Merge remote-tracking branch 'origin/main' into studio-security-headers-pure-asgi

studio/backend/routes/llama.py

@@ -55,6 +55,9 @@ class LlamaUpdateStatusResponse(BaseModel):
     source_build: bool = Field(
         False, description = "True when there is no marker (source build) but a prebuilt is offered."
     )
+    update_size_bytes: Optional[int] = Field(
+        None, description = "Download size of the prebuilt Update would fetch, in bytes."
+    )
     job: LlamaUpdateJob = Field(default_factory = LlamaUpdateJob)
 
 

studio/backend/tests/test_llama_cpp_freshness.py

@@ -520,3 +520,114 @@ def test_in_memory_only_reset_replays_stale_same_base_mix(monkeypatch, tmp_path)
     assert info["latest_tag"] == "b9596-mix-aaa"
     assert info["behind"] is True
     assert info["stale"] is True
+
+
+# update_download_size_bytes (banner download-size lookup).
+
+
+def _patch_assets(monkeypatch, mapping):
+    """Stub latest_release_assets with a per-repo {asset_name: size} lookup."""
+    monkeypatch.setattr(
+        fr,
+        "latest_release_assets",
+        lambda repo, *, force_refresh = False: mapping.get(repo),
+    )
+
+
+def test_update_size_unsloth_prebuilt_exact_match(monkeypatch):
+    # The unsloth fork's own bundle (app-<tag>-<platform>): the want= exact match
+    # on app-<latest>-<suffix> wins.
+    marker = {
+        "asset": "app-b9190-linux-x64-cuda13-newer.tar.gz",
+        "published_repo": "unslothai/llama.cpp",
+    }
+    _patch_assets(
+        monkeypatch,
+        {
+            "unslothai/llama.cpp": {
+                "app-b9300-linux-x64-cuda13-newer.tar.gz": 123_456_789,
+                "app-b9300-windows-x64-cuda13-newer.zip": 999,
+            }
+        },
+    )
+    assert fr.update_download_size_bytes(marker, "b9300", "unslothai/llama.cpp") == 123_456_789
+
+
+def test_update_size_macos_fork_asset_suffix_fallback(monkeypatch):
+    # macOS bundles use the upstream-style llama-<tag>-bin-macos-*, matched via the
+    # endswith fallback in the publish repo.
+    marker = {
+        "asset": "llama-b9190-bin-macos-arm64.tar.gz",
+        "published_repo": "unslothai/llama.cpp",
+    }
+    _patch_assets(
+        monkeypatch,
+        {"unslothai/llama.cpp": {"llama-b9300-bin-macos-arm64.tar.gz": 55_000_000}},
+    )
+    assert fr.update_download_size_bytes(marker, "b9300", "unslothai/llama.cpp") == 55_000_000
+
+
+def test_update_size_upstream_ubuntu_uses_binary_repo(monkeypatch):
+    # #6338 P2: ggml-org ubuntu-* prebuilt lives in binary_repo, not the fork
+    # publish repo. The size must still resolve.
+    marker = {
+        "asset": "llama-b9190-bin-ubuntu-x64.tar.gz",
+        "published_repo": "unslothai/llama.cpp",
+        "binary_repo": "ggml-org/llama.cpp",
+    }
+    _patch_assets(
+        monkeypatch,
+        {
+            "unslothai/llama.cpp": {"app-b9300-linux-x64-cuda13-newer.tar.gz": 1},
+            "ggml-org/llama.cpp": {
+                "llama-b9673-bin-ubuntu-x64.tar.gz": 42_000_000,
+                "llama-b9673-bin-ubuntu-vulkan-x64.tar.gz": 7,
+            },
+        },
+    )
+    assert fr.update_download_size_bytes(marker, "b9300", "unslothai/llama.cpp") == 42_000_000
+
+
+def test_update_size_upstream_windows_uses_binary_repo(monkeypatch):
+    # Regression (#6338 P2): the Windows upstream CPU prebuilt uses a win-* token.
+    marker = {
+        "asset": "llama-b9190-bin-win-cpu-x64.zip",
+        "published_repo": "unslothai/llama.cpp",
+        "binary_repo": "ggml-org/llama.cpp",
+    }
+    _patch_assets(
+        monkeypatch,
+        {"ggml-org/llama.cpp": {"llama-b9673-bin-win-cpu-x64.zip": 33_000_000}},
+    )
+    assert fr.update_download_size_bytes(marker, "b9300", "unslothai/llama.cpp") == 33_000_000
+
+
+def test_update_size_no_matching_asset_fails_open(monkeypatch):
+    # A ROCm version drift (installed 6.4 vs latest 7.2) leaves no suffix match;
+    # the helper fails open to None rather than guessing a wrong artifact.
+    marker = {
+        "asset": "llama-b9190-bin-ubuntu-rocm-6.4-x64.tar.gz",
+        "published_repo": "unslothai/llama.cpp",
+        "binary_repo": "ggml-org/llama.cpp",
+    }
+    _patch_assets(
+        monkeypatch,
+        {"ggml-org/llama.cpp": {"llama-b9673-bin-ubuntu-rocm-7.2-x64.tar.gz": 9}},
+    )
+    assert fr.update_download_size_bytes(marker, "b9300", "unslothai/llama.cpp") is None
+
+
+def test_update_size_missing_inputs_fail_open(monkeypatch):
+    _patch_assets(
+        monkeypatch,
+        {"unslothai/llama.cpp": {"app-b9300-linux-x64-cpu.tar.gz": 5}},
+    )
+    # No marker, no latest tag, or no asset string -> None (never raise).
+    assert fr.update_download_size_bytes(None, "b9300", "unslothai/llama.cpp") is None
+    assert (
+        fr.update_download_size_bytes(
+            {"asset": "app-b9190-linux-x64-cpu.tar.gz"}, None, "unslothai/llama.cpp"
+        )
+        is None
+    )
+    assert fr.update_download_size_bytes({"asset": None}, "b9300", "unslothai/llama.cpp") is None

