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

unified logger

Browse Source
This commit is contained in:
Vladimir Mandic
2026-02-19 09:46:42 +01:00
parent bfe014f5da
commit a3074baf8b
315 changed files with 2507 additions and 2116 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/model_quant.py
+10 -8
View File
@@ -1,3 +1,4 @@
from modules import logger
import os
import re
import sys
@@ -6,7 +7,8 @@ import json
import time
import diffusers
import transformers
from installer import installed, install, log, setup_logging
from installer import installed, install, setup_logging
from modules.logger import log
ao = None
@@ -45,7 +47,7 @@ def dont_quant():
models_list = re.split(r'[ ,]+', shared.opts.models_not_to_quant)
models_list = [m.lower().strip() for m in models_list]
if shared.sd_model_type.lower() in models_list:
shared.log.debug(f'Quantization: model={shared.sd_model_type} skip')
logger.log.debug(f'Quantization: model={shared.sd_model_type} skip')
return True
return False
@@ -169,7 +171,7 @@ def create_sdnq_config(kwargs = None, allow: bool = True, module: str = 'Model',
from modules.sdnq.common import use_torch_compile as sdnq_use_torch_compile
if shared.opts.sdnq_use_quantized_matmul and not sdnq_use_torch_compile:
shared.log.warning('SDNQ Quantized MatMul requires a working Triton install. Disabling Quantized MatMul.')
logger.log.warning('SDNQ Quantized MatMul requires a working Triton install. Disabling Quantized MatMul.')
shared.opts.sdnq_use_quantized_matmul = False
if weights_dtype is None:
@@ -514,7 +516,7 @@ def sdnq_quantize_model(model, op=None, sd_model=None, do_gc: bool = True, weigh
from modules.sdnq.common import use_torch_compile as sdnq_use_torch_compile
if shared.opts.sdnq_use_quantized_matmul and not sdnq_use_torch_compile:
shared.log.warning('SDNQ Quantized MatMul requires a working Triton install. Disabling Quantized MatMul.')
logger.log.warning('SDNQ Quantized MatMul requires a working Triton install. Disabling Quantized MatMul.')
shared.opts.sdnq_use_quantized_matmul = False
if weights_dtype is None:
@@ -807,15 +809,15 @@ def do_post_load_quant(sd_model, allow=True):
if dont_quant():
return sd_model
if shared.opts.sdnq_quantize_weights and (shared.opts.sdnq_quantize_mode == 'post' or (allow and shared.opts.sdnq_quantize_mode == 'auto')):
shared.log.debug('Load model: post_quant=sdnq')
logger.log.debug('Load model: post_quant=sdnq')
sd_model = sdnq_quantize_weights(sd_model)
if len(shared.opts.optimum_quanto_weights) > 0:
shared.log.debug('Load model: post_quant=quanto')
logger.log.debug('Load model: post_quant=quanto')
sd_model = optimum_quanto_weights(sd_model)
if shared.opts.torchao_quantization and (shared.opts.torchao_quantization_mode == 'post' or (allow and shared.opts.torchao_quantization_mode == 'auto')):
shared.log.debug('Load model: post_quant=torchao')
logger.log.debug('Load model: post_quant=torchao')
sd_model = torchao_quantization(sd_model)
if shared.opts.layerwise_quantization:
shared.log.debug('Load model: post_quant=layerwise')
logger.log.debug('Load model: post_quant=layerwise')
apply_layerwise(sd_model)
return sd_model