ppo.py

from collections import namedtuple
from typing import Optional, Tuple
import torch
import torch.nn as nn
from torch.distributions import Independent, Normal
from ding.hpc_rl import hpc_wrapper

ppo_data = namedtuple(
    'ppo_data',
    ['logit_new', 'logit_old', 'action', 'value_new', 'value_old', 'adv', 'return_', 'weight', 'logit_pretrained']
)
ppo_data_continuous = namedtuple(
    'ppo_data_continuous', [
        'mu_sigma_new', 'mu_sigma_old', 'action', 'value_new', 'value_old', 'adv', 'return_', 'weight',
        'logit_pretrained'
    ]
)
ppo_policy_data = namedtuple(
    'ppo_policy_data', ['logit_new', 'logit_old', 'action', 'adv', 'weight', 'logit_pretrained']
)
ppo_policy_data_continuous = namedtuple(
    'ppo_policy_data_continuous', ['mu_sigma_new', 'mu_sigma_old', 'action', 'adv', 'weight', 'logit_pretrained']
)
ppo_value_data = namedtuple('ppo_value_data', ['value_new', 'value_old', 'return_', 'weight'])
ppo_loss = namedtuple('ppo_loss', ['policy_loss', 'value_loss', 'entropy_loss', 'kl_div'])
ppo_policy_loss = namedtuple('ppo_policy_loss', ['policy_loss', 'entropy_loss', 'kl_div'])
ppo_info = namedtuple('ppo_info', ['approx_kl', 'clipfrac'])


def calculate_kl_div(log_ratio: torch.Tensor, kl_type: str) -> torch.Tensor:
    """
    Overview:
        Calculate different Monte-Carlo estimators for KL-divergence KL(q, p) = E_q[log(q/p)],
        where q is the current policy and p is the pretrained policy.
        The implementation is based on John Schulman's blog post "Approximating KL Divergence".
        Reference: http://joschu.net/blog/kl-approx.html
    Arguments:
        - log_ratio (:obj:`torch.Tensor`): The log-ratio of probabilities, which should be
          log(q/p) = logp_new - logp_pretrained.
        - kl_type (:obj:`str`): The type of KL divergence estimator to use.
            - 'k1': The standard, unbiased but high-variance estimator: `E_q[log(q/p)]`.
            - 'k2': A biased, low-variance estimator from a second-order approximation: `E_q[1/2 * (log(p/q))^2]`.
            - 'k3': An unbiased, low-variance estimator: `E_q[(p/q - 1) - log(p/q)]`.
    Returns:
        - kl_div (:obj:`torch.Tensor`): The calculated KL divergence estimate.
    """
    if kl_type == 'k1':
        return log_ratio.mean()
    elif kl_type == 'k2':
        return (log_ratio ** 2 / 2).mean()
    elif kl_type == 'k3':
        return (torch.exp(-log_ratio) - 1 + log_ratio).mean()
    else:
        raise ValueError(f"Unknown kl_type: {kl_type}")


def shape_fn_ppo(args, kwargs):
    r"""
    Overview:
        Return shape of ppo for hpc
    Returns:
        shape: [B, N]
    """
    if len(args) <= 0:
        tmp = kwargs['data'].logit_new.shape
    else:
        tmp = args[0].logit_new.shape
    return tmp