studio/backend/tests/test_llama_cpp_update.py

@@ -901,3 +901,49 @@ def test_start_update_marked_refuses_when_not_behind(monkeypatch, tmp_path):
     res = upd.start_update()
     assert res["started"] is False
     assert res["reason"] == "up_to_date"
+
+
+def test_status_update_available_includes_size(monkeypatch, tmp_path):
+    # Marker (prebuilt) update path attaches the download size of the asset the
+    # banner would fetch.
+    binary = _write_install(tmp_path, "b9493", asset = "app-b9493-linux-x64-cuda13-newer.tar.gz")
+    monkeypatch.setattr(upd, "_find_binary", lambda: binary)
+    monkeypatch.setattr(freshness, "_fetch_latest_release_tag", lambda repo, timeout = 5.0: "b9518")
+    monkeypatch.setattr(
+        freshness,
+        "latest_release_assets",
+        lambda repo, *, force_refresh = False: {
+            "app-b9518-linux-x64-cuda13-newer.tar.gz": 88_000_000
+        },
+    )
+    st = upd.get_update_status(force_refresh = True)
+    assert st["update_available"] is True
+    assert st["update_size_bytes"] == 88_000_000
+
+
+def test_status_source_build_includes_update_size(monkeypatch, tmp_path):
+    # #6338 P3: a source build offered a prebuilt must carry the asset size too.
+    binary = tmp_path / "llama.cpp" / "build" / "bin" / "llama-server"
+    binary.parent.mkdir(parents = True)
+    binary.write_text("stub")  # no marker -> source build
+    monkeypatch.setattr(upd, "_find_binary", lambda: str(binary))
+    _prebuilt(
+        monkeypatch,
+        repo = "unslothai/llama.cpp",
+        release_tag = "b9585",
+        asset = "app-b9585-linux-x64-cpu.tar.gz",
+    )
+    monkeypatch.setattr(upd, "_installed_build_number", lambda b: None)
+    monkeypatch.setattr(
+        upd,
+        "latest_release_assets",
+        lambda repo, *, force_refresh = False: (
+            {"app-b9585-linux-x64-cpu.tar.gz": 77_000_000}
+            if repo == "unslothai/llama.cpp"
+            else None
+        ),
+    )
+    st = upd.get_update_status()
+    assert st["source_build"] is True
+    assert st["update_available"] is True
+    assert st["update_size_bytes"] == 77_000_000

studio/backend/tests/test_llama_cpp_wait_for_vram_settle.py

@@ -281,6 +281,7 @@ def test_kill_process_records_timestamp_on_actual_kill():
     backend = LlamaCppBackend.__new__(LlamaCppBackend)
     backend._process = None
     backend._healthy = False
+    backend._stats_logger = None  # _kill_process stops it in finally
     backend._stdout_thread = None
     backend._llama_log_fh = None
     backend._last_kill_monotonic = 0.0

studio/backend/tests/test_llama_route.py

@@ -78,6 +78,27 @@ def test_status_response_exposes_source_build():
     rl.LlamaUpdateStatusResponse(**{**payload, "unexpected": 1})
 
