Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 01:04:32 +02:00
Code Issues Packages Projects Releases Wiki Activity

docs(ui): add tooltips for Nunchaku attention and offload settings

Browse Source
This commit is contained in:
CalamitousFelicitousness
2026-02-06 00:13:06 +00:00
parent 33de04a0c7
commit 5000401647
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
html/locale_en.json
+2
View File
@@ -337,6 +337,8 @@
{"id":"","label":"Model Options","localized":"","reload":"","hint":"Settings related to behavior of specific models"},
{"id":"","label":"Model Offloading","localized":"","reload":"","hint":"Settings related to model offloading and memory management"},
{"id":"","label":"Model Quantization","localized":"","reload":"","hint":"Settings related to model quantization which is used to reduce memory usage"},
{"id":"","label":"Nunchaku attention","localized":"","reload":"","hint":"Replaces default attention with Nunchaku's custom FP16 attention kernel for faster inference on consumer NVIDIA GPUs.<br>Might provide performance improvement on GPUs which have higher FP16 tensor cores throughput than BF16.<br><br>Currently only affects Flux-based models (Dev, Schnell, Kontext, Fill, Depth, etc.). Has no effect on Qwen, SDXL, Sana, or other architectures.<br><br>Disabled by default."},
{"id":"","label":"Nunchaku offloading","localized":"","reload":"","hint":"Enables Nunchaku's own per-block CPU offloading with asynchronous CUDA streams to reduce VRAM usage.<br>Uses a ping-pong buffer strategy: while one transformer block computes on GPU, the next block preloads from CPU in the background, hiding most of the transfer latency.<br><br>Can reduce VRAM usage at the cost of slower inference.<br>This replaces SD.Next's pipeline offloading for the transformer component.<br><br>Only useful on low-VRAM GPUs. If your GPU has enough memory to hold the quantized model (16+ GB), keep this disabled for maximum speed.<br>Supports Flux and Qwen models. Not supported for SDXL where this setting is ignored.<br>Disabled by default."},
{"id":"","label":"Image Metadata","localized":"","reload":"","hint":"Settings related to handling of metadata that is created with generated images"},
{"id":"","label":"Legacy Options","localized":"","reload":"","hint":"Settings related to legacy options - should not be used"},
{"id":"","label":"Restart server","localized":"","reload":"","hint":"Restart server"},