_dict[training.EPOCHS_KEY] # on mismatch, warn the user and prevent the override if self.seed != ckpt_dict[training.SEED_KEY]: warn( message=( "Config value for seed does not match the checkpoint value, " f"using the checkpoint value: {ckpt_dict[training.SEED_KEY]}" ) ) self.seed = ckpt_dict[training.SEED_KEY] if self.max_steps_per_epoch != ckpt_dict[training.MAX_STEPS_KEY]: warn( message=( "Config value for max_steps_per_epoch does not match the checkpoint value, " f"using the checkpoint value: {ckpt_dict[training.MAX_STEPS_KEY]}" ) ) self.max_steps_per_epoch = ckpt_dict[training.MAX_STEPS_KEY] # on mismatch, warn the user but allow the override if self.total_epochs != ckpt_dict[training.TOTAL_EPOCHS_KEY]: warn( message=( "Config value for total_epochs does not match the checkpoint value, " f"using the config value: {self.total_epochs}" ) ) except KeyError as e: raise KeyError( "Checkpoint does not contain the required keys needed for updating recipe state. " "Are you sure you passed in the right recipe checkpoint?" ) from e def setup(self, cfg: DictConfig) -> None: """ Setup the recipe state. This includes recipe state (if resume_from_checkpoint is True), model, tokenizer, loss, optimizer, learning rate scheduler, sampler, and dataloader. """ if self._is_rank_zero: self._metric_logger = config.instantiate(cfg.metric_logger) # log config with parameter override self._metric_logger.log_config(cfg) self._compile = cfg.get("compile", False) checkpoint_dict = self.load_checkpoint(cfg_checkpointer=cfg.checkpointer) teacher_checkpoint_dict = self.load_teacher_checkpoint( cfg_checkpointer=cfg.teacher_checkpointer ) # set up model self._model = self._setup_model( cfg_model=cfg.model, enable_activation_checkpointing=cfg.enable_activation_checkpointing, fsdp_cpu_offload=cfg.get("fsdp_cpu_offload", False), reshard_after_forward=cfg.get("fsdp_reshard_after_forward", True), base_model_state_dict=checkpoint_dict[training.MODEL_KEY], lora_weights_state_dict=( checkpoint_dict[training.ADAPTER_KEY] if self._resume_from_checkpoint else None ), ) self._teacher_model = self._setup_teacher_model( model_cfg=cfg.teacher_model, custom_sharded_layers=cfg.get("custom_sharded_layers", None), fsdp_cpu_offload=cfg.get("fsdp_cpu_offload", False), reshard_after_forward=cfg.get("fsdp_reshard_after_forward", True), model_state_dict=teacher_checkpoint_dict[training.MODEL_KEY], ) self._tokenizer = config.instantiate(cfg.tokenizer) utils.log_rank_zero(log, "Tokenizer is initialized from file.") self._optimizer = self._setup_optimizer( cfg_optimizer=cfg.optimizer, opt_state_dict=( checkpoint_dict[training.OPT_KEY] if self._resume_from_checkpoint else None ), ) # initialize loss self._loss_fn = config.instantiate(cfg.loss) self._kd_loss_fn = config.instantiate(cfg.kd_loss) if self._compile: self._loss_fn = training.compile_loss( self._loss_fn, verbose=self._is_rank_zero ) self._kd_loss_fn = training.compile_loss( self._kd_loss_fn, verbose=self._is_rank_zero ) if self._loss_fn.__class__.__name__ == "CEWithChunkedOutputLoss": # set num_output_chunks for model self._model.set_num_output_chunks(self._loss_fn.num_output_chunks) self._teacher_model.set_num_output_chunks(self._loss_fn.num_output_chunks) # assert _loss_fn and _kd_loss_fn have the same num_output_chunks assert ( self._loss_fn.num_output_chunks == self._kd_loss_fn.num_output_chunks ), "Number of output chunks for loss_fn and kd_loss_fn must be the same." utils.log_rank_zero(log, "Loss is initialized.") # Dataloader depends on the tokenizer and loss_fn and should be # setup after all of these are setup self._sampler, self._dataloader = self._setup_data( cfg_dataset=cfg.dataset, shuffle=cfg.shuffle, batch_size=cfg.batch_size, ) # Finally update the recipe state which can only be correctly set after all of the # other components have been initialized and updated. # Number of training steps in each epoch depends on the number of batches