aljaz/DALLE2-pytorch
1
1
Fork 0
mirror of https://github.com/lucidrains/DALLE2-pytorch.git synced 2025-12-19 01:34:19 +01:00
Code Issues Packages Projects Releases Wiki Activity

more shots in the dark regarding fp16 with learned variance for deepspeed issue

Browse Source
This commit is contained in:
Phil Wang
2022-07-06 19:05:50 -07:00
parent a6cdbe0b9c
commit 6a59c7093d
2 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
dalle2_pytorch/dalle2_pytorch.py
View File

@@ -337,7 +337,7 @@ def discretized_gaussian_log_likelihood(x, *, means, log_scales, thres = 0.999):
# attempting to correct nan gradients when learned variance is turned on # attempting to correct nan gradients when learned variance is turned on
# in the setting of deepspeed fp16 # in the setting of deepspeed fp16
eps = 1e-12 if x.dtype == torch.float32 else 1e-5 eps = 1e-12 if x.dtype == torch.float32 else 1e-3
centered_x = x - means centered_x = x - means
inv_stdv = torch.exp(-log_scales) inv_stdv = torch.exp(-log_scales)
@@ -345,8 +345,8 @@ def discretized_gaussian_log_likelihood(x, *, means, log_scales, thres = 0.999):
cdf_plus = approx_standard_normal_cdf(plus_in) cdf_plus = approx_standard_normal_cdf(plus_in)
min_in = inv_stdv * (centered_x - 1. / 255.) min_in = inv_stdv * (centered_x - 1. / 255.)
cdf_min = approx_standard_normal_cdf(min_in) cdf_min = approx_standard_normal_cdf(min_in)
log_cdf_plus = log(cdf_plus) log_cdf_plus = log(cdf_plus, eps = eps)
log_one_minus_cdf_min = log(1. - cdf_min) log_one_minus_cdf_min = log(1. - cdf_min, eps = eps)
cdf_delta = cdf_plus - cdf_min cdf_delta = cdf_plus - cdf_min
log_probs = torch.where(x < -thres, log_probs = torch.where(x < -thres,

2
dalle2_pytorch/version.py
View File

@@ -1 +1 @@
__version__ = '0.16.13' __version__ = '0.16.14'