@hpc_wrapper(
    shape_fn=shape_fn_ppo,
    namedtuple_data=True,
    include_args=[0, 1, 2, 3],
    include_kwargs=['data', 'clip_ratio', 'use_value_clip', 'dual_clip']
)
def ppo_error(
        data: namedtuple,
        clip_ratio: float = 0.2,
        use_value_clip: bool = True,
        dual_clip: Optional[float] = None,
        kl_type: str = 'k1'
) -> Tuple[namedtuple, namedtuple]:
    """
    Overview:
        Implementation of Proximal Policy Optimization (arXiv:1707.06347) with value_clip and dual_clip
    Arguments:
        - data (:obj:`namedtuple`): the ppo input data with fieids shown in ``ppo_data``
        - clip_ratio (:obj:`float`): the ppo clip ratio for the constraint of policy update, defaults to 0.2
        - use_value_clip (:obj:`bool`): whether to use clip in value loss with the same ratio as policy
        - dual_clip (:obj:`float`): a parameter c mentioned in arXiv:1912.09729 Equ. 5, shoule be in [1, inf),\
        defaults to 5.0, if you don't want to use it, set this parameter to None
        - kl_type (:obj:`str`): which kl loss to use, default set to 'k1'.
    Returns:
        - ppo_loss (:obj:`namedtuple`): the ppo loss item, all of them are the differentiable 0-dim tensor
        - ppo_info (:obj:`namedtuple`): the ppo optim information for monitoring, all of them are Python scalar
    Shapes:
        - logit_new (:obj:`torch.FloatTensor`): :math:`(B, N)`, where B is batch size and N is action dim
        - logit_old (:obj:`torch.FloatTensor`): :math:`(B, N)`
        - action (:obj:`torch.LongTensor`): :math:`(B, )`
        - value_new (:obj:`torch.FloatTensor`): :math:`(B, )`
        - value_old (:obj:`torch.FloatTensor`): :math:`(B, )`
        - adv (:obj:`torch.FloatTensor`): :math:`(B, )`
        - return (:obj:`torch.FloatTensor`): :math:`(B, )`
        - weight (:obj:`torch.FloatTensor` or :obj:`None`): :math:`(B, )`
        - policy_loss (:obj:`torch.FloatTensor`): :math:`()`, 0-dim tensor
        - value_loss (:obj:`torch.FloatTensor`): :math:`()`
        - entropy_loss (:obj:`torch.FloatTensor`): :math:`()`
    Examples:
        >>> action_dim = 4
        >>> data = ppo_data(
        >>>     logit_new=torch.randn(3, action_dim),
        >>>     logit_old=torch.randn(3, action_dim),
        >>>     action=torch.randint(0, action_dim, (3,)),
        >>>     value_new=torch.randn(3),
        >>>     value_old=torch.randn(3),
        >>>     adv=torch.randn(3),
        >>>     return_=torch.randn(3),
        >>>     weight=torch.ones(3),
        >>> )
        >>> loss, info = ppo_error(data)

    .. note::

        adv is already normalized value (adv - adv.mean()) / (adv.std() + 1e-8), and there are many
        ways to calculate this mean and std, like among data buffer or train batch, so we don't couple
        this part into ppo_error, you can refer to our examples for different ways.
    """
    assert dual_clip is None or dual_clip > 1.0, "dual_clip value must be greater than 1.0, but get value: {}".format(
        dual_clip
    )
    logit_new, logit_old, action, value_new, value_old, adv, return_, weight, logit_pretrained = data
    policy_data = ppo_policy_data(logit_new, logit_old, action, adv, weight, logit_pretrained)
    policy_output, policy_info = ppo_policy_error(policy_data, clip_ratio, dual_clip, kl_type=kl_type)
    value_data = ppo_value_data(value_new, value_old, return_, weight)
    value_loss = ppo_value_error(value_data, clip_ratio, use_value_clip)

    return ppo_loss(
        policy_output.policy_loss, value_loss, policy_output.entropy_loss, policy_output.kl_div
    ), policy_info