produced # by the dataloader and the max_steps_per_epoch param set by the user and is used # for logging and tracking training state. This should be computed after the dataloader # has been setup self._steps_per_epoch = ( len(self._dataloader) // self._gradient_accumulation_steps ) if ( self.max_steps_per_epoch is not None and self.max_steps_per_epoch < self._steps_per_epoch ): self._steps_per_epoch = self.max_steps_per_epoch self.global_step = self.epochs_run * self._steps_per_epoch # Learning rate scheduler can only be set up after number of steps # has been computed self._lr_scheduler = self._setup_lr_scheduler( cfg_lr_scheduler=cfg.lr_scheduler, num_training_steps=self.total_epochs * self._steps_per_epoch, last_epoch=self.global_step - 1, ) # Set up profiler, returns DummyProfiler (nullcontext object with no-op `step` method) # if cfg is missing profiler key or if `cfg.profiler.enabled = False` self._profiler = self._setup_profiler(cfg.get(PROFILER_KEY, None)) # Used to ignore labels for loss computation self.ignore_labels_cache = torch.full( (cfg.batch_size, 1), self._loss_fn.ignore_index, device=self._device ) def _setup_profiler( self, cfg_profiler: Optional[DictConfig] = None ) -> Union[torch.profiler.profile, DummyProfiler]: """ Parses the `profiler` section of top-level `cfg` and sets up profiler Args: cfg_profiler (Optional[DictConfig]): ``profiler`` section of the top-level ``cfg`` (the main config passed to `recipe.main`). Default None. Returns: profiler: Union[torch.profiler.profile, DummyProfiler] - DummyProfiler is a nullcontext with no-op methods for `start`, `stop`, and `step` that can be used in place of `torch.profiler.profile` if profiler is not enabled such that the instrumented training loop does not need to be changed profiling is disabled. The profiler config can be provided in configs under the `profiler` key with the following layout: .. code-block:: yaml profiler: enabled: bool #Output directory of trace artifacts output_dir: str #`torch.profiler.ProfilerActivity` types to trace cpu: bool cuda: bool #Trace options profile_memory: bool with_stack: bool record_shapes: bool with_flops: bool # `torch.profiler.schedule` options: # wait_steps -> wait, warmup_steps -> warmup, active_steps -> active, num_cycles -> repeat wait_steps: int warmup_steps: int active_steps: int num_cycles: int """ # Missing profiler section in config, assume disabled if cfg_profiler is None: cfg_profiler = DictConfig({"enabled": False}) # Check that component is included and set correctly if cfg_profiler.get("_component_", None) is None: cfg_profiler["_component_"] = "torchtune.training.setup_torch_profiler" else: assert ( cfg_profiler.get("_component_") == "torchtune.training.setup_torch_profiler" ), "Only torch profiler supported currently: component must be `torchtune.training.setup_torch_profiler`" profiler, profiler_cfg = config.instantiate(cfg_profiler) utils.log_rank_zero( log, f" Profiler config after instantiation: {profiler_cfg}" ) if self._is_rank_zero: self.profiler_profile_memory = profiler_cfg.get("profile_memory", False) if profiler_cfg["enabled"]: self.profiler_wait_steps = profiler_cfg["wait_steps"] self.profiler_warmup_steps = profiler_cfg["warmup_steps"] self.profiler_active_steps = profiler_cfg["active_steps"] return profiler def _setup_model( self, cfg_model: DictConfig, enable_activation_checkpointing: bool, fsdp_cpu_offload: bool, reshard_after_forward: bool, base_model_state_dict: Dict[str, Any], custom_sharded_layers: Optional[List[str]] = None, lora_weights_state_dict: Optional[Dict[str, Any]] = None, ) -> nn.Module: """ Model initialization has some important considerations: a. To minimize GPU peak memory, we initialize the model on meta device with the right dtype b. All ranks calls ``load_state_dict`` without peaking CPU RAMs since full state dicts are loaded with ``torch.load(mmap=True)`` c. We register (pre-)forward hooks with ``fully_shard`` instead of wrapping `nn.Module` """ self._lora_rank = cfg_model.lora_rank self._lora_alpha = cfg_model.lora_alpha self._lora_attn_modules = list(cfg_model.lora_attn_modules) self._apply_lora_to_mlp = cfg_model.apply_lora_to_mlp self._apply_lora_to_output = getattr(cfg_model, "apply_lora_to_output", False) utils.log_rank_zero( log, "FSDP is enabled. Instantiating model and loading checkpoint on Rank 0 ...", ) init_start = time.perf_counter() with training.set_default_dtype(self._dtype), torch.device("meta"): model = config.instantiate(cfg_model) self.adapter_params = get_adapter_params(model) set_trainable_params(model, self.adapter_params) if self._compile: training.compile_model(model, verbose=self._is_rank_zero) if enable_activation_checkpointing: training.set_activation_checkpointing( model, auto_wrap_policy={modules.TransformerSelfAttentionLayer} ) # For FSDP sharding fsdp_shard_conditions = [ partial( training.get_shard_conditions, names_to_match=custom_sharded_layers, ) ] training.shard_model( model=model, shard_conditions=fsdp_shard_conditions, cpu_offload=fsdp_cpu_offload, reshard_after_forward=reshard_after_forward, ) if lora_weights_state_dict: lora_missing, lora_unexpected = training.load_from_full_model_state_dict( model, lora_weights_state_dict, self._device, cpu_offload=fsdp_cpu_offload, ) else: lora_missing, lora_unexpected = None, None # Initializer for LoRA params and RoPE buffers with training.set_default_dtype(self._dtype), self._device: lora_device = "cpu" if fsdp_cpu_offload else self._device for m in model.modules(): if ( isinstance(m, LoRALinear) or isinstance(m, DoRALinear) ) and not lora_weights_state_dict: # lora may not be covered in state dict # if finetune for the 1st time m.to_empty(device=lora_device) m.initialize_parameters() # RoPE is not covered in state dict if hasattr(m, "rope_init"): m.rope_init() base_missing, base_unexpected = training.load_from_full_model_state_dict( model, base_model_state_dict, self._device, cpu_offload=fsdp_cpu_offload, ) for m in model.modules(): if hasattr(m, "initialize_dora_magnitude"): m.initialize_dora_magnitude() validate_missing_and_unexpected_for_lora( lora_attn_modules=self._lora_attn_modules, apply_lora_to_mlp=self._apply_lora_to_mlp, apply_lora_to_output=self._apply_lora_to_output, base_missing=base_missing, base_unexpected=base_unexpected, lora_missing=lora_missing, lora_unexpected=lora_unexpected, ) # Ensure no params and buffers are on meta device training.validate_no_params_on_meta_device(model) utils.log_rank_zero( log, f"Instantiating student model and loading checkpoint took {time.perf_counter() - init_start:.2f} secs", ) if self._is_rank_zero: memory_stats = training.get_memory_stats(device=self._device) training.log_memory_stats( memory_stats, message="Memory stats after student model init:" ) # synchronize before training begins torch.distributed.barrier() return model def _setup_teacher_model( self, model_cfg: DictConfig, custom_sharded_layers: Optional[List[str]], fsdp_cpu_offload: bool, reshard_after_forward: bool, model_state_dict: Dict[str, Any], ) -> nn.Module: """ Model initialization for teacher model has some important considerations: a. To minimize GPU peak memory, we initialize the model on meta device with the right dtype b. All ranks calls ``load_state_dict`` without peaking CPU RAMs since full state dicts are loaded with ``torch.load(mmap=True)`` """ utils.log_rank_zero( log, "FSDP enabled. Instantiating teacher model and loading checkpoint on Rank 0 ...", ) init_start = time.perf_counter() with training.set_default_dtype(self._dtype), torch.device("meta"): model = config.instantiate(model_cfg) if self._compile: training.compile_model(model, verbose=self._is_rank_zero) # For FSDP sharding fsdp_shard_conditions = [ partial( training.get_shard_conditions, names_to_match=custom_sharded_layers, ) ] training.shard_model( model=model, shard_conditions=fsdp_shard_conditions, cpu_offload=fsdp_cpu_offload, reshard_after_forward=reshard_after_forward, ) with