ai-toolkit/toolkit/optimizers/adafactor.py

import math
from typing import List
import torch
from toolkit.optimizers.optimizer_utils import copy_stochastic, stochastic_grad_accummulation
from optimum.quanto import QBytesTensor
import random


class Adafactor(torch.optim.Optimizer):
    """
    Adafactor implementation with stochastic rounding accumulation and stochastic rounding on apply.
    Modified from transformers Adafactor implementation to support stochastic rounding accumulation and apply.

    AdaFactor pytorch implementation can be used as a drop in replacement for Adam original fairseq code:
    https://github.com/pytorch/fairseq/blob/master/fairseq/optim/adafactor.py

    Paper: *Adafactor: Adaptive Learning Rates with Sublinear Memory Cost* https://arxiv.org/abs/1804.04235 Note that
    this optimizer internally adjusts the learning rate depending on the `scale_parameter`, `relative_step` and
    `warmup_init` options. To use a manual (external) learning rate schedule you should set `scale_parameter=False` and
    `relative_step=False`.

    Arguments:
        params (`Iterable[nn.parameter.Parameter]`):
            Iterable of parameters to optimize or dictionaries defining parameter groups.
        lr (`float`, *optional*):
            The external learning rate.
        eps (`Tuple[float, float]`, *optional*, defaults to `(1e-30, 0.001)`):
            Regularization constants for square gradient and parameter scale respectively
        clip_threshold (`float`, *optional*, defaults to 1.0):
            Threshold of root mean square of final gradient update
        decay_rate (`float`, *optional*, defaults to -0.8):
            Coefficient used to compute running averages of square
        beta1 (`float`, *optional*):
            Coefficient used for computing running averages of gradient
        weight_decay (`float`, *optional*, defaults to 0.0):
            Weight decay (L2 penalty)
        scale_parameter (`bool`, *optional*, defaults to `True`):
            If True, learning rate is scaled by root mean square
        relative_step (`bool`, *optional*, defaults to `True`):
            If True, time-dependent learning rate is computed instead of external learning rate
        warmup_init (`bool`, *optional*, defaults to `False`):
            Time-dependent learning rate computation depends on whether warm-up initialization is being used

    This implementation handles low-precision (FP16, bfloat) values, but we have not thoroughly tested.

    Recommended T5 finetuning settings (https://discuss.huggingface.co/t/t5-finetuning-tips/684/3):

        - Training without LR warmup or clip_threshold is not recommended.

           - use scheduled LR warm-up to fixed LR
           - use clip_threshold=1.0 (https://arxiv.org/abs/1804.04235)
        - Disable relative updates
        - Use scale_parameter=False
        - Additional optimizer operations like gradient clipping should not be used alongside Adafactor

    Example:

    ```python
    Adafactor(model.parameters(), scale_parameter=False, relative_step=False, warmup_init=False, lr=1e-3)
    ```

    Others reported the following combination to work well:

    ```python
    Adafactor(model.parameters(), scale_parameter=True, relative_step=True, warmup_init=True, lr=None)
    ```

    When using `lr=None` with [`Trainer`] you will most likely need to use [`~optimization.AdafactorSchedule`]
    scheduler as following:

    ```python
    from transformers.optimization import Adafactor, AdafactorSchedule

    optimizer = Adafactor(model.parameters(), scale_parameter=True, relative_step=True, warmup_init=True, lr=None)
    lr_scheduler = AdafactorSchedule(optimizer)
    trainer = Trainer(..., optimizers=(optimizer, lr_scheduler))
    ```

    Usage:

    ```python
    # replace AdamW with Adafactor
    optimizer = Adafactor(
        model.parameters(),
        lr=1e-3,
        eps=(1e-30, 1e-3),
        clip_threshold=1.0,
        decay_rate=-0.8,
        beta1=None,
        weight_decay=0.0,
        relative_step=False,
        scale_parameter=False,
        warmup_init=False,
    )
    ```"""

    def __init__(
        self,
        params,
        lr=None,
        eps=(1e-30, 1e-3),
        clip_threshold=1.0,
        decay_rate=-0.8,
        beta1=None,
        weight_decay=0.0,
        scale_parameter=True,
        relative_step=True,
        warmup_init=False,
        do_paramiter_swapping=False,
        paramiter_swapping_factor=0.1,
        stochastic_accumulation=True,
    ):
        if lr is not None and relative_step:
            raise ValueError(
                "Cannot combine manual `lr` and `relative_step=True` options")
        if warmup_init and not relative_step:
            raise ValueError(
                "`warmup_init=True` requires `relative_step=True`")

