truetype/transcriber
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: optimize Apple NPU backend with safety checker detection, dynamic CFG enforcement, and improved model loading feedback

Browse Source
This commit is contained in:
jarves committed 2026-06-25 06:53:34 +05:30
1 parent bf71e8b389
commit 8f9ab93bee
8 files changed
+431 -183
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

app/frontend/src/components/Generator.jsx
+5 -4
View File
@@ -186,12 +186,13 @@ function Generator({
const requestedBackend = constraints.backendType || (constraints.useGpu === false ? "cpu" : "auto");
const currentBackend = settings.backendType || (settings.useGpu === false ? "cpu" : "auto");
const shouldCompareModel = requestedBackend !== "openvino-npu";
const isCoreMLBackend = requestedBackend === "apple-npu";
if ((shouldCompareModel && currentModelName !== targetModelName) ||
parseInt(settings.steps) !== parseInt(constraints.steps) ||
Math.abs(parseFloat(settings.cfgScale) - parseFloat(constraints.cfgScale)) > 0.05 ||
settings.sampler !== constraints.sampler ||
parseInt(settings.threads) !== parseInt(constraints.threads) ||
(!isCoreMLBackend && parseInt(settings.steps) !== parseInt(constraints.steps)) ||
(!isCoreMLBackend && Math.abs(parseFloat(settings.cfgScale) - parseFloat(constraints.cfgScale)) > 0.05) ||
(!isCoreMLBackend && settings.sampler !== constraints.sampler) ||
(!isCoreMLBackend && parseInt(settings.threads) !== parseInt(constraints.threads)) ||
Boolean(settings.useGpu) !== (constraints.useGpu !== false) ||
parseInt(settings.width || 512) !== parseInt(constraints.width || 512) ||
parseInt(settings.height || 512) !== parseInt(constraints.height || 512) ||
app/frontend/src/components/ModelManager.jsx
+29 -4
View File
@@ -233,6 +233,24 @@ const COREML_MODEL_LIBRARY = [
notes: "Apple Silicon ANE (Neural Engine) optimized 6-bit palettized Stable Diffusion 1.5 model. Unzips automatically on download completion and runs extremely fast on Mac NPUs.",
url: "https://huggingface.co/apple/coreml-stable-diffusion-v1-5-palettized/resolve/main/coreml-stable-diffusion-v1-5-palettized_split_einsum_v2_compiled.zip",
},
{
name: "CyberRealistic v1.5 CoreML (6-bit palettized)",
filename: "cyberrealistic-6bit.coreml",
format: "CoreML",
approxSize: "815 MB",
resolution: "512x512",
notes: "6-bit palettized CoreML version of CyberRealistic. Extremely fast and optimized for Apple Silicon Neural Engine (ANE).",
url: "https://huggingface.co/orailnooor/cyberrealistic-coreml/resolve/main/cyberrealistic-6bit.coreml.zip",
},
{
name: "CyberRealistic v1.5 CoreML (Standard)",
filename: "cyberrealistic.coreml",
format: "CoreML",
approxSize: "1.8 GB",
resolution: "512x512",
notes: "Standard CoreML version of CyberRealistic. Provides high-fidelity realism optimized for Apple Silicon NPUs.",
url: "https://huggingface.co/orailnooor/cyberrealistic-coreml/resolve/main/cyberrealistic.coreml.zip",
},
],
},
];
@@ -991,7 +1009,7 @@ function ModelManager({
width: 512,
height: 512,
steps: activeConstraints.steps || 20,
cfgScale: activeConstraints.cfgScale || 7,
cfgScale: Number(activeConstraints.cfgScale) > 1 ? activeConstraints.cfgScale : 7,
}
: modelInfo?.backendType === "openvino-npu"
? {
@@ -1462,9 +1480,16 @@ function ModelManager({
<div className="model-progress-fill" style={{ width: `${Math.min(100, Math.max(0, modelLoadProgress.progress))}%`, transition: "width 0.2s ease" }}></div>
</div>
{(modelLoadProgress.backendMode || modelLoadProgress.device) && (
<div style={{ fontSize: "0.75rem", color: "var(--md-sys-color-outline)", marginTop: "6px" }}>
Initializing {modelLoadProgress.backendMode || "backend"}
{modelLoadProgress.device ? ` • ${modelLoadProgress.device}` : ""}
<div style={{ fontSize: "0.75rem", color: "var(--md-sys-color-outline)", marginTop: "6px", lineHeight: "1.4" }}>
<div>
Initializing {modelLoadProgress.backendMode || "backend"}
{modelLoadProgress.device ? ` • ${modelLoadProgress.device}` : ""}
</div>
{(modelLoadProgress.backendMode === "Apple NPU" || modelLoadProgress.backendMode === "Apple Neural Engine (NPU)") && (
<div style={{ color: "var(--md-sys-color-primary)", marginTop: "4px", fontWeight: "500" }}>
⚠️ Note: The first time loading this model, compilation on the Apple Neural Engine (NPU) can take 3–4 minutes. Subsequent loads will be almost instant.
</div>
)}
</div>
)}
</div>