Studio: keep automatic model loading toast visible until completion (#7425)

studio/frontend/src/components/ui/sonner.tsx

@@ -8,8 +8,8 @@ import {
   MultiplicationSignCircleIcon,
 } from "@hugeicons/core-free-icons";
 import { HugeiconsIcon } from "@hugeicons/react";
-import { Spinner } from "@/components/ui/spinner";
 import { useTheme } from "@/features/settings/stores/theme-store";
+import { createLoadingToastIcon } from "@/lib/toast";
 import { Toaster as Sonner, type ToasterProps } from "sonner";
 
 // Make toast text selectable. Sonner's onPointerDown calls setPointerCapture(),
@@ -78,7 +78,7 @@ const Toaster = ({ ...props }: ToasterProps) => {
             />
           ),
           // App-wide arc spinner so loading toasts match the "Downloading model" toast.
-          loading: <Spinner className="size-4 text-muted-foreground" />,
+          loading: createLoadingToastIcon(),
         }}
         style={
           {

studio/frontend/src/features/chat/api/chat-adapter.ts

@@ -6,7 +6,7 @@ import { resolveInitialConfig } from "@/features/model-picker";
 import { projectHasSources } from "@/features/rag/api/rag-api";
 import { apiUrl } from "@/lib/api-base";
 import { parseParamCountB } from "@/lib/model-size";
-import { toast } from "@/lib/toast";
+import { createLoadingToastIcon, toast } from "@/lib/toast";
 import type { MessageTiming, ToolCallMessagePart } from "@assistant-ui/core";
 import type { ChatModelAdapter } from "@assistant-ui/react";
 import { parsePartialJsonObject } from "assistant-stream/utils";
@@ -1512,13 +1512,38 @@ async function autoLoadSmallestModel(): Promise<{
   const trustRemoteCode = store.params.trustRemoteCode ?? false;
   const specSettings = resolveSpeculativeSettingsForLoad();
   const lastLoaded = readLastLocalModelLoad();
-  const toastId = toast("Loading a model…", {
+  let autoLoadToastDismissed = false;
+  const toastId = toast.message("Loading a model…", {
     description: lastLoaded
       ? "Loading last used model."
       : "Auto-selecting the smallest downloaded model.",
-    duration: 5000,
+    duration: Number.POSITIVE_INFINITY,
     closeButton: true,
+    icon: createLoadingToastIcon(),
+    onDismiss: () => {
+      autoLoadToastDismissed = true;
+    },
   });
+  const updateAutoLoadToast = (message: string, description: string): void => {
+    if (autoLoadToastDismissed) return;
+    toast.message(message, {
+      id: toastId,
+      description,
+      duration: Number.POSITIVE_INFINITY,
+    });
+  };
+  const showAutoLoadSuccess = (message: string): void => {
+    const options = {
+      description: undefined,
+      duration: 5000,
+      icon: undefined,
+    };
+    if (autoLoadToastDismissed) {
+      toast.success(message, options);
+      return;
+    }
+    toast.success(message, { ...options, id: toastId });
+  };
   let blockedByTrustRemoteCode = false;
   let hadNonTrustFailure = false;
   let loadAttempts = 0;
@@ -1774,7 +1799,7 @@ async function autoLoadSmallestModel(): Promise<{
         ggufVariant: candidate.ggufVariant,
       });
     }
-    toast.success(candidate.successLabel, { id: toastId });
+    showAutoLoadSuccess(candidate.successLabel);
     return true;
   }
   try {
@@ -1800,11 +1825,10 @@ async function autoLoadSmallestModel(): Promise<{
                 isAutoLoadableGgufVariant(entry),
             );
             if (variant) {
-              toast("Loading last used model…", {
-                id: toastId,
-                description: `${repo.repo_id} (${variant.quant})`,
-                duration: 5000,
-              });
+              updateAutoLoadToast(
+                "Loading last used model…",
+                `${repo.repo_id} (${variant.quant})`,
+              );
               if (
                 await loadAutoLoadCandidate({
                   id: repo.repo_id,
@@ -1829,11 +1853,7 @@ async function autoLoadSmallestModel(): Promise<{
         const repo = findCachedRepo(modelRepos, lastLoaded.id);
         if (repo) {
           try {
-            toast("Loading last used model…", {
-              id: toastId,
-              description: repo.repo_id,
-              duration: 5000,
-            });
+            updateAutoLoadToast("Loading last used model…", repo.repo_id);
             if (
               await loadAutoLoadCandidate({
                 id: repo.repo_id,
@@ -1854,11 +1874,10 @@ async function autoLoadSmallestModel(): Promise<{
           }
         }
       }
-      toast("Loading a model…", {
-        id: toastId,
-        description: "Auto-selecting the smallest downloaded model.",
-        duration: 5000,
-      });
+      updateAutoLoadToast(
+        "Loading a model…",
+        "Auto-selecting the smallest downloaded model.",
+      );
     }
 