def ppo_policy_error(
        data: namedtuple,
        clip_ratio: float = 0.2,
        dual_clip: Optional[float] = None,
        entropy_bonus: bool = True,
        kl_type: str = 'k1'
) -> Tuple[namedtuple, namedtuple]:
    """
    Overview:
        Get PPO policy loss (both for classical RL in control/video games and LLM/VLM RLHF).
    Arguments:
        - data (:obj:`namedtuple`): Ppo input data with fieids shown in ``ppo_policy_data``.
        - clip_ratio (:obj:`float`): Clip value for ratio, defaults to 0.2.
        - dual_clip (:obj:`float`): A parameter c mentioned in arXiv:1912.09729 Equ. 5, shoule be in [1, inf), \
            defaults to 5.0, if you don't want to use it, set this parameter to None
        - entropy_bonus (:obj:`bool`): Whether to use entropy bonus, defaults to True. LLM RLHF usually does not use it.
        - kl_type (:obj:`str`): which kl loss to use, default set to 'k1'.
    Returns:
        - ppo_policy_loss (:obj:`namedtuple`): the ppo policy loss item, all of them are the differentiable 0-dim tensor
        - ppo_info (:obj:`namedtuple`): the ppo optim information for monitoring, all of them are Python scalar
    Shapes:
        - logit_new (:obj:`torch.FloatTensor`): :math:`(B, N)`, where B is batch size and N is action dim
        - logit_old (:obj:`torch.FloatTensor`): :math:`(B, N)`
        - action (:obj:`torch.LongTensor`): :math:`(B, )`
        - adv (:obj:`torch.FloatTensor`): :math:`(B, )`
        - weight (:obj:`torch.FloatTensor` or :obj:`None`): :math:`(B, )`
        - policy_loss (:obj:`torch.FloatTensor`): :math:`()`, 0-dim tensor
        - entropy_loss (:obj:`torch.FloatTensor`): :math:`()`
    Examples:
        >>> action_dim = 4
        >>> data = ppo_policy_data(
        >>>     logit_new=torch.randn(3, action_dim),
        >>>     logit_old=torch.randn(3, action_dim),
        >>>     action=torch.randint(0, action_dim, (3,)),
        >>>     adv=torch.randn(3),
        >>>     weight=torch.ones(3),
        >>> )
        >>> loss, info = ppo_policy_error(data)

    .. note::
        This function can be extended from `B` to more parallel dimensions, like `(B, S)`, where `S` is the
        sequence length in LLM/VLM.

    .. note::
        For the action mask often used in LLM/VLM, users can set the `weight` to the action mask.
    """
    logit_new, logit_old, action, adv, weight, logit_pretrained = data
    if weight is None:
        weight = torch.ones_like(adv)
    dist_new = torch.distributions.categorical.Categorical(logits=logit_new)
    dist_old = torch.distributions.categorical.Categorical(logits=logit_old)
    logp_new = dist_new.log_prob(action)
    logp_old = dist_old.log_prob(action)

    if entropy_bonus:
        dist_new_entropy = dist_new.entropy()
        if dist_new_entropy.shape != weight.shape:  # for the multi-agent rl case
            dist_new_entropy = dist_new.entropy().mean(dim=1)
        entropy_loss = (dist_new_entropy * weight).mean()
    else:
        entropy_loss = torch.tensor(0.0)
    # policy_loss
    ratio = torch.exp(logp_new - logp_old)
    if ratio.shape != adv.shape:
        ratio = ratio.mean(dim=1)
    surr1 = ratio * adv
    surr2 = ratio.clamp(1 - clip_ratio, 1 + clip_ratio) * adv
    if dual_clip is not None:
        clip1 = torch.min(surr1, surr2)
        clip2 = torch.max(clip1, dual_clip * adv)
        # only use dual_clip when adv < 0
        policy_loss = -(torch.where(adv < 0, clip2, clip1) * weight).mean()
    else:
        policy_loss = (-torch.min(surr1, surr2) * weight).mean()
    with torch.no_grad():
        approx_kl = (logp_old - logp_new).mean().item()
        clipped = ratio.gt(1 + clip_ratio) | ratio.lt(1 - clip_ratio)
        clipfrac = torch.as_tensor(clipped).float().mean().item()

    if logit_pretrained is not None:
        dist_pretrained = torch.distributions.categorical.Categorical(logits=logit_pretrained)
        logp_pretrained = dist_pretrained.log_prob(action)
        log_ratio = logp_new - logp_pretrained
        kl_div = calculate_kl_div(log_ratio, kl_type)
    else:
        kl_div = 0

    return ppo_policy_loss(policy_loss, entropy_loss, kl_div), ppo_info(approx_kl, clipfrac)


