implement VRAM usage warning, enable Linux CUDA backend support, and optimize flash attention flags.
This commit is contained in:
1 parent
74b0c3ba46
commit
54f4809990
6 files changed
+343
-142
No files matched your search
+5
-1
@@ -1484,7 +1484,11 @@ async function startBackend(settings = {}) {
|
||||
args.push("--vae-on-cpu");
|
||||
}
|
||||
if (currentSettings.flashAttn) {
|
||||
args.push("--fa");
|
||||
if (requestedBackend === "cuda") {
|
||||
args.push("--diffusion-fa");
|
||||
} else {
|
||||
args.push("--fa");
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
Reference in new issue
Block a user