Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-20 01:31:13 +02:00
Code Issues Packages Projects Releases Wiki Activity

Remove NNCF device hijack

Browse Source
This commit is contained in:
Disty0
2025-05-11 18:30:10 +03:00
parent 0673689d5b
commit 9cfdc3c079
3 changed files with 5 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/model_quant.py
+2 -4
View File
@@ -327,13 +327,13 @@ def apply_layerwise(sd_model, quiet:bool=False):
log.error(f'Quantization: type=layerwise {e}')
def nncf_compress_model(model, op=None, sd_model=None, send_to_device=True, do_gc=True):
def nncf_compress_model(model, op=None, sd_model=None, do_gc=True):
global quant_last_model_name, quant_last_model_device # pylint: disable=global-statement
from modules import devices, shared
from accelerate import init_empty_weights
load_nncf('Quantize model: type=NNCF')
from modules.model_quant_nncf import apply_nncf_to_module, nncf_send_to_device
from modules.model_quant_nncf import apply_nncf_to_module
from nncf.torch.nncf_module_replacement import replace_modules_by_nncf_modules # get around lazy import
model.eval()
@@ -358,8 +358,6 @@ def nncf_compress_model(model, op=None, sd_model=None, send_to_device=True, do_g
is_asym_mode = shared.opts.nncf_compress_weights_mode in {"INT8", "INT4", "INT8_ASYM", "INT4_ASYM"}
model = apply_nncf_to_module(model, num_bits, is_asym_mode, quant_conv=shared.opts.nncf_quantize_conv_layers)
model.quantization_method = 'NNCF'
if send_to_device:
nncf_send_to_device(model, devices.device)
if hasattr(model, "set_input_embeddings") and backup_embeddings is not None:
model.set_input_embeddings(backup_embeddings)