 
+def test_status_response_exposes_update_size_bytes():
+    payload = {
+        "supported": True,
+        "update_available": True,
+        "stale": False,
+        "installed_tag": "b9493",
+        "latest_tag": "b9518",
+        "published_repo": "unslothai/llama.cpp",
+        "installed_at_utc": None,
+        "age_days": None,
+        "source_build": False,
+        "update_size_bytes": 123_456_789,
+        "job": {"state": "idle"},
+    }
+    model = rl.LlamaUpdateStatusResponse(**payload)
+    assert model.model_dump()["update_size_bytes"] == 123_456_789
+    # Omitted -> defaults to None (the offline / no-matching-asset case).
+    without = {k: v for k, v in payload.items() if k != "update_size_bytes"}
+    assert rl.LlamaUpdateStatusResponse(**without).model_dump()["update_size_bytes"] is None
+
+
 def test_status_handler_runs_off_event_loop(monkeypatch):
     seen = {}
 

studio/backend/utils/llama_cpp_freshness.py

@@ -33,6 +33,8 @@ _INSTALL_MARKER_NAME = "UNSLOTH_PREBUILT_INFO.json"
 
 _marker_cache: dict[str, Optional[dict]] = {}
 _release_memo: dict[str, tuple[float, Optional[str]]] = {}
+# Newest-release asset sizes (name -> bytes), memoized like the tag (24h TTL).
+_assets_memo: dict[str, tuple[float, dict[str, int]]] = {}
 
 
 def _cache_dir() -> Path:
@@ -180,6 +182,113 @@ def latest_published_release(repo: str, *, force_refresh: bool = False) -> Optio
     return latest
 
 
+def _fetch_latest_release_assets(repo: str, timeout: float = 5.0) -> Optional[dict[str, int]]:
+    """Asset name -> size (bytes) for the newest published release of `repo`,
+    selected exactly like _fetch_latest_release_tag. None on any failure."""
+    import urllib.error
+    import urllib.request
+
+    url = f"https://api.github.com/repos/{repo}/releases?per_page=30"
+    headers = {
+        "Accept": "application/vnd.github+json",
+        "User-Agent": "unsloth-studio-freshness-check",
+    }
+    token = os.environ.get("GITHUB_TOKEN") or os.environ.get("GH_TOKEN")
+    if token:
+        headers["Authorization"] = f"Bearer {token}"
+    req = urllib.request.Request(url, headers = headers)
+    try:
+        with urllib.request.urlopen(req, timeout = timeout) as resp:
+            data = json.loads(resp.read().decode("utf-8"))
+    except (
+        urllib.error.URLError,
+        urllib.error.HTTPError,
+        OSError,
+        json.JSONDecodeError,
+    ) as exc:
+        logger.debug("freshness asset fetch failed", repo = repo, error = str(exc))
+        return None
+    if not isinstance(data, list):
+        return None
+    published = [
+        r
+        for r in data
+        if isinstance(r, dict)
+        and not r.get("draft")
+        and not r.get("prerelease")
+        and isinstance(r.get("tag_name"), str)
+        and r.get("tag_name")
+    ]
+    if not published:
+        return None
+    newest = max(published, key = lambda r: r.get("published_at") or "")
+    assets: dict[str, int] = {}
+    for a in newest.get("assets") or []:
+        name, size = a.get("name"), a.get("size")
+        if isinstance(name, str) and isinstance(size, int):
+            assets[name] = size
+    return assets
+
+
+def latest_release_assets(repo: str, *, force_refresh: bool = False) -> Optional[dict[str, int]]:
+    """Newest-release asset sizes for `repo`, memoized (24h TTL). None when
+    offline and never fetched. In-memory only -- a restart simply re-fetches."""
+    if not repo:
+        return None
+    now = time.time()
+    if not force_refresh:
+        memo = _assets_memo.get(repo)
+        if memo and now - memo[0] < _RELEASE_CACHE_TTL_SECONDS:
+            return memo[1]
+    assets = _fetch_latest_release_assets(repo)
+    if assets is None:
+        memo = _assets_memo.get(repo)
+        return memo[1] if memo else None
+    _assets_memo[repo] = (now, assets)
+    return assets
+
+
+def update_download_size_bytes(
+    marker: Optional[dict],
+    latest_tag: Optional[str],
+    repo: Optional[str],
+    *,
+    force_refresh: bool = False,
+) -> Optional[int]:
+    """Download size of the latest-release asset matching this host's installed
+    bundle (same platform/arch/runtime suffix as the installed asset). None when
+    there is no marker asset, the latest assets can't be read, or no match."""
+    if not marker or not latest_tag or not repo:
+        return None
+    installed_asset = marker.get("asset")
+    if not isinstance(installed_asset, str):
+        return None
+    # Tag-independent platform suffix: accept the fork's "app-*" bundles and the
+    # upstream ggml-org "ubuntu-*"/"win-*" prebuilts ("windows" before "win").
+    m = re.search(r"-((?:linux|ubuntu|windows|win|macos|darwin)-.*)$", installed_asset)
+    if not m:
+        return None
+    suffix = m.group(1)
+    # Upstream ubuntu/win assets live in the marker's binary_repo, not the fork
+    # publish repo; try the publish repo first, then it.
+    repos = [repo]
+    binary_repo = marker.get("binary_repo")
+    if isinstance(binary_repo, str) and binary_repo and binary_repo != repo:
+        repos.append(binary_repo)
+    want = f"app-{latest_tag}-{suffix}"
+    for r in repos:
+        assets = latest_release_assets(r, force_refresh = force_refresh)
+        if not assets:
+            continue
+        if want in assets:
+            return assets[want]
+        # Tag formatting can vary (mix suffixes); fall back to the platform suffix.
+        for name, size in assets.items():
+            if name.endswith(suffix):
+                return size
+    return None
+
+
 def _parse_installed_at(value: object) -> Optional[datetime]:
     if not isinstance(value, str) or not value:
         return None
@@ -313,6 +422,7 @@ def reset_caches(*, drop_disk: bool = False) -> None:
     open (off) instead of pointing at the just-replaced build."""
     _marker_cache.clear()
     _release_memo.clear()
+    _assets_memo.clear()
     if drop_disk:
         import shutil
 