def ppo_value_error(
        data: namedtuple,
        clip_ratio: float = 0.2,
        use_value_clip: bool = True,
) -> torch.Tensor:
    '''
    Overview:
        Get PPO value loss
    Arguments:
        - data (:obj:`namedtuple`): ppo input data with fieids shown in ``ppo_value_data``
        - clip_ratio (:obj:`float`): clip value for ratio
        - use_value_clip (:obj:`bool`): whether use value clip
    Returns:
        - value_loss (:obj:`torch.FloatTensor`): the ppo value loss item, \
            all of them are the differentiable 0-dim tensor
    Shapes:
        - value_new (:obj:`torch.FloatTensor`): :math:`(B, )`, where B is batch size
        - value_old (:obj:`torch.FloatTensor`): :math:`(B, )`
        - return (:obj:`torch.FloatTensor`): :math:`(B, )`
        - weight (:obj:`torch.FloatTensor` or :obj:`None`): :math:`(B, )`
        - value_loss (:obj:`torch.FloatTensor`): :math:`()`, 0-dim tensor
    Examples:
        >>> action_dim = 4
        >>> data = ppo_value_data(
        >>>     value_new=torch.randn(3),
        >>>     value_old=torch.randn(3),
        >>>     return_=torch.randn(3),
        >>>     weight=torch.ones(3),
        >>> )
        >>> loss, info = ppo_value_error(data)
    '''
    value_new, value_old, return_, weight = data
    if weight is None:
        weight = torch.ones_like(value_old)
    # value_loss
    if use_value_clip:
        value_clip = value_old + (value_new - value_old).clamp(-clip_ratio, clip_ratio)
        v1 = (return_ - value_new).pow(2)
        v2 = (return_ - value_clip).pow(2)
        value_loss = 0.5 * (torch.max(v1, v2) * weight).mean()
    else:
        value_loss = 0.5 * ((return_ - value_new).pow(2) * weight).mean()
    return value_loss


def ppo_error_continuous(
        data: namedtuple,
        clip_ratio: float = 0.2,
        use_value_clip: bool = True,
        dual_clip: Optional[float] = None,
        kl_type: str = 'k1'
) -> Tuple[namedtuple, namedtuple]:
    """
    Overview:
        Implementation of Proximal Policy Optimization (arXiv:1707.06347) with value_clip and dual_clip
    Arguments:
        - data (:obj:`namedtuple`): the ppo input data with fieids shown in ``ppo_data``
        - clip_ratio (:obj:`float`): the ppo clip ratio for the constraint of policy update, defaults to 0.2
        - use_value_clip (:obj:`bool`): whether to use clip in value loss with the same ratio as policy
        - dual_clip (:obj:`float`): a parameter c mentioned in arXiv:1912.09729 Equ. 5, shoule be in [1, inf),\
        defaults to 5.0, if you don't want to use it, set this parameter to None
        - kl_type (:obj:`str`): which kl loss to use, default set to 'k1'.
    Returns:
        - ppo_loss (:obj:`namedtuple`): the ppo loss item, all of them are the differentiable 0-dim tensor
        - ppo_info (:obj:`namedtuple`): the ppo optim information for monitoring, all of them are Python scalar
    Shapes:
        - mu_sigma_new (:obj:`tuple`): :math:`((B, N), (B, N))`, where B is batch size and N is action dim
        - mu_sigma_old (:obj:`tuple`): :math:`((B, N), (B, N))`, where B is batch size and N is action dim
        - action (:obj:`torch.LongTensor`): :math:`(B, )`
        - value_new (:obj:`torch.FloatTensor`): :math:`(B, )`
        - value_old (:obj:`torch.FloatTensor`): :math:`(B, )`
        - adv (:obj:`torch.FloatTensor`): :math:`(B, )`
        - return (:obj:`torch.FloatTensor`): :math:`(B, )`
        - weight (:obj:`torch.FloatTensor` or :obj:`None`): :math:`(B, )`
        - policy_loss (:obj:`torch.FloatTensor`): :math:`()`, 0-dim tensor
        - value_loss (:obj:`torch.FloatTensor`): :math:`()`
        - entropy_loss (:obj:`torch.FloatTensor`): :math:`()`
    Examples:
        >>> action_dim = 4
        >>> data = ppo_data_continuous(
        >>>     mu_sigma_new= dict(mu=torch.randn(3, action_dim), sigma=torch.randn(3, action_dim)**2),
        >>>     mu_sigma_old= dict(mu=torch.randn(3, action_dim), sigma=torch.randn(3, action_dim)**2),
        >>>     action=torch.randn(3, action_dim),
        >>>     value_new=torch.randn(3),
        >>>     value_old=torch.randn(3),
        >>>     adv=torch.randn(3),
        >>>     return_=torch.randn(3),
        >>>     weight=torch.ones(3),
        >>> )
        >>> loss, info = ppo_error(data)

