Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 17:24:32 +02:00
Code Issues Packages Projects Releases Wiki Activity

add torchao

Browse Source 
Signed-off-by: Vladimir Mandic <mandic00@live.com>
This commit is contained in:
Vladimir Mandic
2024-12-17 13:29:36 -05:00
parent 36909a9f25
commit fd7fe8cea5
10 changed files with 43 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/shared.py
+12 -12
View File
@@ -583,23 +583,23 @@ options_templates.update(options_section(('backends', "Backend Settings"), {
options_templates.update(options_section(('quantization', "Quantization Settings"), {
"bnb_sep": OptionInfo("<h2>BitsAndBytes</h2>", "", gr.HTML),
"bnb_quantization": OptionInfo([], "Enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder"], "visible": native}),
"bnb_quantization_type": OptionInfo("nf4", "Type", gr.Radio, {"choices": ['nf4', 'fp8', 'fp4'], "visible": native}),
"bnb_quantization_storage": OptionInfo("uint8", "Backend storage", gr.Radio, {"choices": ["float16", "float32", "int8", "uint8", "float64", "bfloat16"], "visible": native}),
"bnb_quantization": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder"], "visible": native}),
"bnb_quantization_type": OptionInfo("nf4", "Quantization type", gr.Dropdown, {"choices": ['nf4', 'fp8', 'fp4'], "visible": native}),
"bnb_quantization_storage": OptionInfo("uint8", "Backend storage", gr.Dropdown, {"choices": ["float16", "float32", "int8", "uint8", "float64", "bfloat16"], "visible": native}),
"optimum_quanto_sep": OptionInfo("<h2>Optimum Quanto</h2>", "", gr.HTML),
"optimum_quanto_weights": OptionInfo([], "Enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder", "ControlNet"], "visible": native}),
"optimum_quanto_weights_type": OptionInfo("qint8", "Type", gr.Radio, {"choices": ['qint8', 'qfloat8_e4m3fn', 'qfloat8_e5m2', 'qint4', 'qint2'], "visible": native}),
"optimum_quanto_activations_type": OptionInfo("none", "Activations ", gr.Radio, {"choices": ['none', 'qint8', 'qfloat8_e4m3fn', 'qfloat8_e5m2'], "visible": native}),
"optimum_quanto_weights": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder", "ControlNet"], "visible": native}),
"optimum_quanto_weights_type": OptionInfo("qint8", "Quantization weights type", gr.Dropdown, {"choices": ['qint8', 'qfloat8_e4m3fn', 'qfloat8_e5m2', 'qint4', 'qint2'], "visible": native}),
"optimum_quanto_activations_type": OptionInfo("none", "Quantization activations type ", gr.Dropdown, {"choices": ['none', 'qint8', 'qfloat8_e4m3fn', 'qfloat8_e5m2'], "visible": native}),
"torchao_sep": OptionInfo("<h2>TorchAO</h2>", "", gr.HTML),
"torchao_quantization": OptionInfo([], "Enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder"], "visible": native}),
"torchao_quantization_mode": OptionInfo("pre", "Mode", gr.Radio, {"choices": ['pre', 'post'], "visible": native}),
"torchao_quantization_type": OptionInfo("int8", "Type", gr.Radio, {"choices": ["int8+act", "int8", "int4", "fp8+act", "fp8", "fpx"], "visible": native}),
"torchao_quantization": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder"], "visible": native}),
"torchao_quantization_mode": OptionInfo("pre", "Quantization mode", gr.Dropdown, {"choices": ['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}),
"nncf_sep": OptionInfo("<h2>NNCF</h2>", "", gr.HTML),
"nncf_compress_weights": OptionInfo([], "Enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder", "ControlNet"], "visible": native}),
"nncf_compress_weights_mode": OptionInfo("INT8", "Mode", gr.Radio, {"choices": ['INT8', 'INT8_SYM', 'INT4_ASYM', 'INT4_SYM', 'NF4'] if cmd_opts.use_openvino else ['INT8']}),
"nncf_compress_weights": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder", "ControlNet"], "visible": native}),
"nncf_compress_weights_mode": OptionInfo("INT8", "Quantization type", gr.Dropdown, {"choices": ['INT8', 'INT8_SYM', 'INT4_ASYM', 'INT4_SYM', 'NF4'] if cmd_opts.use_openvino else ['INT8']}),
"nncf_compress_weights_raito": OptionInfo(1.0, "Compress ratio", gr.Slider, {"minimum": 0, "maximum": 1, "step": 0.01, "visible": cmd_opts.use_openvino}),
"nncf_quantize": OptionInfo([], "OpenVINO enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder"], "visible": cmd_opts.use_openvino}),
"nncf_quant_mode": OptionInfo("INT8", "OpenVINO mode", gr.Radio, {"choices": ['INT8', 'FP8_E4M3', 'FP8_E5M2'], "visible": cmd_opts.use_openvino}),
"nncf_quant_mode": OptionInfo("INT8", "OpenVINO mode", gr.Dropdown, {"choices": ['INT8', 'FP8_E4M3', 'FP8_E5M2'], "visible": cmd_opts.use_openvino}),
"quant_shuffle_weights": OptionInfo(False, "Shuffle weights", gr.Checkbox, {"visible": native}),
}))