feat: lower memory dedupe threshold for more aggressive cleaning

This commit is contained in:
root
2026-07-07 16:45:30 +02:00
parent c92f238d2d
commit 97be0f1d00
68 changed files with 15783 additions and 15783 deletions
+68 -68
View File
@@ -1,68 +1,68 @@
import { api } from "@/lib/api"
import { useJobs, useQueryClient, qk } from "@/lib/queries"
import { useDialog } from "@/lib/useDialog"
import { cn } from "@/lib/utils"
import { fmtBytes, fmtEta } from "@/lib/format"
export function JobsBar() {
const qc = useQueryClient()
const { data: jobs = [] } = useJobs(2_000)
const { showAlert, dialogElement } = useDialog()
async function cancelJob(jobId: string) {
try {
await api(`/api/jobs/${jobId}/cancel`, { method: "POST" })
qc.invalidateQueries({ queryKey: qk.jobs })
} catch (e: any) {
showAlert("Fehler", e.message)
}
}
const active = jobs.filter((j) => j.state === "running" || j.state === "queued")
const recent = jobs.filter((j) => j.state !== "running" && j.state !== "queued").slice(-3)
if (active.length === 0 && recent.length === 0) return null
return (
<div className="space-y-3 rounded-2xl border border-border/60 bg-card/45 backdrop-blur-md p-4 shadow-lg shadow-black/10">
<div className="text-[11px] font-semibold text-muted-foreground uppercase tracking-wider">Aktive Downloads</div>
{active.map((j) => (
<div key={j.id} className="space-y-1.5 p-3 rounded-xl bg-background/20 border border-border/40">
<div className="flex justify-between items-center text-xs">
<span className="font-semibold truncate max-w-[250px]">{j.label}</span>
<div className="flex items-center gap-3">
<span className="text-muted-foreground font-mono">
{j.progress ?? 0}% {fmtBytes(j.done_bytes)}/{fmtBytes(j.total_bytes)}
{j.eta_s ? ` • ETA ${fmtEta(j.eta_s)}` : ""}
</span>
<button
onClick={() => cancelJob(j.id)}
className="text-[10px] text-red-400 hover:text-red-300 font-semibold border border-red-500/25 bg-red-500/5 px-2 py-0.5 rounded transition-all cursor-pointer"
>
Abbrechen
</button>
</div>
</div>
<div className="h-1.5 overflow-hidden rounded-full bg-muted">
<div className="h-full rounded-full bg-primary transition-all duration-500" style={{ width: `${j.progress ?? 0}%` }} />
</div>
</div>
))}
{recent.map((j) => (
<div key={j.id} className="flex justify-between items-center text-xs text-muted-foreground px-1">
<span className="truncate">{j.label}</span>
<span className={cn(
"font-semibold text-[10px] px-1.5 py-0.5 rounded uppercase font-mono",
j.state === "done" ? "bg-emerald-500/10 text-emerald-400" : "bg-amber-500/10 text-amber-400"
)}>
{j.state}
</span>
</div>
))}
{dialogElement}
</div>
)
}
import { api } from "@/lib/api"
import { useJobs, useQueryClient, qk } from "@/lib/queries"
import { useDialog } from "@/lib/useDialog"
import { cn } from "@/lib/utils"
import { fmtBytes, fmtEta } from "@/lib/format"
export function JobsBar() {
const qc = useQueryClient()
const { data: jobs = [] } = useJobs(2_000)
const { showAlert, dialogElement } = useDialog()
async function cancelJob(jobId: string) {
try {
await api(`/api/jobs/${jobId}/cancel`, { method: "POST" })
qc.invalidateQueries({ queryKey: qk.jobs })
} catch (e: any) {
showAlert("Fehler", e.message)
}
}
const active = jobs.filter((j) => j.state === "running" || j.state === "queued")
const recent = jobs.filter((j) => j.state !== "running" && j.state !== "queued").slice(-3)
if (active.length === 0 && recent.length === 0) return null
return (
<div className="space-y-3 rounded-2xl border border-border/60 bg-card/45 backdrop-blur-md p-4 shadow-lg shadow-black/10">
<div className="text-[11px] font-semibold text-muted-foreground uppercase tracking-wider">Aktive Downloads</div>
{active.map((j) => (
<div key={j.id} className="space-y-1.5 p-3 rounded-xl bg-background/20 border border-border/40">
<div className="flex justify-between items-center text-xs">