    .. note::

        adv is already normalized value (adv - adv.mean()) / (adv.std() + 1e-8), and there are many
        ways to calculate this mean and std, like among data buffer or train batch, so we don't couple
        this part into ppo_error, you can refer to our examples for different ways.
    """
    assert dual_clip is None or dual_clip > 1.0, "dual_clip value must be greater than 1.0, but get value: {}".format(
        dual_clip
    )
    mu_sigma_new, mu_sigma_old, action, value_new, value_old, adv, return_, weight, logit_pretrained = data
    if weight is None:
        weight = torch.ones_like(adv)

    dist_new = Independent(Normal(mu_sigma_new['mu'], mu_sigma_new['sigma']), 1)
    if len(mu_sigma_old['mu'].shape) == 1:
        dist_old = Independent(Normal(mu_sigma_old['mu'].unsqueeze(-1), mu_sigma_old['sigma'].unsqueeze(-1)), 1)
    else:
        dist_old = Independent(Normal(mu_sigma_old['mu'], mu_sigma_old['sigma']), 1)
    logp_new = dist_new.log_prob(action)
    logp_old = dist_old.log_prob(action)
    entropy_loss = (dist_new.entropy() * weight).mean()
    # policy_loss
    ratio = torch.exp(logp_new - logp_old)
    surr1 = ratio * adv
    surr2 = ratio.clamp(1 - clip_ratio, 1 + clip_ratio) * adv
    if dual_clip is not None:
        policy_loss = (-torch.max(torch.min(surr1, surr2), dual_clip * adv) * weight).mean()
    else:
        policy_loss = (-torch.min(surr1, surr2) * weight).mean()
    with torch.no_grad():
        approx_kl = (logp_old - logp_new).mean().item()
        clipped = ratio.gt(1 + clip_ratio) | ratio.lt(1 - clip_ratio)
        clipfrac = torch.as_tensor(clipped).float().mean().item()
    # value_loss
    if use_value_clip:
        value_clip = value_old + (value_new - value_old).clamp(-clip_ratio, clip_ratio)
        v1 = (return_ - value_new).pow(2)
        v2 = (return_ - value_clip).pow(2)
        value_loss = 0.5 * (torch.max(v1, v2) * weight).mean()
    else:
        value_loss = 0.5 * ((return_ - value_new).pow(2) * weight).mean()

    if logit_pretrained is not None:
        dist_pretrained = Independent(Normal(logit_pretrained['mu'], logit_pretrained['sigma']), 1)
        logp_pretrained = dist_pretrained.log_prob(action)
        log_ratio = logp_new - logp_pretrained
        kl_div = calculate_kl_div(log_ratio, kl_type)
    else:
        kl_div = 0

    return ppo_loss(policy_loss, value_loss, entropy_loss, kl_div), ppo_info(approx_kl, clipfrac)