studio/backend/utils/llama_cpp_update.py

@@ -37,9 +37,11 @@ from utils.llama_cpp_freshness import (
     _INSTALL_MARKER_NAME,
     check_prebuilt_freshness,
     latest_published_release,
+    latest_release_assets,
     parse_base_build,
     read_install_marker,
     reset_caches,
+    update_download_size_bytes,
 )
 
 logger = structlog.get_logger(__name__)
@@ -291,6 +293,18 @@ def _source_build_status(binary: str, *, force_refresh: bool) -> Optional[dict]:
         update_available = False
     # Display the mix tag when that's what makes it newer; otherwise the base.
     latest = release_tag if latest_is_mix else base_tag
+    # Size of the resolved prebuilt, so source builds show it like the marker
+    # path. Fails open to None (offline / asset absent from the release).
+    update_size_bytes = None
+    if update_available:
+        asset_name = res.get("asset")
+        if isinstance(asset_name, str) and asset_name:
+            try:
+                assets = latest_release_assets(res.get("repo"), force_refresh = force_refresh)
+                if assets:
+                    update_size_bytes = assets.get(asset_name)
+            except Exception as exc:  # pragma: no cover - network defensive
+                logger.debug("llama update: source-build size lookup failed", error = str(exc))
     with _job_lock:
         job = dict(_job)
     return {
@@ -303,6 +317,7 @@ def _source_build_status(binary: str, *, force_refresh: bool) -> Optional[dict]:
         "installed_at_utc": None,
         "age_days": None,
         "source_build": True,
+        "update_size_bytes": update_size_bytes,
         "job": job,
     }
 
@@ -345,6 +360,20 @@ def get_update_status(*, force_refresh: bool = False) -> dict:
     # (see llama_cpp_freshness.is_behind).
     update_available = bool(freshness.get("has_marker") and freshness.get("behind"))
 