<span className="font-semibold truncate max-w-[250px]">{j.label}</span>
<div className="flex items-center gap-3">
<span className="text-muted-foreground font-mono">
{j.progress ?? 0}% {fmtBytes(j.done_bytes)}/{fmtBytes(j.total_bytes)}
{j.eta_s ? ` • ETA ${fmtEta(j.eta_s)}` : ""}
</span>
<button
onClick={() => cancelJob(j.id)}
className="text-[10px] text-red-400 hover:text-red-300 font-semibold border border-red-500/25 bg-red-500/5 px-2 py-0.5 rounded transition-all cursor-pointer"
>
Abbrechen
</button>
</div>
</div>
<div className="h-1.5 overflow-hidden rounded-full bg-muted">
<div className="h-full rounded-full bg-primary transition-all duration-500" style={{ width: `${j.progress ?? 0}%` }} />
</div>
</div>
))}
{recent.map((j) => (
<div key={j.id} className="flex justify-between items-center text-xs text-muted-foreground px-1">
<span className="truncate">{j.label}</span>
<span className={cn(
"font-semibold text-[10px] px-1.5 py-0.5 rounded uppercase font-mono",
j.state === "done" ? "bg-emerald-500/10 text-emerald-400" : "bg-amber-500/10 text-amber-400"
)}>
{j.state}
</span>
</div>
))}
{dialogElement}
</div>
)
}
+60 -60
View File
@@ -1,60 +1,60 @@
import { type Fit } from "@/lib/api"
import { cn } from "@/lib/utils"
import { ROLES, roleTone, roleMeta } from "@/lib/roleMeta"
// Klartext-Schicht: Rollen-Namen/Icons/Farben leben jetzt zentral in lib/roleMeta.ts.
// Re-Export hält bestehende Importe (Cockpit, ModelsCard, …) stabil.
export { ROLES, roleTone, roleMeta }
// Rollen-Badge in Klartext: Icon + Zweck-Name, technisches Kürzel als Tooltip.
// `dense` = nur Kurzform (für enge Badges).
export function RoleLabel({
role,
dense = false,
className,
}: {
role?: string | null
dense?: boolean
className?: string
}) {
const meta = roleMeta(role)
const Icon = meta.icon
return (
<span
className={cn(
"inline-flex items-center gap-1 rounded px-1.5 py-0.5 text-[10px] font-semibold border tracking-wide",
meta.tone,
className,
)}
title={`${meta.desc} (${meta.role})`}
>
<Icon className="h-3 w-3 shrink-0" aria-hidden="true" />
{dense ? meta.short : meta.label}
</span>
)
}
export function FitBadge({ fit }: { fit: Fit }) {
const tone = {
perfect: "bg-emerald-500/15 text-emerald-400 border border-emerald-500/20",
marginal: "bg-amber-500/15 text-amber-400 border border-amber-500/20",
too_tight: "bg-red-500/15 text-red-400 border border-red-500/20",
}[fit.level]
return (
<span className={cn("rounded px-2 py-0.5 text-[10px] font-semibold uppercase tracking-wider font-mono", tone)}>
{fit.text} {fit.req_gb} GB RAM
</span>
)
}
export function getBrandInfo(name: string) {
const low = name.toLowerCase()
if (low.includes("qwen")) return { name: "Qwen", color: "bg-purple-500/20 text-purple-300 border-purple-500/30", initial: "Q" }
if (low.includes("gemma")) return { name: "Gemma", color: "bg-blue-500/20 text-blue-300 border-blue-500/30", initial: "G" }
if (low.includes("llama")) return { name: "Llama", color: "bg-red-500/20 text-red-300 border-red-500/30", initial: "🦙" }
if (low.includes("mistral") || low.includes("mixtral")) return { name: "Mistral", color: "bg-orange-500/20 text-orange-300 border-orange-500/30", initial: "M" }
if (low.includes("deepseek")) return { name: "DeepSeek", color: "bg-cyan-500/20 text-cyan-300 border-cyan-500/30", initial: "D" }
if (low.includes("hermes") || low.includes("nous")) return { name: "Hermes", color: "bg-amber-500/20 text-amber-300 border-amber-500/30", initial: "H" }
if (low.includes("phi")) return { name: "Phi", color: "bg-emerald-500/20 text-emerald-300 border-emerald-500/30", initial: "Φ" }
return { name: "Other", color: "bg-slate-500/20 text-slate-300 border-slate-500/30", initial: "AI" }
}
import { type Fit } from "@/lib/api"
import { cn } from "@/lib/utils"
import { ROLES, roleTone, roleMeta } from "@/lib/roleMeta"
// Klartext-Schicht: Rollen-Namen/Icons/Farben leben jetzt zentral in lib/roleMeta.ts.