     // GGUF first: smallest-total-size repo, then its smallest variant.
@@ -1949,12 +1968,10 @@ async function autoLoadSmallestModel(): Promise<{
     }
 
     // No cached models — try downloading a small default GGUF.
-    toast("Downloading a small model…", {
-      id: toastId,
-      description:
-        "No downloaded models found. Fetching Qwen3.5-4B-MTP (UD-Q4_K_XL).",
-      duration: 30000,
-    });
+    updateAutoLoadToast(
+      "Downloading a small model…",
+      "No downloaded models found. Fetching Qwen3.5-4B-MTP (UD-Q4_K_XL).",
+    );
     try {
       const rt = useChatRuntimeStore.getState();
       if (
@@ -2050,7 +2067,7 @@ async function autoLoadSmallestModel(): Promise<{
         kind: "gguf",
         ggufVariant: "UD-Q4_K_XL",
       });
-      toast.success("Loaded Qwen3.5-4B-MTP (UD-Q4_K_XL)", { id: toastId });
+      showAutoLoadSuccess("Loaded Qwen3.5-4B-MTP (UD-Q4_K_XL)");
       return { loaded: true, blockedByTrustRemoteCode: false };
     } catch {
       toast.dismiss(toastId);

studio/frontend/src/features/recipe-studio/hooks/use-recipe-executions.ts

@@ -2,7 +2,7 @@
 // Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
 
 import { getInferenceStatus, loadModel } from "@/features/chat";
-import { toast } from "@/lib/toast";
+import { createLoadingToastIcon, toast } from "@/lib/toast";
 import { toastError } from "@/shared/toast";
 import { useCallback, useEffect, useState } from "react";
 import { useShallow } from "zustand/react/shallow";
@@ -238,8 +238,15 @@ async function loadLocalModelSelection(
 ): Promise<string | null> {
   const { target, ggufVariant } = selection;
   const modelLabel = ggufVariant ? `${target} (${ggufVariant})` : target;
-  const toastId = toast.loading(`Loading ${modelLabel}...`, {
+  let loadToastDismissed = false;
+  const toastId = toast.message(`Loading ${modelLabel}...`, {
     description: "Starting the local inference server for this recipe.",
+    duration: Number.POSITIVE_INFINITY,
+    closeButton: true,
+    icon: createLoadingToastIcon(),
+    onDismiss: () => {
+      loadToastDismissed = true;
+    },
   });
   try {
     const isGguf = GGUF_MODEL_PATTERN.test(target) || Boolean(ggufVariant);
@@ -267,7 +274,16 @@ async function loadLocalModelSelection(
       // biome-ignore lint/style/useNamingConvention: api schema
       tensor_parallel: false,
     });
-    toast.success(`Loaded ${modelLabel}`, { id: toastId, duration: 2000 });
+    const successOptions = {
+      description: undefined,
+      duration: 2000,
+      icon: undefined,
+    };
+    if (loadToastDismissed) {
+      toast.success(`Loaded ${modelLabel}`, successOptions);
+    } else {
+      toast.success(`Loaded ${modelLabel}`, { ...successOptions, id: toastId });
+    }
     return null;
   } catch (error) {
     toast.dismiss(toastId);

studio/frontend/src/lib/toast.ts

@@ -4,5 +4,15 @@
 // Re-export of sonner. Swipe blocking lives on the Toaster via
 // `swipeDirections={[]}`, so no per-toast dismissible override.
 
