enso/diffusion/__pycache__/diffusion_utils.cpython-310.pyc

o
<zduã@s4dd
lZ
dd
lZdd„Zdd„Zdd„Zdd	„Zd
S)
éNcsŠd
‰||
||fD]}t|t
jƒr|‰
n
qˆd
u
sJdƒ
‚
‡f
dd„|
|fDƒ
\}
}dd||
t
 |
|¡
||dt
 |¡
S)z¦
    Compute the KL divergence between two gaussians.
    Shapes are automatically broadcasted, so batches can be compared to
    scalars, among other use cases.
    Nz&at least one argument must be a Tensorc
s,g|]}
t|
t
jƒr|
nt
 |
¡
 ˆ¡
‘qS©)Ú
isinstanceÚthÚTensorÚtensorÚto)Ú.0Ú
x©
rrú]/maindata/data/shared/multimodal/zhengcong.fei/code/dit-moe/code/diffusion/diffusion_utils.pyÚ
<listcomp>sÿÿznormal_kl.<locals>.<listcomp>çà?gð¿é)rrrÚexp)Úmean1Úlogvar1Úmean2Úlogvar2Úobjrr
rÚ	normal_kl
s*



þ

þ

ÿþýüÿrc

Cs2d
dt 
t dtj¡
|dt |d¡¡
S)zb
    A fast approximation of the cumulative distribution function of the
    standard normal.
    r
çð?g@g÷
Hmâä¦?é)rÚtanhÚnpÚsqrtÚpiÚpow)
r	rrrÚapprox_standard_normal_cdf's2rc
Cs@||
}t 
|¡
}||}tj t |¡
t |¡
¡ |¡
}|S)
a

    Compute the log-likelihood of a continuous Gaussian distribution.
    :param x: the targets
    :param means: the Gaussian mean Tensor.
    :param log_scales: the Gaussian log stddev Tensor.
    :return: a tensor like x of log probabilities (in nats).
    )rrÚ
distributionsÚNormalÚ
zeros_likeÚ	ones_likeÚlog_prob)r	ÚmeansÚ
log_scalesÚ
centered_xÚinv_stdvZnormalized_xÚ	log_probsrrrÚ"continuous_gaussian_log_likelihood/s


 

r(c


CsÐ|j|
jkr|jksJ‚

J‚
||
}t
 |¡
}||d
}t|ƒ
}||d
}t|ƒ
}t
 |jdd�
¡
}	t
 d|jdd�
¡
}
||}t
 |dk|	t
 |dk|
t
 |jdd�
¡
¡¡}|j|jksfJ‚
|S)az

    Compute the log-likelihood of a Gaussian distribution discretizing to a
    given image.
    :param x: the target images. It is assumed that this was uint8 values,
              rescaled to the range [-1, 1].
    :param means: the Gaussian mean Tensor.