+    # Size of the prebuilt that Update would download, for the banner. Only when
+    # an update is offered; fails open to None (offline / no matching asset).
+    update_size_bytes = None
+    if update_available:
+        try:
+            update_size_bytes = update_download_size_bytes(
+                marker,
+                latest,
+                freshness.get("published_repo") or repo,
+                force_refresh = force_refresh,
+            )
+        except Exception as exc:  # pragma: no cover - network defensive
+            logger.debug("llama update: size lookup failed", error = str(exc))
+
     with _job_lock:
         job = dict(_job)
 
@@ -358,6 +387,7 @@ def get_update_status(*, force_refresh: bool = False) -> dict:
         "installed_at_utc": freshness.get("installed_at_utc"),
         "age_days": freshness.get("age_days"),
         "source_build": False,
+        "update_size_bytes": update_size_bytes,
         "job": job,
     }
 

studio/frontend/src/components/llama-update-banner.tsx

@@ -7,10 +7,7 @@ import { useShowLlamaUpdateBanner } from "@/hooks/use-llama-update-pref";
 import { toast } from "@/lib/toast";
 import { cn } from "@/lib/utils";
 import { Download } from "lucide-react";
-import { AnimatePresence, motion } from "motion/react";
 import { type ReactElement, useEffect, useRef, useState } from "react";
-
-const EASE_OUT_QUART: [number, number, number, number] = [0.165, 0.84, 0.44, 1];
 // Backend progress is coarse (5% steps, ~0.9 max) and the extract tail emits no
 // signal. Creep toward this cap so the bar keeps moving rather than freezing.
 const RUNNING_CAP = 0.95;
@@ -114,6 +111,12 @@ export function LlamaUpdateBanner({
 
   const show =
     visible && status != null && (status.update_available || applying);
+  const sizeBytes = status?.update_size_bytes ?? null;
+  // Round to whole MB; these prebuilts are hundreds of MB.
+  const sizeLabel =
+    sizeBytes && sizeBytes > 0
+      ? `${Math.round(sizeBytes / (1024 * 1024))} MB`
+      : null;
   const updateProgress = status?.job.progress ?? null;
   const jobSucceeded = status?.job.state === "success";
   // Drives the bar so it animates continuously; aria reports the real value.
@@ -123,113 +126,110 @@ export function LlamaUpdateBanner({
     jobSucceeded,
   );
 
-  return (
-    <AnimatePresence>
-      {show ? (
-        <motion.div
-          initial={{ opacity: 0, y: 12, scale: 0.96 }}
-          animate={{ opacity: 1, y: 0, scale: 1 }}
-          exit={{ opacity: 0, y: 8, scale: 0.97 }}
-          transition={{ duration: 0.35, ease: EASE_OUT_QUART }}
-          className={cn(
-            positioned
-              ? "fixed bottom-4 right-4 z-[9998] w-[calc(100vw-2rem)] max-w-[400px]"
-              : "pointer-events-auto w-full",
-          )}
-          data-testid="llama-update-banner"
-        >
-          <div className="relative overflow-hidden rounded-[24px] bg-white px-5 pb-4 pt-5 shadow-[0_2px_8px_-2px_rgba(0,0,0,0.16)] dark:bg-card dark:shadow-[0_8px_28px_-6px_rgba(0,0,0,0.28)]">
-            {applying ? null : (
-              <button
-                type="button"
-                onClick={dismiss}
-                className="absolute top-2.5 right-3 flex size-6 items-center justify-center rounded-full text-muted-foreground/60 transition-colors hover:bg-muted hover:text-foreground"
-                aria-label="Dismiss llama.cpp update notification"
-              >
-                <svg
-                  aria-hidden="true"
-                  width="12"
-                  height="12"
-                  viewBox="0 0 14 14"
-                  fill="none"
-                  xmlns="http://www.w3.org/2000/svg"
-                >
-                  <path
-                    d="M11 3L3 11M3 3l8 8"
-                    stroke="currentColor"
-                    strokeWidth="1.5"
-                    strokeLinecap="round"
-                  />
-                </svg>
-              </button>
-            )}
-
-            <div className="flex min-w-0 items-start gap-4 pr-6">
-              <Download
-                aria-hidden="true"
-                className="mt-1 size-5 shrink-0 text-foreground"
-                strokeWidth={1.75}
+  // Render with no enter/exit animation. An opacity/transform transition (in or
+  // out) promotes a GPU compositing layer whose creation or teardown can flash
+  // for a frame on real displays, which reads as a flicker on appear and on
+  // dismiss. A plain conditional mount appears and leaves cleanly.
+  return show ? (
+    <div
+      className={cn(
+        positioned
+          ? "fixed bottom-4 right-4 z-[9998] w-[calc(100vw-2rem)] max-w-[400px]"
+          : "pointer-events-auto w-full",
+      )}
+      data-testid="llama-update-banner"
+    >
+      <div className="relative overflow-hidden rounded-[24px] bg-white px-5 pb-4 pt-5 shadow-[0_2px_8px_-2px_rgba(0,0,0,0.16)] dark:bg-card dark:shadow-[0_8px_28px_-6px_rgba(0,0,0,0.28)]">
+        {applying ? null : (
+          <button
+            type="button"
+            onClick={dismiss}
+            className="absolute top-2.5 right-3 flex size-6 items-center justify-center rounded-full text-muted-foreground/60 transition-colors hover:bg-muted hover:text-foreground"
+            aria-label="Dismiss llama.cpp update notification"
+          >
+            <svg
+              aria-hidden="true"
+              width="12"
+              height="12"
+              viewBox="0 0 14 14"
+              fill="none"
+              xmlns="http://www.w3.org/2000/svg"
+            >
+              <path
+                d="M11 3L3 11M3 3l8 8"
+                stroke="currentColor"
+                strokeWidth="1.5"
+                strokeLinecap="round"
               />
-              <div className="min-w-0">
-                <p className="font-heading text-base font-medium text-foreground">
-                  {applying ? "Updating llama.cpp..." : "New llama.cpp update"}
-                </p>
-                <p className="mt-0.5 text-xs text-muted-foreground">
-                  {status?.installed_tag ?? "unknown"} &rarr;{" "}
-                  <span className="font-medium text-foreground">
-                    {status?.latest_tag ?? ""}
-                  </span>
-                </p>
-                <p className="mt-1 text-[11px] text-muted-foreground/70">
-                  No restart needed after update
-                </p>
-              </div>
-            </div>
+            </svg>
+          </button>
+        )}
 
