Merge branch 'master' into dr-support-pip-cm

This commit is contained in:
Dr.Lt.Data 2025-10-14 12:34:58 +09:00
commit b180f47d0e
3 changed files with 15 additions and 6 deletions

View File

@ -345,9 +345,9 @@ try:
if torch_version_numeric >= (2, 7): # works on 2.6 but doesn't actually seem to improve much if torch_version_numeric >= (2, 7): # works on 2.6 but doesn't actually seem to improve much
if any((a in arch) for a in ["gfx90a", "gfx942", "gfx1100", "gfx1101", "gfx1151"]): # TODO: more arches, TODO: gfx950 if any((a in arch) for a in ["gfx90a", "gfx942", "gfx1100", "gfx1101", "gfx1151"]): # TODO: more arches, TODO: gfx950
ENABLE_PYTORCH_ATTENTION = True ENABLE_PYTORCH_ATTENTION = True
# if torch_version_numeric >= (2, 8): if rocm_version >= (7, 0):
# if any((a in arch) for a in ["gfx1201"]): if any((a in arch) for a in ["gfx1201"]):
# ENABLE_PYTORCH_ATTENTION = True ENABLE_PYTORCH_ATTENTION = True
if torch_version_numeric >= (2, 7) and rocm_version >= (6, 4): if torch_version_numeric >= (2, 7) and rocm_version >= (6, 4):
if any((a in arch) for a in ["gfx1200", "gfx1201", "gfx942", "gfx950"]): # TODO: more arches if any((a in arch) for a in ["gfx1200", "gfx1201", "gfx942", "gfx950"]): # TODO: more arches
SUPPORT_FP8_OPS = True SUPPORT_FP8_OPS = True

View File

@ -276,8 +276,13 @@ class VAE:
if 'decoder.up_blocks.0.resnets.0.norm1.weight' in sd.keys(): #diffusers format if 'decoder.up_blocks.0.resnets.0.norm1.weight' in sd.keys(): #diffusers format
sd = diffusers_convert.convert_vae_state_dict(sd) sd = diffusers_convert.convert_vae_state_dict(sd)
self.memory_used_encode = lambda shape, dtype: (1767 * shape[2] * shape[3]) * model_management.dtype_size(dtype) #These are for AutoencoderKL and need tweaking (should be lower) if model_management.is_amd():
self.memory_used_decode = lambda shape, dtype: (2178 * shape[2] * shape[3] * 64) * model_management.dtype_size(dtype) VAE_KL_MEM_RATIO = 2.73
else:
VAE_KL_MEM_RATIO = 1.0
self.memory_used_encode = lambda shape, dtype: (1767 * shape[2] * shape[3]) * model_management.dtype_size(dtype) * VAE_KL_MEM_RATIO #These are for AutoencoderKL and need tweaking (should be lower)
self.memory_used_decode = lambda shape, dtype: (2178 * shape[2] * shape[3] * 64) * model_management.dtype_size(dtype) * VAE_KL_MEM_RATIO
self.downscale_ratio = 8 self.downscale_ratio = 8
self.upscale_ratio = 8 self.upscale_ratio = 8
self.latent_channels = 4 self.latent_channels = 4

View File

@ -39,7 +39,11 @@ if hasattr(torch.serialization, "add_safe_globals"): # TODO: this was added in
pass pass
ModelCheckpoint.__module__ = "pytorch_lightning.callbacks.model_checkpoint" ModelCheckpoint.__module__ = "pytorch_lightning.callbacks.model_checkpoint"
from numpy.core.multiarray import scalar def scalar(*args, **kwargs):
from numpy.core.multiarray import scalar as sc
return sc(*args, **kwargs)
scalar.__module__ = "numpy.core.multiarray"
from numpy import dtype from numpy import dtype
from numpy.dtypes import Float64DType from numpy.dtypes import Float64DType
from _codecs import encode from _codecs import encode