        defaults = {
            "lr": lr,
            "eps": eps,
            "clip_threshold": clip_threshold,
            "decay_rate": decay_rate,
            "beta1": beta1,
            "weight_decay": weight_decay,
            "scale_parameter": scale_parameter,
            "relative_step": relative_step,
            "warmup_init": warmup_init,
        }
        super().__init__(params, defaults)
        
        self.base_lrs: List[float] = [
            lr for group in self.param_groups
        ]

        self.is_stochastic_rounding_accumulation = False

        # setup stochastic grad accum hooks
        if stochastic_accumulation:
            for group in self.param_groups:
                for param in group['params']:
                    if param.requires_grad and param.dtype != torch.float32:
                        self.is_stochastic_rounding_accumulation = True
                        param.register_post_accumulate_grad_hook(
                            stochastic_grad_accummulation
                        )
    
        self.do_paramiter_swapping = do_paramiter_swapping
        self.paramiter_swapping_factor = paramiter_swapping_factor
        self._total_paramiter_size = 0
        # count total paramiters
        for group in self.param_groups:
            for param in group['params']:
                self._total_paramiter_size += torch.numel(param)
        # pretty print total paramiters with comma seperation
        print(f"Total training paramiters: {self._total_paramiter_size:,}")
        
        # needs to be enabled to count paramiters
        if self.do_paramiter_swapping:
            self.enable_paramiter_swapping(self.paramiter_swapping_factor)
        
    
    def enable_paramiter_swapping(self, paramiter_swapping_factor=0.1):
        self.do_paramiter_swapping = True
        self.paramiter_swapping_factor = paramiter_swapping_factor
        # call it an initial time
        self.swap_paramiters()
                    
    def swap_paramiters(self):
        all_params = []
        # deactivate all paramiters
        for group in self.param_groups:
            for param in group['params']:
                param.requires_grad_(False)
                # remove any grad
                param.grad = None
                all_params.append(param)
        # shuffle all paramiters
        random.shuffle(all_params)
        
        # keep activating paramiters until we are going to go over the target paramiters
        target_paramiters = int(self._total_paramiter_size * self.paramiter_swapping_factor)
        total_paramiters = 0
        for param in all_params:
            total_paramiters += torch.numel(param)
            if total_paramiters >= target_paramiters:
                break
            else:
                param.requires_grad_(True)

    @staticmethod
    def _get_lr(param_group, param_state):
        rel_step_sz = param_group["lr"]
        if param_group["relative_step"]:
            min_step = 1e-6 * \
                param_state["step"] if param_group["warmup_init"] else 1e-2
            rel_step_sz = min(min_step, 1.0 / math.sqrt(param_state["step"]))
        param_scale = 1.0
        if param_group["scale_parameter"]:
            param_scale = max(param_group["eps"][1], param_state["RMS"])
        return param_scale * rel_step_sz

    @staticmethod
    def _get_options(param_group, param_shape):
        factored = len(param_shape) >= 2
        use_first_moment = param_group["beta1"] is not None
        return factored, use_first_moment

    @staticmethod
    def _rms(tensor):
        return tensor.norm(2) / (tensor.numel() ** 0.5)

    @staticmethod
    def _approx_sq_grad(exp_avg_sq_row, exp_avg_sq_col):
        # copy from fairseq's adafactor implementation:
        # https://github.com/huggingface/transformers/blob/8395f14de6068012787d83989c3627c3df6a252b/src/transformers/optimization.py#L505
        r_factor = (exp_avg_sq_row / exp_avg_sq_row.mean(dim=-
                    1, keepdim=True)).rsqrt_().unsqueeze(-1)
        c_factor = exp_avg_sq_col.unsqueeze(-2).rsqrt()
        return torch.mul(r_factor, c_factor)

    def step_hook(self):
        if not self.is_stochastic_rounding_accumulation:
            return
        # copy over stochastically rounded grads
        for group in self.param_groups:
            for param in group['params']:
                if param.requires_grad and hasattr(param, "_accum_grad"):
                    param.grad = param._accum_grad
                    del param._accum_grad