// Re-Export hält bestehende Importe (Cockpit, ModelsCard, …) stabil.
export { ROLES, roleTone, roleMeta }
// Rollen-Badge in Klartext: Icon + Zweck-Name, technisches Kürzel als Tooltip.
// `dense` = nur Kurzform (für enge Badges).
export function RoleLabel({
role,
dense = false,
className,
}: {
role?: string | null
dense?: boolean
className?: string
}) {
const meta = roleMeta(role)
const Icon = meta.icon
return (
<span
className={cn(
"inline-flex items-center gap-1 rounded px-1.5 py-0.5 text-[10px] font-semibold border tracking-wide",
meta.tone,
className,
)}
title={`${meta.desc} (${meta.role})`}
>
<Icon className="h-3 w-3 shrink-0" aria-hidden="true" />
{dense ? meta.short : meta.label}
</span>
)
}
export function FitBadge({ fit }: { fit: Fit }) {
const tone = {
perfect: "bg-emerald-500/15 text-emerald-400 border border-emerald-500/20",
marginal: "bg-amber-500/15 text-amber-400 border border-amber-500/20",
too_tight: "bg-red-500/15 text-red-400 border border-red-500/20",
}[fit.level]
return (
<span className={cn("rounded px-2 py-0.5 text-[10px] font-semibold uppercase tracking-wider font-mono", tone)}>
{fit.text} {fit.req_gb} GB RAM
</span>
)
}
export function getBrandInfo(name: string) {
const low = name.toLowerCase()
if (low.includes("qwen")) return { name: "Qwen", color: "bg-purple-500/20 text-purple-300 border-purple-500/30", initial: "Q" }
if (low.includes("gemma")) return { name: "Gemma", color: "bg-blue-500/20 text-blue-300 border-blue-500/30", initial: "G" }
if (low.includes("llama")) return { name: "Llama", color: "bg-red-500/20 text-red-300 border-red-500/30", initial: "🦙" }
if (low.includes("mistral") || low.includes("mixtral")) return { name: "Mistral", color: "bg-orange-500/20 text-orange-300 border-orange-500/30", initial: "M" }
if (low.includes("deepseek")) return { name: "DeepSeek", color: "bg-cyan-500/20 text-cyan-300 border-cyan-500/30", initial: "D" }
if (low.includes("hermes") || low.includes("nous")) return { name: "Hermes", color: "bg-amber-500/20 text-amber-300 border-amber-500/30", initial: "H" }
if (low.includes("phi")) return { name: "Phi", color: "bg-emerald-500/20 text-emerald-300 border-emerald-500/30", initial: "Φ" }
return { name: "Other", color: "bg-slate-500/20 text-slate-300 border-slate-500/30", initial: "AI" }
}
+203 -203
View File
@@ -1,203 +1,203 @@
import { useState } from "react"
import { Zap, Plus, X, Check, HardDrive, AlertTriangle } from "lucide-react"
import { setGroup, type ModelInfo } from "@/lib/api"
import { useModels, useGroups, useSystemStatus, useHermesBrain, useQueryClient, qk } from "@/lib/queries"
import { useDialog } from "@/lib/useDialog"
import { RoleLabel, roleMeta } from "@/components/models/ModelBadges"
import { fmtSize } from "@/lib/format"
import { cn } from "@/lib/utils"
// „Immer-bereit-Set" = die ko-residente llama-swap-Gruppe `brains`. Ihre Mitglieder
// bleiben gemeinsam warm (verdrängen sich nicht). Hier in Klartext verwaltbar,
// mit Speicher-Budget und Geländer (Hirn/Gedächtnis nicht entfernbar).