def ppo_policy_error_continuous(
        data: namedtuple,
        clip_ratio: float = 0.2,
        dual_clip: Optional[float] = None,
        kl_type: str = 'k1'
) -> Tuple[namedtuple, namedtuple]:
    """
    Overview:
        Implementation of Proximal Policy Optimization (arXiv:1707.06347) with dual_clip
    Arguments:
        - data (:obj:`namedtuple`): the ppo input data with fieids shown in ``ppo_data``
        - clip_ratio (:obj:`float`): the ppo clip ratio for the constraint of policy update, defaults to 0.2
        - dual_clip (:obj:`float`): a parameter c mentioned in arXiv:1912.09729 Equ. 5, shoule be in [1, inf),\
        defaults to 5.0, if you don't want to use it, set this parameter to None
        - kl_type (:obj:`str`): which kl loss to use, default set to 'k1'.
    Returns:
        - ppo_loss (:obj:`namedtuple`): the ppo loss item, all of them are the differentiable 0-dim tensor
        - ppo_info (:obj:`namedtuple`): the ppo optim information for monitoring, all of them are Python scalar
    Shapes:
        - mu_sigma_new (:obj:`tuple`): :math:`((B, N), (B, N))`, where B is batch size and N is action dim
        - mu_sigma_old (:obj:`tuple`): :math:`((B, N), (B, N))`, where B is batch size and N is action dim
        - action (:obj:`torch.LongTensor`): :math:`(B, )`
        - adv (:obj:`torch.FloatTensor`): :math:`(B, )`
        - weight (:obj:`torch.FloatTensor` or :obj:`None`): :math:`(B, )`
        - policy_loss (:obj:`torch.FloatTensor`): :math:`()`, 0-dim tensor
        - entropy_loss (:obj:`torch.FloatTensor`): :math:`()`
    Examples:
        >>> action_dim = 4
        >>> data = ppo_policy_data_continuous(
        >>>     mu_sigma_new=dict(mu=torch.randn(3, action_dim), sigma=torch.randn(3, action_dim)**2),
        >>>     mu_sigma_old=dict(mu=torch.randn(3, action_dim), sigma=torch.randn(3, action_dim)**2),
        >>>     action=torch.randn(3, action_dim),
        >>>     adv=torch.randn(3),
        >>>     weight=torch.ones(3),
        >>> )
        >>> loss, info = ppo_policy_error_continuous(data)
    """
    assert dual_clip is None or dual_clip > 1.0, "dual_clip value must be greater than 1.0, but get value: {}".format(
        dual_clip
    )
    mu_sigma_new, mu_sigma_old, action, adv, weight, logit_pretrained = data
    if weight is None:
        weight = torch.ones_like(adv)

    dist_new = Independent(Normal(mu_sigma_new['mu'], mu_sigma_new['sigma']), 1)
    if len(mu_sigma_old['mu'].shape) == 1:
        dist_old = Independent(Normal(mu_sigma_old['mu'].unsqueeze(-1), mu_sigma_old['sigma'].unsqueeze(-1)), 1)
    else:
        dist_old = Independent(Normal(mu_sigma_old['mu'], mu_sigma_old['sigma']), 1)
    logp_new = dist_new.log_prob(action)
    logp_old = dist_old.log_prob(action)
    entropy_loss = (dist_new.entropy() * weight).mean()
    # policy_loss
    ratio = torch.exp(logp_new - logp_old)
    surr1 = ratio * adv
    surr2 = ratio.clamp(1 - clip_ratio, 1 + clip_ratio) * adv
    if dual_clip is not None:
        policy_loss = (-torch.max(torch.min(surr1, surr2), dual_clip * adv) * weight).mean()
    else:
        policy_loss = (-torch.min(surr1, surr2) * weight).mean()
    with torch.no_grad():
        approx_kl = (logp_old - logp_new).mean().item()
        clipped = ratio.gt(1 + clip_ratio) | ratio.lt(1 - clip_ratio)
        clipfrac = torch.as_tensor(clipped).float().mean().item()

    if logit_pretrained is not None:
        dist_pretrained = Independent(Normal(logit_pretrained['mu'], logit_pretrained['sigma']), 1)
        logp_pretrained = dist_pretrained.log_prob(action)
        log_ratio = logp_new - logp_pretrained
        kl_div = calculate_kl_div(log_ratio, kl_type)
    else:
        kl_div = 0

    return ppo_policy_loss(policy_loss, entropy_loss, kl_div), ppo_info(approx_kl, clipfrac)