Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 09:14:35 +02:00
Code Issues Packages Projects Releases Wiki Activity

enhance automated testing

Browse Source 
Signed-off-by: Vladimir Mandic <mandic00@live.com>
This commit is contained in:
Vladimir Mandic
2026-06-03 09:27:34 +02:00
parent c47d2e0dd8
commit ec2e37ee6e
52 changed files with 192 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/model_quant.py
+4 -1
View File
@@ -172,7 +172,10 @@ def create_sdnq_config(kwargs = None, allow: bool = True, module: str = 'Model',
if quantized_matmul_dtype is None:
quantized_matmul_dtype = "auto" # set for logging
svd = f'{shared.opts.sdnq_use_svd} rank={shared.opts.sdnq_svd_rank} steps={shared.opts.sdnq_svd_steps}' if shared.opts.sdnq_use_svd else f'{shared.opts.sdnq_use_svd}'
log.debug(f'Quantization: module="{module}" type=sdnq mode=pre dtype={weights_dtype} svd={svd} hadamard={shared.opts.sdnq_use_hadamard} dynamic={shared.opts.sdnq_use_dynamic_quantization} group={shared.opts.sdnq_group_size} hadamard_group={shared.opts.sdnq_hadamard_group_size} loss={shared.opts.sdnq_dynamic_loss_threshold} matmul_dtype={quantized_matmul_dtype} matmul_quant={shared.opts.sdnq_use_quantized_matmul} matmul_conv={shared.opts.sdnq_use_quantized_matmul_conv} quant_conv={shared.opts.sdnq_quantize_conv_layers} quant_embedding={shared.opts.sdnq_quantize_embedding_layers} fp32={shared.opts.sdnq_dequantize_fp32} device={quantization_device} return={return_device} use_gpu={shared.opts.sdnq_quantize_with_gpu} map={shared.opts.device_map} offload={shared.opts.diffusers_offload_mode} non_blocking={shared.opts.diffusers_offload_nonblocking} skip_modules={modules_to_not_convert} dict={modules_dtype_dict}')
hadamard = f'{shared.opts.sdnq_use_hadamard} group={shared.opts.sdnq_hadamard_group_size}' if shared.opts.sdnq_use_hadamard else f'{shared.opts.sdnq_use_hadamard}'
log.debug(f'Quantization: module="{module}" type=sdnq mode=pre dtype={weights_dtype} svd={svd} hadamard={hadamard} dynamic={shared.opts.sdnq_use_dynamic_quantization} group={shared.opts.sdnq_group_size} loss={shared.opts.sdnq_dynamic_loss_threshold} matmul_dtype={quantized_matmul_dtype} matmul_quant={shared.opts.sdnq_use_quantized_matmul} matmul_conv={shared.opts.sdnq_use_quantized_matmul_conv} quant_conv={shared.opts.sdnq_quantize_conv_layers} quant_embed={shared.opts.sdnq_quantize_embedding_layers} fp32={shared.opts.sdnq_dequantize_fp32} device={quantization_device} return={return_device} gpu={shared.opts.sdnq_quantize_with_gpu} map={shared.opts.device_map}')
if len(modules_to_not_convert) > 0 or modules_dtype_dict:
log.debug(f'Quantization: module={module} type=sdnq skip_modules={modules_to_not_convert} modules_dtype_dict={modules_dtype_dict}')
if kwargs is None:
return sdnq_config
else: