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

Add auto quantization mode

Browse Source
This commit is contained in:
Disty0
2025-06-27 18:54:15 +03:00
parent 70193c5c64
commit 3b8ced444c
4 changed files with 33 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/shared.py
+2 -2
View File
@@ -518,7 +518,7 @@ options_templates.update(options_section(('backends', "Backend Settings"), {
options_templates.update(options_section(("quantization", "Quantization Settings"), {
"sdnq_quantize_sep": OptionInfo("<h2>SDNQ: SD.Next Quantization</h2>", "", gr.HTML),
"sdnq_quantize_weights": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "Transformer", "VAE", "TE", "Video", "LLM", "ControlNet"], "visible": native}),
"sdnq_quantize_mode": OptionInfo("pre", "Quantization mode", gr.Dropdown, {"choices": ["pre", "post"], "visible": native}),
"sdnq_quantize_mode": OptionInfo("auto", "Quantization mode", gr.Dropdown, {"choices": ["auto", "pre", "post"], "visible": native}),
"sdnq_quantize_weights_mode": OptionInfo("int8", "Quantization type", gr.Dropdown, {"choices": sdnq_quant_modes, "visible": native}),
"sdnq_quantize_weights_mode_te": OptionInfo("default", "Quantization type for Text Encoders", gr.Dropdown, {"choices": ['default'] + sdnq_quant_modes, "visible": native}),
"sdnq_quantize_weights_group_size": OptionInfo(0, "Group size", gr.Slider, {"minimum": -1, "maximum": 4096, "step": 1, "visible": native}),
@@ -547,7 +547,7 @@ options_templates.update(options_section(("quantization", "Quantization Settings
"torchao_sep": OptionInfo("<h2>TorchAO</h2>", "", gr.HTML),
"torchao_quantization": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "Transformer", "VAE", "TE", "Video", "LLM"], "visible": native}),
"torchao_quantization_mode": OptionInfo("pre", "Quantization mode", gr.Dropdown, {"choices": ["pre", "post"], "visible": native}),
"torchao_quantization_mode": OptionInfo("auto", "Quantization mode", gr.Dropdown, {"choices": ["auto", "pre", "post"], "visible": native}),
"torchao_quantization_type": OptionInfo("int8_weight_only", "Quantization type", gr.Dropdown, {"choices": ["int4_weight_only", "int8_dynamic_activation_int4_weight", "int8_weight_only", "int8_dynamic_activation_int8_weight", "float8_weight_only", "float8_dynamic_activation_float8_weight", "float8_static_activation_float8_weight"], "visible": native}),
"layerwise_quantization_sep": OptionInfo("<h2>Layerwise Casting</h2>", "", gr.HTML),