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

Add Hadamard support to SDNQ Atten

Browse Source
This commit is contained in:
Disty0
2026-06-23 14:17:34 +03:00
parent 904925c1ef
commit 72a4d2c2bd
4 changed files with 66 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/ui_definitions.py
+2
View File
@@ -254,6 +254,8 @@ def create_settings(cmd_opts):
"xformers_options": OptionInfo(['Flash attention'], "xFormers options", gr.CheckboxGroup, {"choices": ['Flash attention'] }),
"dynamic_attention_slice_rate": OptionInfo(0.5, "Dynamic Attention slicing rate", gr.Slider, {"minimum": 0.01, "maximum": max(gpu_memory,4), "step": 0.01}),
"dynamic_attention_trigger_rate": OptionInfo(1, "Dynamic Attention trigger rate", gr.Slider, {"minimum": 0.01, "maximum": max(gpu_memory,4)*2, "step": 0.01}),
"sdnq_attention_use_hadamard": OptionInfo(False, "SDNQ Attention use Hadamard", gr.Checkbox),
"sdnq_attention_smooth_k": OptionInfo(False, "SDNQ Attention use Smooth K", gr.Checkbox),
"sdnq_attention_matmul_type": OptionInfo("auto", "SDNQ Attention MatMul type", gr.Radio, {"choices": sdnq_matmul_modes}),
"sdnq_attention_pv_matmul_type": OptionInfo("auto", "SDNQ Attention PV MatMul type", gr.Radio, {"choices": sdnq_matmul_modes}),
"sdnq_attention_quant_group_size": OptionInfo(128, "SDNQ Attention Quantization Group Size", gr.Number, {"minimum": 32, "maximum": 1024, "step": 1}),