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

NNCF experimental direct INT8 MatMul support

Browse Source
This commit is contained in:
Disty0
2025-05-12 21:41:49 +03:00
parent c3def9b601
commit f4e3a81a84
4 changed files with 225 additions and 111 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/shared.py
+1
View File
@@ -551,6 +551,7 @@ options_templates.update(options_section(('quantization', "Quantization Settings
"nncf_quantize_conv_layers": OptionInfo(False, "Quantize the convolutional layers", gr.Checkbox, {"visible": native and not cmd_opts.use_openvino}),
"nncf_decompress_fp32": OptionInfo(False, "Decompress using full precision", gr.Checkbox, {"visible": native and not cmd_opts.use_openvino}),
"nncf_decompress_compile": OptionInfo(devices.has_triton(), "Decompress using torch.compile", gr.Checkbox, {"visible": native and not cmd_opts.use_openvino}),
"nncf_decompress_int8_matmul": OptionInfo(False, "Use direct INT8 MatMul", gr.Checkbox, {"visible": native and not cmd_opts.use_openvino}),
"nncf_quantize_shuffle_weights": OptionInfo(False, "Shuffle weights in post mode", gr.Checkbox, {"visible": native and not cmd_opts.use_openvino}),
"layerwise_quantization_sep": OptionInfo("<h2>Layerwise Casting</h2>", "", gr.HTML),