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

SDNQ INT8 matmul support for Conv2d

Browse Source
This commit is contained in:
Disty0
2025-06-05 13:36:49 +03:00
parent c7fb5b1690
commit e25890bb1d
2 changed files with 99 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/shared.py
+2 -1
View File
@@ -521,10 +521,11 @@ options_templates.update(options_section(("quantization", "Quantization Settings
"sdnq_quantize_weights_mode": OptionInfo("int8", "Quantization type", gr.Dropdown, {"choices": ["int8", "int6", "uint4", "float8_e4m3fn", "uint8", "uint6", "int4", "float8_e5m2", "float8_e4m3fnuz", "float8_e5m2fnuz", "int2", "uint2", "uint1"], "visible": native}),
"sdnq_quantize_weights_group_size": OptionInfo(0, "Group size", gr.Slider, {"minimum": -1, "maximum": 4096, "step": 1, "visible": native}),
"sdnq_quantize_conv_layers": OptionInfo(False, "Quantize the convolutional layers", gr.Checkbox, {"visible": native}),
"sdnq_decompress_fp32": OptionInfo(False, "Decompress using full precision", gr.Checkbox, {"visible": native}),
"sdnq_decompress_compile": OptionInfo(devices.has_triton(), "Decompress using torch.compile", gr.Checkbox, {"visible": native}),
"sdnq_use_quantized_matmul": OptionInfo(False, "Use Quantized MatMul", gr.Checkbox, {"visible": native}),
"sdnq_use_quantized_matmul_conv": OptionInfo(False, "Use Quantized MatMul with convolutional layers", gr.Checkbox, {"visible": native}),
"sdnq_quantize_with_gpu": OptionInfo(True, "Quantize with the GPU", gr.Checkbox, {"visible": native}),
"sdnq_decompress_fp32": OptionInfo(False, "Decompress using full precision", gr.Checkbox, {"visible": native}),
"sdnq_quantize_shuffle_weights": OptionInfo(False, "Shuffle weights in post mode", gr.Checkbox, {"visible": native}),
"bnb_quantization_sep": OptionInfo("<h2>BitsAndBytes</h2>", "", gr.HTML),