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

SDNQ expose Dyn quant on settings

Browse Source
This commit is contained in:
Disty0
2026-01-14 16:54:50 +03:00
parent 034b0844fe
commit c2bc47e0c1
2 changed files with 8 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/model_quant.py
+6 -2
View File
@@ -221,10 +221,12 @@ def create_sdnq_config(kwargs = None, allow: bool = True, module: str = 'Model',
group_size=shared.opts.sdnq_quantize_weights_group_size,
svd_rank=shared.opts.sdnq_svd_rank,
svd_steps=shared.opts.sdnq_svd_steps,
dynamic_loss_threshold=shared.opts.sdnq_dynamic_loss_threshold,
use_svd=shared.opts.sdnq_use_svd,
quant_conv=shared.opts.sdnq_quantize_conv_layers,
use_quantized_matmul=shared.opts.sdnq_use_quantized_matmul,
use_quantized_matmul_conv=shared.opts.sdnq_use_quantized_matmul_conv,
use_dynamic_quantization=shared.opts.sdnq_use_dynamic_quantization,
dequantize_fp32=shared.opts.sdnq_dequantize_fp32,
non_blocking=shared.opts.diffusers_offload_nonblocking,
quantization_device=quantization_device,
@@ -234,7 +236,7 @@ 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
log.debug(f'Quantization: module="{module}" type=sdnq mode=pre dtype={weights_dtype} matmul_dtype={quantized_matmul_dtype} matmul={shared.opts.sdnq_use_quantized_matmul} group_size={shared.opts.sdnq_quantize_weights_group_size} svd_rank={shared.opts.sdnq_svd_rank} svd_steps={shared.opts.sdnq_svd_steps} use_svd={shared.opts.sdnq_use_svd} quant_conv={shared.opts.sdnq_quantize_conv_layers} matmul_conv={shared.opts.sdnq_use_quantized_matmul_conv} dequantize_fp32={shared.opts.sdnq_dequantize_fp32} quantize_with_gpu={shared.opts.sdnq_quantize_with_gpu} quantization_device={quantization_device} return_device={return_device} device_map={shared.opts.device_map} offload_mode={shared.opts.diffusers_offload_mode} non_blocking={shared.opts.diffusers_offload_nonblocking} modules_to_not_convert={modules_to_not_convert} modules_dtype_dict={modules_dtype_dict}')
log.debug(f'Quantization: module="{module}" type=sdnq mode=pre dtype={weights_dtype} matmul_dtype={quantized_matmul_dtype} matmul={shared.opts.sdnq_use_quantized_matmul} group_size={shared.opts.sdnq_quantize_weights_group_size} svd_rank={shared.opts.sdnq_svd_rank} svd_steps={shared.opts.sdnq_svd_steps} use_svd={shared.opts.sdnq_use_svd} dynamic={shared.opts.sdnq_use_dynamic_quantization} loss={shared.opts.sdnq_dynamic_loss_threshold} quant_conv={shared.opts.sdnq_quantize_conv_layers} matmul_conv={shared.opts.sdnq_use_quantized_matmul_conv} dequantize_fp32={shared.opts.sdnq_dequantize_fp32} quantize_with_gpu={shared.opts.sdnq_quantize_with_gpu} quantization_device={quantization_device} return_device={return_device} device_map={shared.opts.device_map} offload_mode={shared.opts.diffusers_offload_mode} non_blocking={shared.opts.diffusers_offload_nonblocking} modules_to_not_convert={modules_to_not_convert} modules_dtype_dict={modules_dtype_dict}')
if kwargs is None:
return sdnq_config
else:
@@ -556,10 +558,12 @@ def sdnq_quantize_model(model, op=None, sd_model=None, do_gc: bool = True, weigh
group_size=shared.opts.sdnq_quantize_weights_group_size,
svd_rank=shared.opts.sdnq_svd_rank,
svd_steps=shared.opts.sdnq_svd_steps,
dynamic_loss_threshold=shared.opts.sdnq_dynamic_loss_threshold,
