Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Re-measure the Studio startup budget after the managed engine options (#12560)

Daniel Hanfaca4e2278f3
diff --git a/studio/frontend/scripts/check-bundle-budget.ts b/studio/frontend/scripts/check-bundle-budget.ts
index 47cf0f74fe3..8ae5189de24 100644
--- a/studio/frontend/scripts/check-bundle-budget.ts
+++ b/studio/frontend/scripts/check-bundle-budget.ts
@@ -32,11 +32,13 @@ const DIST = resolve(HERE, "..", "dist");
* machine.
*/
export const BUDGET = {
- // Re-measured together on one machine and build: 5,813.8 KB raw / 1,730.2 KB transfer at d7795365c,
- // plus the margin the previous raise chose (210.8 KB raw, 65.1 KB transfer). The growth since is
- // chat attachment cards, chips and their layout (#12017), which the first chat screen renders.
- transferBytes: 1_840_000,
- rawBytes: 6_170_000,
+ // Re-measured together on one machine and build: 6,023.4 KB raw / 1,802.1 KB transfer at 5b78bafd04,
+ // plus the margin the previous raises chose (210.8 KB raw, 65.1 KB transfer). The growth since is
+ // the vLLM / SGLang engine options inside the model config page and Resources tab (#11491), which
+ // live inline in the chat and settings chunks: lazy-loading the separable inference-engines
+ // module recovers only 0.6 KB transfer, so what ran out is headroom, not laziness.
+ transferBytes: 1_912_000,
+ rawBytes: 6_384_000,
};
// The chunk count is reported but not budgeted. Splitting a page out of the entry raises it while lowering the