-            {applying ? (
-              <div
-                className="mb-1.5 mt-4 h-1 overflow-hidden rounded-full bg-muted"
-                role="progressbar"
-                aria-label="Updating llama.cpp"
-                aria-valuemin={0}
-                aria-valuemax={100}
-                aria-valuenow={
-                  updateProgress != null
-                    ? Math.round(updateProgress * 100)
-                    : Math.round(displayProgress * 100)
-                }
-                data-testid="llama-update-progress"
-              >
-                <div
-                  className="h-full rounded-full bg-primary"
-                  style={{ width: `${Math.max(displayProgress * 100, 2)}%` }}
-                />
-              </div>
-            ) : (
-              <div className="mt-2 flex flex-wrap items-center justify-end gap-x-1 gap-y-2">
-                <Button
-                  size="sm"
-                  variant="ghost"
-                  className="h-auto rounded-full px-3 py-2 text-[13px] font-medium text-foreground"
-                  onClick={snooze}
-                  data-testid="llama-update-snooze-button"
-                >
-                  Remind me later
-                </Button>
-                <Button
-                  size="sm"
-                  // -mr optically aligns the filled pill's edge with the card padding
-                  className="-mr-1 h-auto rounded-full px-3.5 py-2 text-[13px]"
-                  onClick={handleUpdate}
-                  data-testid="llama-update-button"
-                >
-                  Update
-                </Button>
-              </div>
-            )}
+        <div className="flex min-w-0 items-start gap-4 pr-6">
+          <Download
+            aria-hidden="true"
+            className="mt-1 size-5 shrink-0 text-foreground"
+            strokeWidth={1.75}
+          />
+          <div className="min-w-0">
+            <p className="font-heading text-base font-medium text-foreground">
+              {applying ? "Updating llama.cpp..." : "New llama.cpp update"}
+            </p>
+            <p className="mt-0.5 text-xs text-muted-foreground">
+              {status?.installed_tag ?? "unknown"} &rarr;{" "}
+              <span className="font-medium text-foreground">
+                {status?.latest_tag ?? ""}
+              </span>
+            </p>
+            <p className="mt-1 text-[11px] text-muted-foreground/70">
+              {sizeLabel ? `${sizeLabel} download · ` : ""}No restart needed
+              after update
+            </p>
           </div>
-        </motion.div>
-      ) : null}
-    </AnimatePresence>
-  );
+        </div>
+
+        {applying ? (
+          <div
+            className="mb-1.5 mt-4 h-1 overflow-hidden rounded-full bg-muted"
+            role="progressbar"
+            aria-label="Updating llama.cpp"
+            aria-valuemin={0}
+            aria-valuemax={100}
+            aria-valuenow={
+              updateProgress != null
+                ? Math.round(updateProgress * 100)
+                : Math.round(displayProgress * 100)
+            }
+            data-testid="llama-update-progress"
+          >
+            <div
+              className="h-full rounded-full bg-primary"
+              style={{ width: `${Math.max(displayProgress * 100, 2)}%` }}
+            />
+          </div>
+        ) : (
+          <div className="mt-2 flex flex-wrap items-center justify-end gap-x-1 gap-y-2">
+            <Button
+              size="sm"
+              variant="ghost"
+              className="h-auto rounded-full px-3 py-2 text-[13px] font-medium text-foreground"
+              onClick={snooze}
+              data-testid="llama-update-snooze-button"
+            >
+              Remind me later
+            </Button>
+            <Button
+              size="sm"
+              // -mr optically aligns the filled pill's edge with the card padding
+              className="-mr-1 h-auto rounded-full px-3.5 py-2 text-[13px]"
+              onClick={handleUpdate}
+              data-testid="llama-update-button"
+            >
+              Update
+            </Button>
+          </div>
+        )}
+      </div>
+    </div>
+  ) : null;
 }