use_svd=shared.opts.sdnq_use_svd,
quant_conv=shared.opts.sdnq_quantize_conv_layers,
use_quantized_matmul=shared.opts.sdnq_use_quantized_matmul,
use_quantized_matmul_conv=shared.opts.sdnq_use_quantized_matmul_conv,
use_dynamic_quantization=shared.opts.sdnq_use_dynamic_quantization,
dequantize_fp32=shared.opts.sdnq_dequantize_fp32,
non_blocking=shared.opts.diffusers_offload_nonblocking,
quantization_device=quantization_device,
@@ -594,7 +598,7 @@ def sdnq_quantize_model(model, op=None, sd_model=None, do_gc: bool = True, weigh
if quantized_matmul_dtype is None:
quantized_matmul_dtype = "auto" # set for logging
log.debug(f'Quantization: module="{op if op is not None else model.__class__}" type=sdnq mode=post dtype={weights_dtype} matmul_dtype={quantized_matmul_dtype} matmul={shared.opts.sdnq_use_quantized_matmul} svd={shared.opts.sdnq_use_svd}:group={shared.opts.sdnq_quantize_weights_group_size}:rank={shared.opts.sdnq_svd_rank}:steps={shared.opts.sdnq_svd_steps} quant_conv={shared.opts.sdnq_quantize_conv_layers} matmul_conv={shared.opts.sdnq_use_quantized_matmul_conv} fp32={shared.opts.sdnq_dequantize_fp32} gpu={shared.opts.sdnq_quantize_with_gpu} device={quantization_device} return={return_device} map={shared.opts.device_map} non_blocking={shared.opts.diffusers_offload_nonblocking} modules_skip={modules_to_not_convert} modules_dtype={modules_dtype_dict}')
log.debug(f'Quantization: module="{op if op is not None else model.__class__}" type=sdnq mode=post dtype={weights_dtype} matmul_dtype={quantized_matmul_dtype} matmul={shared.opts.sdnq_use_quantized_matmul} svd={shared.opts.sdnq_use_svd}dynamic={shared.opts.sdnq_use_dynamic_quantization}:group={shared.opts.sdnq_quantize_weights_group_size}:rank={shared.opts.sdnq_svd_rank}:steps={shared.opts.sdnq_svd_steps}:loss={shared.opts.sdnq_dynamic_loss_threshold} quant_conv={shared.opts.sdnq_quantize_conv_layers} matmul_conv={shared.opts.sdnq_use_quantized_matmul_conv} fp32={shared.opts.sdnq_dequantize_fp32} gpu={shared.opts.sdnq_quantize_with_gpu} device={quantization_device} return={return_device} map={shared.opts.device_map} non_blocking={shared.opts.diffusers_offload_nonblocking} modules_skip={modules_to_not_convert} modules_dtype={modules_dtype_dict}')
return model
modules/shared.py
+2
View File
@@ -232,7 +232,9 @@ options_templates.update(options_section(("quantization", "Model Quantization"),
"sdnq_quantize_weights_group_size": OptionInfo(0, "Group size", gr.Slider, {"minimum": -1, "maximum": 4096, "step": 1}),
"sdnq_svd_rank": OptionInfo(32, "SVD rank size", gr.Slider, {"minimum": 1, "maximum": 512, "step": 1}),
"sdnq_svd_steps": OptionInfo(8, "SVD steps", gr.Slider, {"minimum": 1, "maximum": 128, "step": 1}),
"sdnq_dynamic_loss_threshold": OptionInfo(1e-2, "Dynamic loss threshold", gr.Slider, {"minimum": 1e-4, "maximum": 1e-1, "step": 1e-4}),
"sdnq_use_svd": OptionInfo(False, "Use SVD quantization", gr.Checkbox),
"sdnq_use_dynamic_quantization": OptionInfo(False, "Use Dynamic quantization", gr.Checkbox),
"sdnq_quantize_conv_layers": OptionInfo(False, "Quantize convolutional layers", gr.Checkbox),
"sdnq_dequantize_compile": OptionInfo(devices.has_triton(early=True), "Dequantize using torch.compile", gr.Checkbox),
"sdnq_use_quantized_matmul": OptionInfo(False, "Use quantized MatMul", gr.Checkbox),