export function WarmSetManager() {
const qc = useQueryClient()
const { data: modelsResp } = useModels(4_000)
const { data: groupsResp } = useGroups()
const { data: sysStatus } = useSystemStatus()
const { data: brain } = useHermesBrain()
const { showAlert, dialogElement } = useDialog()
const [picking, setPicking] = useState(false)
const [busy, setBusy] = useState(false)
const models = modelsResp?.models ?? []
const running = modelsResp?.running ?? []
const group = groupsResp?.groups?.brains
const members = group?.members ?? []
const byName = (name: string) => models.find((m) => m.name === name)
const shortName = (name: string) => name.split("/").pop()?.replace(/\.gguf$/i, "") || name
const memberModels = members.map(byName).filter(Boolean) as ModelInfo[]
const B = 1024 ** 3
const bud = brain?.budget
const weightsGb = memberModels.reduce((a, m) => a + (m.size_bytes || 0), 0) / B
const gttTotal = sysStatus?.gpu?.gtt_total || sysStatus?.gpu?.vram_total || 0
const totalGb = bud?.gtt_gb || (gttTotal ? gttTotal / B : 0)
// Ehrlicher Fußabdruck inkl. Arbeitsspeicher (KV-Cache) aus dem Budget; sonst nur Gewichte.
const reservedGb = bud?.warm_projected_gb ?? weightsGb
const pct = totalGb > 0 ? Math.min(100, (reservedGb / totalGb) * 100) : 0
const eligible = models.filter((m) => !members.includes(m.name) && !m.incomplete)
async function save(next: string[]) {
setBusy(true)
try {
await setGroup("brains", next, group?.swap ?? false, group?.persist ?? true)
qc.invalidateQueries({ queryKey: qk.groups })
qc.invalidateQueries({ queryKey: qk.models })
qc.invalidateQueries({ queryKey: qk.hermesBrain })
} catch (e: any) {
showAlert("Fehler", `Immer-bereit-Set konnte nicht geändert werden: ${e?.message || e}`)
} finally {
setBusy(false)
}
}
function remove(name: string) {
const m = byName(name)
if (m && roleMeta(m.role).protected) {
showAlert("Geschützt", `${roleMeta(m.role).label}" ist lebenswichtig und bleibt immer bereit.`)
return
}
void save(members.filter((x) => x !== name))
}
function add(name: string) {
setPicking(false)
void save([...members, name])
}
if (!group) {
return (
<div className="rounded-2xl border border-border/60 bg-card/45 backdrop-blur-md p-5 shadow-lg shadow-black/10">
<Header />
<p className="mt-3 text-xs text-muted-foreground">
Es gibt noch kein Immer-bereit-Set. Es entsteht automatisch, sobald Lucys Hirn zugewiesen ist.