training.set_default_dtype(self._dtype), self._device: for m in model.modules(): # RoPE is not covered in state dict if hasattr(m, "rope_init"): m.rope_init() # This method will convert the full model state dict into a sharded state # dict and load into the model training.load_from_full_model_state_dict( model, model_state_dict, self._device, strict=True, cpu_offload=fsdp_cpu_offload, ) # Put model in eval mode. # Note: This will not disable the dropout applied in SDPA, # see https://github.com/pytorch/pytorch/issues/124464 model.eval() for p in model.parameters(): p.requires_grad = False # Ensure no params and buffers are on meta device training.validate_no_params_on_meta_device(model) utils.log_rank_zero( log, f"Instantiating teacher model and loading checkpoint took {time.perf_counter() - init_start:.2f} secs", ) if self._is_rank_zero: memory_stats = training.get_memory_stats(device=self._device) training.log_memory_stats( memory_stats, message="Memory stats after teacher model init:" ) # synchronize before training begins torch.distributed.barrier() return model def _setup_optimizer( self, cfg_optimizer: DictConfig, opt_state_dict: Optional[Dict[str, Any]] = None ) -> Optimizer: optimizer = config.instantiate(cfg_optimizer, self._model.parameters()) if opt_state_dict: training.load_from_full_optimizer_state_dict( self._model, optimizer, opt_state_dict, self._device, ) utils.log_rank_zero(log, "Optimizer is initialized.") return optimizer def _setup_lr_scheduler( self, cfg_lr_scheduler: DictConfig, num_training_steps: int, last_epoch: int, ) -> Optimizer: lr_scheduler = config.instantiate( cfg_lr_scheduler, self._optimizer, num_training_steps=num_training_steps, last_epoch=last_epoch, ) utils.log_rank_zero(log, "Learning rate scheduler is initialized.") return lr_scheduler def _setup_data( self, cfg_dataset: DictConfig, shuffle: bool, batch_size: int, ) -> Tuple[DistributedSampler, DataLoader]: """ All data related setup happens here. Currently this recipe only supports Map-style Datasets which fit into memory and an option for random shuffling. Samplers, iterable datasets, and streaming datasets are not supported. """ if isinstance(cfg_dataset, ListConfig): datasets = [ config.instantiate(single_cfg_dataset, self._tokenizer) for single_cfg_dataset in cfg_dataset ] ds = ConcatDataset(datasets=datasets) packed = getattr(ds, "packed", False) else: ds = config.instantiate(cfg_dataset, self._tokenizer) packed = cfg_dataset.get("packed", False) sampler = DistributedSampler( ds, num_replicas=self.world_size, rank=self.rank, shuffle=shuffle, seed=0, ) dataloader = DataLoader( dataset=ds, batch_size=batch_size, sampler=sampler, collate_fn=( partial( padded_collate_sft, padding_idx=self._tokenizer.pad_id, ignore_idx=self._loss_fn.ignore_index, ) if not packed else partial( padded_collate_packed, ) ), ) utils.log_rank_zero(log, "Dataset and Sampler are initialized.") return sampler, dataloader def save_checkpoint(self, epoch: int) -> None: """ Checkpoint the state of the recipe. The constructed checkpoint state dict contains the following information: - Merged weights with key MODEL_KEY - Adapter weights with key ADAPTER_KEY - Relevant recipe state if training is not complete - If the `self._save_adapter_weights_only` option is True, the checkpointer will save only the adapter weights To correctly resume from training, the adapter weights and recipe state must be provided along with the base model weights. """ # final dict passed onto the checkpointer checkpoint_dict = {} intermediate_checkpoint = epoch + 1 < self.total_epochs # To prevent GPU memory from spiking during checkpoint save, # we consolidate the full model and optim state dicts on CPU for rank 0 cpu_state_dict = training.gather_cpu_state_dict( self._model, self._is_rank_zero, device=self._device, ) if