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

more granular quantization modules options

Browse Source 
Signed-off-by: Vladimir Mandic <mandic00@live.com>
This commit is contained in:
Vladimir Mandic
2025-03-28 14:46:52 -04:00
parent a6e8e88974
commit f4fdd496b9
14 changed files with 30 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/shared.py
+9 -9
View File
@@ -494,7 +494,7 @@ options_templates.update(options_section(('backends', "Backend Settings"), {
"olive_cache_optimized": OptionInfo(True, 'Olive cache optimized models'),
"ipex_sep": OptionInfo("<h2>IPEX</h2>", "", gr.HTML, {"visible": devices.backend == "ipex"}),
"ipex_optimize": OptionInfo([], "IPEX Optimize", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder", "Upscaler"], "visible": devices.backend == "ipex"}),
"ipex_optimize": OptionInfo([], "IPEX Optimize", gr.CheckboxGroup, {"choices": ["Model", "VAE", "TE", "Upscaler"], "visible": devices.backend == "ipex"}),
"openvino_sep": OptionInfo("<h2>OpenVINO</h2>", "", gr.HTML, {"visible": cmd_opts.use_openvino}),
"openvino_devices": OptionInfo([], "OpenVINO devices to use", gr.CheckboxGroup, {"choices": get_openvino_device_list() if cmd_opts.use_openvino else [], "visible": cmd_opts.use_openvino}), # pylint: disable=E0606
@@ -509,36 +509,36 @@ options_templates.update(options_section(('backends', "Backend Settings"), {
options_templates.update(options_section(('quantization', "Quantization Settings"), {
"bnb_quantization_sep": OptionInfo("<h2>BitsAndBytes</h2>", "", gr.HTML),
"bnb_quantization": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder", "LLM"], "visible": native}),
"bnb_quantization": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "Transformer", "VAE", "TE", "Video", "LLM"], "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}),
"quanto_quantization_sep": OptionInfo("<h2>Optimum Quanto</h2>", "", gr.HTML),
"quanto_quantization": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder"], "visible": native}),
"quanto_quantization": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "Transformer", "VAE", "TE", "Video", "LLM"], "visible": native}),
"quanto_quantization_type": OptionInfo("int8", "Quantization weights type", gr.Dropdown, {"choices": ["float8", "int8", "int4", "int2"], "visible": native}),
"optimum_quanto_sep": OptionInfo("<h2>Optimum Quanto: post-load</h2>", "", gr.HTML),
"optimum_quanto_weights": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder", "ControlNet"], "visible": native}),
"optimum_quanto_weights": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "Transformer", "VAE", "TE", "Video", "LLM", "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}),
"optimum_quanto_shuffle_weights": OptionInfo(False, "Shuffle weights", gr.Checkbox, {"visible": native}),
"torchao_sep": OptionInfo("<h2>TorchAO</h2>", "", gr.HTML),
"torchao_quantization": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder"], "visible": native}),
"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_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_compress_sep": OptionInfo("<h2>NNCF: Neural Network Compression Framework</h2>", "", gr.HTML),
"nncf_compress_weights": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder", "ControlNet"], "visible": native}),
"nncf_compress_weights": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "Transformer", "VAE", "TE", "Video", "LLM", "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(0, "Compress ratio", gr.Slider, {"minimum": 0, "maximum": 1, "step": 0.01, "visible": cmd_opts.use_openvino}),
"nncf_compress_weights_group_size": OptionInfo(0, "Group size", gr.Slider, {"minimum": -1, "maximum": 512, "step": 1, "visible": cmd_opts.use_openvino}),
"nncf_quantize": OptionInfo([], "OpenVINO enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder"], "visible": cmd_opts.use_openvino}),
"nncf_quantize": OptionInfo([], "OpenVINO enabled", gr.CheckboxGroup, {"choices": ["Model", "VAE", "TE"], "visible": cmd_opts.use_openvino}),
"nncf_quantize_mode": OptionInfo("INT8", "OpenVINO mode", gr.Dropdown, {"choices": ['INT8', 'FP8_E4M3', 'FP8_E5M2'], "visible": cmd_opts.use_openvino}),
"nncf_quantize_shuffle_weights": OptionInfo(False, "Shuffle weights", gr.Checkbox, {"visible": native}),
"layerwise_quantization_sep": OptionInfo("<h2>Layerwise Casting</h2>", "", gr.HTML),
"layerwise_quantization": OptionInfo([], "Layerwise casting enabled", gr.CheckboxGroup, {"choices": ["Model", "Transformer", "Text Encoder"], "visible": native}),
"layerwise_quantization": OptionInfo([], "Layerwise casting enabled", gr.CheckboxGroup, {"choices": ["Model", "Transformer", "TE"], "visible": native}),
"layerwise_quantization_storage": OptionInfo("float8_e4m3fn", "Layerwise casting storage", gr.Dropdown, {"choices": ["float8_e4m3fn", "float8_e5m2"], "visible": native}),
"layerwise_quantization_nonblocking": OptionInfo(False, "Layerwise non-blocking operations", gr.Checkbox, {"visible": native}),
}))
@@ -612,7 +612,7 @@ options_templates.update(options_section(('advanced', "Pipeline Modifiers"), {
options_templates.update(options_section(('compile', "Model Compile"), {
"cuda_compile_sep": OptionInfo("<h2>Model Compile</h2>", "", gr.HTML),
"cuda_compile": OptionInfo([] if not cmd_opts.use_openvino else ["Model", "VAE", "Upscaler"], "Compile Model", gr.CheckboxGroup, {"choices": ["Model", "VAE", "Text Encoder", "Upscaler"]}),
"cuda_compile": OptionInfo([] if not cmd_opts.use_openvino else ["Model", "VAE", "Upscaler"], "Compile Model", gr.CheckboxGroup, {"choices": ["Model", "VAE", "TE", "Upscaler"]}),
"cuda_compile_backend": OptionInfo("none" if not cmd_opts.use_openvino else "openvino_fx", "Model compile backend", gr.Radio, {"choices": ['none', 'inductor', 'cudagraphs', 'aot_ts_nvfuser', 'hidet', 'migraphx', 'ipex', 'onediff', 'stable-fast', 'deep-cache', 'olive-ai', 'openvino_fx']}),
"cuda_compile_mode": OptionInfo("default", "Model compile mode", gr.Radio, {"choices": ['default', 'reduce-overhead', 'max-autotune', 'max-autotune-no-cudagraphs']}),
"cuda_compile_fullgraph": OptionInfo(True if not cmd_opts.use_openvino else False, "Model compile fullgraph"),