</p>
{dialogElement}
</div>
)
}
return (
<div className="rounded-2xl border border-border/60 bg-card/45 backdrop-blur-md p-5 shadow-lg shadow-black/10 space-y-4">
<Header />
{/* Mitglieder */}
<div className="space-y-1.5">
{memberModels.length === 0 && (
<div className="text-xs text-muted-foreground">Noch keine Modelle im Set.</div>
)}
{memberModels.map((m) => {
const isWarm = running.includes(m.name)
const prot = roleMeta(m.role).protected
return (
<div key={m.name} className="flex items-center justify-between gap-2 rounded-xl border border-border/30 bg-background/20 p-2.5">
<div className="flex min-w-0 items-center gap-2.5">
<span className={cn("h-2 w-2 shrink-0 rounded-full ring-2 ring-black/40", isWarm ? "bg-emerald-500 animate-pulse" : "bg-muted-foreground/40")}
title={isWarm ? "Gerade warm (bereit)" : "Reserviert — lädt bei Bedarf sofort"} />
{m.role && <RoleLabel role={m.role} dense className="shrink-0" />}
<span className="truncate font-mono text-xs font-semibold text-foreground" title={m.name}>{shortName(m.name)}</span>
<span className="shrink-0 font-mono text-[10px] text-muted-foreground/70">{fmtSize(m.size_bytes)}</span>
</div>
{prot ? (
<span className="shrink-0 text-[10px] font-semibold text-muted-foreground" title="Lebenswichtig — bleibt immer im Set.">🔒</span>
) : (
<button
onClick={() => remove(m.name)}
disabled={busy}
className="flex h-7 w-7 shrink-0 items-center justify-center rounded-lg border border-border/40 text-muted-foreground transition-colors hover:bg-red-500/5 hover:text-red-400 cursor-pointer disabled:opacity-50"
title="Aus dem Immer-bereit-Set nehmen (lädt dann nur noch bei Bedarf)"
>
<X className="h-3.5 w-3.5" />
</button>
)}
</div>
)
})}
</div>
{/* Hinzufügen */}
<button
onClick={() => setPicking(true)}
disabled={busy || eligible.length === 0}
className="flex h-8 w-full items-center justify-center gap-1.5 rounded-lg border border-dashed border-border/50 text-[11px] font-semibold uppercase tracking-wide text-muted-foreground transition-colors hover:border-primary/40 hover:text-primary cursor-pointer disabled:opacity-50"
>
<Plus className="h-3.5 w-3.5" /> Modell dauerhaft bereithalten
</button>
{/* Speicher-Budget */}
<div className="rounded-xl border border-border/30 bg-background/20 p-3">
<div className="mb-1.5 flex items-center justify-between text-[11px]">
<span className="flex items-center gap-1.5 font-semibold uppercase tracking-wider text-muted-foreground" title="Modellgewichte + KV-Cache, berechnet aus den echten Architektur-Daten jedes Modells (Layer × KV-Köpfe × Kontext) und seiner KV-Quantisierung. So viel legt llama.cpp beim Laden wirklich für den eingestellten Kontext an — kein Aufschlag mehr.">
<HardDrive className="h-3.5 w-3.5" /> Reserviert fürs Set
</span>
<span className="font-mono font-bold text-foreground">
{reservedGb.toFixed(1)}{totalGb > 0 ? ` / ${totalGb.toFixed(0)}` : ""} GB
</span>
</div>
<div className="h-2 w-full overflow-hidden rounded-full border border-border/30 bg-background/50">
<div className={cn("h-full rounded-full transition-all duration-500", pct >= 85 ? "bg-red-500" : pct >= 65 ? "bg-amber-500" : "bg-teal-500")} style={{ width: `${pct}%` }} />
</div>
{bud && !bud.fits && (
<p className="mt-2 flex items-start gap-1.5 text-[11px] text-amber-400">
<AlertTriangle className="mt-0.5 h-3.5 w-3.5 shrink-0" />
<span>
Zusammen mit dem größten Gelegenheits-Modell (~{bud.largest_ondemand_gb} GB) wird der Speicher knapp. Das Set
bleibt dabei immer geladen wird es wirklich eng, schlägt das Laden des großen Modells fehl (es wartet dann,
statt das Set zu verdrängen).
</span>
</p>
)}
{bud && bud.fits && (
<p className="mt-2 flex items-center gap-1.5 text-[11px] text-emerald-400">
<Check className="h-3.5 w-3.5 shrink-0" /> Passt auch neben dem größten Gelegenheits-Modell nichts wird verdrängt.