studio/frontend/src/hooks/use-llama-update-check.ts

@@ -31,6 +31,8 @@ export interface LlamaUpdateStatus {
   update_available: boolean;
   installed_tag: string | null;
   latest_tag: string | null;
+  // Download size of the prebuilt Update would fetch, in bytes (null if unknown).
+  update_size_bytes: number | null;
   job: LlamaUpdateJob;
 }
 
@@ -43,6 +45,8 @@ function parseStatus(value: unknown): LlamaUpdateStatus | null {
     update_available: s.update_available === true,
     installed_tag: typeof s.installed_tag === "string" ? s.installed_tag : null,
     latest_tag: typeof s.latest_tag === "string" ? s.latest_tag : null,
+    update_size_bytes:
+      typeof s.update_size_bytes === "number" ? s.update_size_bytes : null,
     job: {
       state: (job.state as LlamaUpdateJob["state"]) ?? "idle",
       message: typeof job.message === "string" ? job.message : "",

tests/python/test_no_torch_filtering.py

@@ -471,11 +471,15 @@ class TestInstallPythonStackSubprocessMock:
     # -- Normal Linux path (NO_TORCH=False, IS_MACOS=False, IS_WINDOWS=False) --
 
     def test_normal_linux_includes_overrides(self):
-        """Normal Linux: overrides.txt IS called."""
+        """Normal Linux: the torchao override step IS called.
+
+        The override step installs a torch-matched torchao spec via
+        --force-reinstall (uv: --reinstall), not overrides.txt directly.
+        """
         cmds = self._capture_install(no_torch = False, is_macos = False, is_windows = False)
-        assert self._cmds_contain_file(
-            cmds, "overrides.txt"
-        ), "overrides.txt should be called on normal Linux"
+        assert any(
+            "--reinstall" in cmd for cmd in cmds
+        ), "torchao override step (--reinstall) should be called on normal Linux"
 
     def test_normal_linux_includes_triton(self):
         """Normal Linux: triton-kernels.txt IS called."""

tests/studio/playwright_chat_ui.py

@@ -1430,7 +1430,16 @@ with sync_playwright() as p:
     # Re-login through the UI with NEW2 so the browser has a valid
     # access token for the /api/shutdown call (the previous one
     # was invalidated by the CLI rotation above).
-    page.goto(f"{BASE}/login")
+    # The CLI rotation left a stale token, so the SPA auth guard can
+    # client-side-redirect mid-navigation and abort this goto with
+    # net::ERR_ABORTED. Resolve on domcontentloaded and tolerate the
+    # abort; the password-field wait below confirms we reached /login.
+    try:
+        page.goto(f"{BASE}/login", wait_until = "domcontentloaded", timeout = 60_000)
+    except Exception as exc:
+        if "ERR_ABORTED" not in str(exc):
+            raise
+        info(f"goto /login aborted ({exc!r}); password-field wait will confirm /login")
     pw_field = page.locator("#password")
     pw_field.wait_for(state = "visible", timeout = 60_000)
     pw_field.fill(NEW2)