protenix/utils/training.py

# Copyright 2024 ByteDance and/or its affiliates.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#      http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import inspect

import torch
import torch.distributed as dist


def get_adamw(
    model: torch.nn.Module,
    weight_decay: float,
    learning_rate: float,
    betas: tuple[float, float],
    device_type: str,
) -> torch.optim.AdamW:
    """
    Create an AdamW optimizer for the given model with specified parameters.

    Args:
        model (torch.nn.Module): The model for which the optimizer is created.
        weight_decay (float): The weight decay (L2 penalty) for the optimizer.
        learning_rate (float): The learning rate for the optimizer.
        betas (tuple): Coefficients used for computing running averages of gradient and its square.
        device_type (str): The device type ('cuda' or 'cpu') on which the optimizer will operate.

    Returns:
        torch.optim.AdamW: The AdamW optimizer configured with the specified parameters.
    """
    # start with all of the candidate parameters
    param_dict = {pn: p for pn, p in model.named_parameters()}
    # filter out those that do not require grad
    param_dict = {pn: p for pn, p in param_dict.items() if p.requires_grad}
    # create optim groups. Any parameters that is 2D will be weight decayed, otherwise no.
    # i.e. all weight tensors in matmuls + embeddings decay, all biases and layernorms don't.
    decay_params = [p for n, p in param_dict.items() if p.dim() >= 2]
    nodecay_params = [p for n, p in param_dict.items() if p.dim() < 2]
    optim_groups = [
        {"params": decay_params, "weight_decay": weight_decay},
        {"params": nodecay_params, "weight_decay": 0.0},
    ]
    num_decay_params = sum(p.numel() for p in decay_params)
    num_nodecay_params = sum(p.numel() for p in nodecay_params)
    print(
        f"num decayed parameter tensors: {len(decay_params)}, with {num_decay_params:,} parameters"
    )
    print(
        f"num non-decayed parameter tensors: {len(nodecay_params)}, with {num_nodecay_params:,} parameters"
    )
    # Create AdamW optimizer and use the fused version if it is available
    fused_available = "fused" in inspect.signature(torch.optim.AdamW).parameters
    use_fused = fused_available and device_type == "cuda"
    extra_args = dict(fused=True) if use_fused else dict()
    optimizer = torch.optim.AdamW(
        optim_groups, lr=learning_rate, betas=betas, **extra_args
    )
    print(f"using fused AdamW: {use_fused}")

    return optimizer


def get_optimizer(configs, model: torch.nn.Module) -> torch.optim.Optimizer:
    if configs.adam.use_adamw:
        optimizer = get_adamw(
            model=model,
            weight_decay=configs.adam.weight_decay,
            learning_rate=configs.adam.lr,
            betas=(configs.adam.beta1, configs.adam.beta2),
            device_type="cuda" if torch.cuda.is_available() else "cpu",
        )
    else:
        optimizer = torch.optim.Adam(
            model.parameters(),
            lr=configs.adam.lr,
            weight_decay=configs.adam.weight_decay,
            betas=(configs.adam.beta1, configs.adam.beta2),
        )
    return optimizer


def is_loss_nan_check(loss: torch.Tensor) -> bool:
    """check the validness of the current loss

    Args:
        loss: the loss from the model

    Returns:
        bool: if True, loss is not nan or inf
    """

    def is_nan(x):
        return torch.isnan(x).any() or torch.isinf(x).any()

    def all_reduce_tensor(tensor, op=dist.ReduceOp.SUM):
        if dist.is_initialized():
            dist.all_reduce(tensor, op=op)
        return tensor

    nan_flag = torch.tensor(
        1.0 if is_nan(loss) else 0.0,
        device=loss.device if torch.cuda.is_available() else None,
    )  # support cpu
    # avoid "Watchdog caught collective operation timeout" error
    all_reduce_tensor(nan_flag)
    if nan_flag.item() > 0.0:
        return True
    return False