intermediate_checkpoint: opt_state_dict = training.get_full_optimizer_state_dict( self._model, self._optimizer, self._is_rank_zero, device=self._device, ) else: opt_state_dict = None # Now that we have the model and opt state dict, create the actual checkpoint # to be sent to the checkpointer and ultimately written to file if self._is_rank_zero: # Filter out the adapter keys and weights from the model state dict. These will # be saved separately adapter_state_dict = get_adapter_state_dict(cpu_state_dict) checkpoint_dict.update({training.ADAPTER_KEY: adapter_state_dict}) # merge the adapter weights and base weights to create the model checkpoint merged_state_dict = get_merged_lora_ckpt( cpu_state_dict, rank=self._lora_rank, alpha=self._lora_alpha, ) checkpoint_dict.update({training.MODEL_KEY: merged_state_dict}) # if training is in-progress, checkpoint the optimizer state and recipe state # as well if intermediate_checkpoint: checkpoint_dict.update( { training.OPT_KEY: opt_state_dict, training.SEED_KEY: self.seed, training.EPOCHS_KEY: self.epochs_run, training.TOTAL_EPOCHS_KEY: self.total_epochs, training.MAX_STEPS_KEY: self.max_steps_per_epoch, } ) adapter_config = { "r": self._lora_rank, "lora_alpha": self._lora_alpha, "target_modules": get_lora_module_names( self._lora_attn_modules, self._apply_lora_to_mlp, self._apply_lora_to_output, ), "peft_type": "LORA", } checkpoint_dict.update({training.ADAPTER_CONFIG: adapter_config}) self._checkpointer.save_checkpoint( checkpoint_dict, epoch=epoch, intermediate_checkpoint=intermediate_checkpoint, adapter_only=self._save_adapter_weights_only, ) def _loss_step( self, batch: Dict[str, torch.Tensor] ) -> (torch.Tensor, torch.Tensor): # Both are shape [b, s] tokens, labels = batch["tokens"], batch["labels"] # Get the attention mask and position ids from the dataset if they # exist. Currently, only sample packing in PackedDataset returns these mask = batch.get("mask", None) # shape [b, s, s] input_pos = batch.get("input_pos", None) # shape [b, s] # run model logits = self._model(tokens, mask=mask, input_pos=input_pos) # Shift labels to compute loss # equivalent to doing labels[..., 1:] and logits[..., :-1, :] # But this way we dont need to slice the logits. We just add an ignore index to labels. labels = torch.hstack( (labels[..., 1:], self.ignore_labels_cache[: labels.shape[0]]) ) if not isinstance(logits, list): labels = labels.reshape(-1) logits = logits.reshape(-1, logits.size(-1)) # Compute KD loss with torch.no_grad(): teacher_logits = self._teacher_model(tokens, mask=mask, input_pos=input_pos) # Compute kd loss kd_loss = self._kd_loss_fn(logits, teacher_logits, labels) # Compute loss loss = self._loss_fn(logits, labels) # free logits otherwise it peaks backward memory del logits del teacher_logits return loss, kd_loss def train(self) -> None: """ The core training loop. """ # clean up before training begins training.cleanup_before_training() # zero out the gradients before starting training self._optimizer.zero_grad() # Initialize tokens count and running loss (for grad accumulation) t0 = time.perf_counter() running_class_loss = 0 running_kd_loss = 0 num_tokens = 0 self._profiler.start() # self.epochs_run should be non-zero when we're resuming from a checkpoint for curr_epoch in range(self.epochs_run, self.total_epochs): # Update the sampler to ensure data is correctly shuffled across epochs # in case shuffle is True self._sampler.set_epoch(curr_epoch) pbar = tqdm(total=self._steps_per_epoch, disable=not (self.rank == 0)) for idx, batch in enumerate(self._dataloader): if ( self.max_steps_per_epoch is not None and (idx // self._gradient_accumulation_steps) == self.max_steps_per_epoch ): break # Start tracking CUDA memory for active steps for just the first epoch if ( self._is_rank_zero and curr_epoch == 0 and self.profiler_profile_memory and idx == self.profiler_wait_steps + self.profiler_warmup_steps and self._device.type == "cuda" ): torch.cuda.memory._record_memory_history() utils.batch_to_device(batch, self._device) # Calculate the number of unmasked tokens in the current batch # and increment the total number of tokens seen in the step current_num_tokens = ( batch["labels"] != self._loss_fn.ignore_index ).sum() num_tokens += current_num_tokens class_loss, kd_loss = self._loss_step(batch) running_class_loss += class_loss * current_num_tokens running_kd_loss += kd_loss * current_num_tokens current_loss = ( 1 - self._kd_ratio ) * class_loss + self._kd_ratio * kd_loss current_loss.backward() # Step with optimizer if (idx + 1) % self._gradient_accumulation_steps == 0: # Get total number of tokens across all ranks to normalize gradients torch.distributed.all_reduce(num_tokens) # This will ensure that the logged loss matches what we're optimizing torch.distributed.all_reduce(running_class_loss) torch.distributed.all_reduce(running_kd_loss) # Manually scale the gradients from unnormalized loss by total # of tokens # We multiply by world_size to undo FSDP2 gradient normalization. training.scale_grads(self._model, self.world_size / num_tokens) class_loss_to_log = running_class_loss.item() / num_tokens kd_loss_to_log = running_kd_loss.item() / num_tokens self._optimizer.step() self._optimizer.zero_grad(set_to_none=True) self._lr_scheduler.step() # Update the number of steps when the weights are updated self.global_step += 1 class_loss_to_log = class_loss.item() kd_loss_to_log = kd_loss.item() loss_to_log = ( 1 - self._kd_ratio ) * class_loss_to_log + self._kd_ratio * kd_loss_to_log pbar.update(1) pbar.set_description( f"{curr_epoch + 1}|{self.global_step}|Loss: {loss_to_log}" ) # Log per-step metrics if ( self.global_step % self._log_every_n_steps == 0 and self._is_rank_zero ): time_per_step = time.perf_counter() - t0 log_dict = { "loss": loss_to_log, "class_loss": class_loss_to_log, "kd_loss": kd_loss_to_log, "lr": self._optimizer.param_groups[0]["lr"], "tokens_per_second_per_gpu": num_tokens / (time_per_step * self.world_size), } if self._log_peak_memory_stats: log_dict.update( training.get_memory_stats(device=self._device) ) self._metric_logger.log_dict( log_dict, step=self.global_step, ) # Reset running stats for the next step running_class_loss = 0 running_kd_loss = 0 num_tokens = 0 t0 = time.perf_counter() # Stop tracking CUDA memory now that active steps are complete if ( self._is_rank_zero and curr_epoch == 0 and self.profiler_profile_memory and idx == self.profiler_wait_steps + self.profiler_warmup_steps + self.profiler_active_steps ): torch.cuda.memory._record_memory_history(enabled=None) # Step the profiler # Note we are stepping each batch, which might not include optimizer step in the trace # if the schedule cycle doesn't align with gradient accumulation. self._profiler.step() self.epochs_run += 1 self.save_checkpoint(epoch=curr_epoch) self._profiler.stop() def cleanup(self) -> None: if self._is_rank_zero: self._metric_logger.close() destroy_process_group() @config.parse def recipe_main(cfg: DictConfig) -> None: """ Entry point for the recipe. Configurable parameters are read in the following order: - Parameters specified in config (see available configs through ``tune ls``) - Overwritten by arguments from the command-line """ if not training.is_distributed(): raise RuntimeError( "Distributed finetune recipe should be run via a distributed launcher." "If using tune CLI, please specify --nnodes 1 and --nproc_per_node [num_gpus]" ) init_process_group("cuda:nccl,cpu:gloo") if cfg.get("fsdp_cpu_offload", False): # Utilize all available CPU cores for intra-op parallelism. This provides ~2x # speed up when benchmarking fused AdamW on CPU training.set_torch_num_threads() config.log_config(recipe_name="KDRecipeDistributed", cfg=cfg) recipe = KDRecipeDistributed(cfg=cfg) recipe.setup(cfg=cfg) recipe.train() recipe.cleanup() if __name__ == "__main__": sys.exit(recipe_main())