Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 17:24:32 +02:00
Code Issues Packages Projects Releases Wiki Activity

add native torch fp8 storage dtype

Browse Source 
Signed-off-by: Vladimir Mandic <mandic00@live.com>
This commit is contained in:
Vladimir Mandic
2025-01-23 12:07:08 -05:00
parent 9a67e5aeb2
commit 5a7c1f50c1
5 changed files with 91 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/sd_models.py
+4 -1
View File
@@ -9,7 +9,7 @@ import diffusers
import diffusers.loaders.single_file_utils
import torch
from modules import paths, shared, shared_state, modelloader, devices, script_callbacks, sd_vae, sd_unet, errors, sd_models_config, sd_models_compile, sd_hijack_accelerate, sd_detect
from modules import paths, shared, shared_state, modelloader, devices, script_callbacks, sd_vae, sd_unet, errors, sd_models_config, sd_models_compile, sd_hijack_accelerate, sd_detect, model_quant
from modules.timer import Timer, process as process_timer
from modules.memstats import memory_stats
from modules.modeldata import model_data
@@ -570,9 +570,12 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
sd_model = sd_models_compile.nncf_compress_weights(sd_model) # run this before move model so it can be compressed in CPU
if shared.opts.optimum_quanto_weights:
sd_model = sd_models_compile.optimum_quanto_weights(sd_model) # run this before move model so it can be compressed in CPU
if shared.opts.layerwise_quantization:
model_quant.apply_layerwise(sd_model)
timer.record("options")
set_diffuser_offload(sd_model, op)
if op == 'model' and not (os.path.isdir(checkpoint_info.path) or checkpoint_info.type == 'huggingface'):
if getattr(shared.sd_model, 'sd_checkpoint_info', None) is not None and vae_file is not None:
sd_vae.apply_vae_config(shared.sd_model.sd_checkpoint_info.filename, vae_file, sd_model)