    # adafactor manages its own lr
    def get_learning_rates(self):
        lrs = [
            self._get_lr(group, self.state[group["params"][0]])
            for group in self.param_groups
            if group["params"][0].grad is not None
        ]
        if len(lrs) == 0:
            lrs = self.base_lrs  # if called before stepping
        return lrs

    @torch.no_grad()
    def step(self, closure=None):
        """
        Performs a single optimization step

        Arguments:
            closure (callable, optional): A closure that reevaluates the model
                and returns the loss.
        """
        self.step_hook()
        loss = None
        if closure is not None:
            loss = closure()

        for group in self.param_groups:
            for p in group["params"]:
                if p.grad is None or not p.requires_grad:
                    continue

                grad = p.grad
                if grad.dtype != torch.float32:
                    grad = grad.to(torch.float32)
                if grad.is_sparse:
                    raise RuntimeError(
                        "Adafactor does not support sparse gradients.")
                
                # if p has atts _scale then it is quantized. We need to divide the grad by the scale
                # if hasattr(p, "_scale"):
                #     grad = grad / p._scale

                state = self.state[p]
                grad_shape = grad.shape

                factored, use_first_moment = self._get_options(
                    group, grad_shape)
                # State Initialization
                if len(state) == 0:
                    state["step"] = 0

                    if use_first_moment:
                        # Exponential moving average of gradient values
                        state["exp_avg"] = torch.zeros_like(grad)
                    if factored:
                        state["exp_avg_sq_row"] = torch.zeros(
                            grad_shape[:-1]).to(grad)
                        state["exp_avg_sq_col"] = torch.zeros(
                            grad_shape[:-2] + grad_shape[-1:]).to(grad)
                    else:
                        state["exp_avg_sq"] = torch.zeros_like(grad)

                    state["RMS"] = 0
                else:
                    if use_first_moment:
                        state["exp_avg"] = state["exp_avg"].to(grad)
                    if factored:
                        state["exp_avg_sq_row"] = state["exp_avg_sq_row"].to(
                            grad)
                        state["exp_avg_sq_col"] = state["exp_avg_sq_col"].to(
                            grad)
                    else:
                        state["exp_avg_sq"] = state["exp_avg_sq"].to(grad)

                p_data_fp32 = p
                
                if isinstance(p_data_fp32, QBytesTensor):
                    p_data_fp32 = p_data_fp32.dequantize()
                if p.dtype != torch.float32:
                    p_data_fp32 = p_data_fp32.clone().float()

                state["step"] += 1
                state["RMS"] = self._rms(p_data_fp32)
                lr = self._get_lr(group, state)

                beta2t = 1.0 - math.pow(state["step"], group["decay_rate"])
                eps = group["eps"]
                if isinstance(eps, tuple) or isinstance(eps, list):
                    eps = eps[0]
                update = (grad**2) + eps
                if factored:
                    exp_avg_sq_row = state["exp_avg_sq_row"]
                    exp_avg_sq_col = state["exp_avg_sq_col"]

                    exp_avg_sq_row.mul_(beta2t).add_(
                        update.mean(dim=-1), alpha=(1.0 - beta2t))
                    exp_avg_sq_col.mul_(beta2t).add_(
                        update.mean(dim=-2), alpha=(1.0 - beta2t))

                    # Approximation of exponential moving average of square of gradient
                    update = self._approx_sq_grad(
                        exp_avg_sq_row, exp_avg_sq_col)
                    update.mul_(grad)
                else:
                    exp_avg_sq = state["exp_avg_sq"]

                    exp_avg_sq.mul_(beta2t).add_(update, alpha=(1.0 - beta2t))
                    update = exp_avg_sq.rsqrt().mul_(grad)

                update.div_(
                    (self._rms(update) / group["clip_threshold"]).clamp_(min=1.0))
                update.mul_(lr)

                if use_first_moment:
                    exp_avg = state["exp_avg"]
                    exp_avg.mul_(group["beta1"]).add_(
                        update, alpha=(1 - group["beta1"]))
                    update = exp_avg

                if group["weight_decay"] != 0:
                    p_data_fp32.add_(
                        p_data_fp32, alpha=(-group["weight_decay"] * lr))

                p_data_fp32.add_(-update)

                if p.dtype != torch.float32:
                    # apply stochastic rounding
                    copy_stochastic(p, p_data_fp32)

        return loss