+import { Spinner } from "@/components/ui/spinner";
+import { createElement } from "react";
+
+function createLoadingToastIcon() {
+  return createElement(Spinner, {
+    className: "size-4 text-muted-foreground",
+  });
+}
+
 export { toast } from "sonner";
 export type { ExternalToast } from "sonner";
+export { createLoadingToastIcon };

tests/studio/test_model_picker_contracts.py

@@ -73,6 +73,57 @@ def test_autoload_records_backend_loaded_model_identity():
     assert "m.id === loadedModelId" in autoload
 
 
+def test_chat_autoload_toast_is_persistent_and_dismissible():
+    """Send-triggered autoload stays visible until it settles but remains
+    dismissible, matching the explicit model-loading toast's lifetime."""
+    src = _read("features/chat/api/chat-adapter.ts")
+    auto_load = src.split("async function autoLoadSmallestModel", 1)[1]
+    auto_load = auto_load.split("export function createOpenAIStreamAdapter", 1)[0]
+    assert "toast.loading(" not in auto_load
+    assert "const updateAutoLoadToast =" in auto_load
+    assert "if (autoLoadToastDismissed) return;" in auto_load
+    assert auto_load.count("toast.message(") == 2
+    assert auto_load.count("updateAutoLoadToast(") >= 4
+    assert "duration: Number.POSITIVE_INFINITY" in auto_load
+    assert "closeButton: true" in auto_load
+    assert "icon: createLoadingToastIcon()" in auto_load
+    assert "onDismiss:" in auto_load
+    # Terminal success uses a fresh finite toast after manual progress dismissal.
+    assert "showAutoLoadSuccess" in auto_load
+    assert "description: undefined" in auto_load
+    assert "icon: undefined" in auto_load
+    assert "duration: 5000" in auto_load
+    assert "duration: 30000" not in auto_load
+    assert auto_load.count("toast.dismiss(toastId)") >= 4
+
+    explicit_load = _read("features/chat/hooks/use-chat-model-runtime.ts")
+    assert "duration: Infinity" in explicit_load
+
+
+def test_recipe_model_load_toast_is_persistent_and_dismissible():
+    """Recipe model loading uses the same dismissible persistent lifecycle as
+    chat loading because both call the non-abortable loadModel API."""
+    src = _read("features/recipe-studio/hooks/use-recipe-executions.ts")
+    model_load = src.split("async function loadLocalModelSelection", 1)[1]
+    model_load = model_load.split("function getLocalModelLoadPlanForPayload", 1)[0]
+    assert "toast.loading(" not in model_load
+    assert "toast.message(" in model_load
+    assert "duration: Number.POSITIVE_INFINITY" in model_load
+    assert "closeButton: true" in model_load
+    assert "icon: createLoadingToastIcon()" in model_load
+    assert "onDismiss:" in model_load
+    assert "description: undefined" in model_load
+    assert "icon: undefined" in model_load
+    assert "duration: 2000" in model_load
+
+    toast_lib = _read("lib/toast.ts")
+    assert "createElement(Spinner" in toast_lib
+    assert 'className: "size-4 text-muted-foreground"' in toast_lib
+
+    sonner = _read("components/ui/sonner.tsx")
+    assert "loading: createLoadingToastIcon()" in sonner
+
+
 def test_rollback_restores_native_lease_expiry_with_token():
     """A failed model switch that rolls back to a previously loaded picked GGUF
     must restore the lease expiry paired with the token, never the token alone