Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-18 16:54:33 +02:00
Code Issues Packages Projects Releases Wiki Activity

NNCF set the default group size to 128 for INT4

Browse Source
This commit is contained in:
Disty0
2025-05-11 08:45:27 +03:00
parent 4af570bfd4
commit 0673689d5b
2 changed files with 7 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/shared.py
+1 -1
View File
@@ -546,7 +546,7 @@ options_templates.update(options_section(('quantization', "Quantization Settings
"nncf_compress_weights_mode": OptionInfo("INT8_SYM", "Quantization type", gr.Dropdown, {"choices": ['INT8', 'INT8_SYM', 'INT4_ASYM', 'INT4_SYM', 'NF4'] if cmd_opts.use_openvino else ['INT8', 'INT8_SYM', 'INT4', 'INT4_SYM']}),
"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_compress_weights_num_of_groups": OptionInfo(1, "Number of quantization groups", gr.Slider, {"minimum": 1, "maximum": 128, "step": 1, "visible": native and not cmd_opts.use_openvino}),
"nncf_compress_weights_num_of_groups": OptionInfo(0, "Number of quantization groups", gr.Slider, {"minimum": 0, "maximum": 128, "step": 1, "visible": native and not 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 activations mode", gr.Dropdown, {"choices": ['INT8', 'FP8_E4M3', 'FP8_E5M2'], "visible": cmd_opts.use_openvino}),
"nncf_quantize_conv_layers": OptionInfo(False, "Quantize the convolutional layers", gr.Checkbox, {"visible": native and not cmd_opts.use_openvino}),