</p>
)}
</div>
{/* Auswahl-Modal */}
{picking && (
<div className="fixed inset-0 z-50 flex items-center justify-center bg-black/60 p-4 backdrop-blur-sm" role="dialog" aria-modal="true" aria-label="Modell zum Immer-bereit-Set hinzufügen">
<div className="w-full max-w-md space-y-3 rounded-2xl border border-border/80 bg-card p-6 shadow-2xl animate-in fade-in zoom-in-95 duration-200">
<div className="flex items-center justify-between border-b border-border/20 pb-2">
<h3 className="text-xs font-bold uppercase tracking-wider text-primary">Dauerhaft bereithalten</h3>
<button onClick={() => setPicking(false)} className="rounded p-1 text-muted-foreground hover:bg-accent hover:text-foreground cursor-pointer"><X className="h-4 w-4" /></button>
</div>
<p className="text-xs text-muted-foreground">Wähle ein Modell, das ohne Ladezeit bereitstehen soll:</p>
<div className="max-h-60 space-y-1.5 overflow-y-auto pr-1">
{eligible.map((m) => (
<button
key={m.name}
onClick={() => add(m.name)}
className="flex w-full items-center justify-between gap-2 rounded-lg border border-border/30 bg-background/20 px-3 py-2.5 text-left transition-colors hover:bg-accent cursor-pointer"
>
<span className="flex min-w-0 items-center gap-2">
{m.role && <RoleLabel role={m.role} dense className="shrink-0" />}
<span className="truncate font-mono text-xs font-semibold text-foreground">{shortName(m.name)}</span>
</span>
<span className="shrink-0 font-mono text-[10px] text-muted-foreground/70">{fmtSize(m.size_bytes)}</span>
</button>
))}
</div>
</div>
</div>
)}
{dialogElement}
</div>
)
}
function Header() {
return (
<div className="flex items-center gap-2">
<Zap className="h-4.5 w-4.5 text-primary" />
<div>
<h2 className="text-sm font-bold uppercase tracking-wide text-foreground">Immer-bereit-Set</h2>
<p className="text-[11px] text-muted-foreground">Diese Modelle hält Lucy immer bereit sie antworten ohne Ladezeit.</p>
</div>
</div>
)
}
import { useState } from "react"
import { Zap, Plus, X, Check, HardDrive, AlertTriangle } from "lucide-react"
import { setGroup, type ModelInfo } from "@/lib/api"
import { useModels, useGroups, useSystemStatus, useHermesBrain, useQueryClient, qk } from "@/lib/queries"
import { useDialog } from "@/lib/useDialog"
import { RoleLabel, roleMeta } from "@/components/models/ModelBadges"
import { fmtSize } from "@/lib/format"
import { cn } from "@/lib/utils"
// „Immer-bereit-Set" = die ko-residente llama-swap-Gruppe `brains`. Ihre Mitglieder
// bleiben gemeinsam warm (verdrängen sich nicht). Hier in Klartext verwaltbar,
// mit Speicher-Budget und Geländer (Hirn/Gedächtnis nicht entfernbar).
export function WarmSetManager() {
const qc = useQueryClient()
const { data: modelsResp } = useModels(4_000)
const { data: groupsResp } = useGroups()
const { data: sysStatus } = useSystemStatus()
const { data: brain } = useHermesBrain()
const { showAlert, dialogElement } = useDialog()
const [picking, setPicking] = useState(false)
const [busy, setBusy] = useState(false)
const models = modelsResp?.models ?? []
const running = modelsResp?.running ?? []
const group = groupsResp?.groups?.brains
const members = group?.members ?? []
const byName = (name: string) => models.find((m) => m.name === name)
const shortName = (name: string) => name.split("/").pop()?.replace(/\.gguf$/i, "") || name
const memberModels = members.map(byName).filter(Boolean) as ModelInfo[]
const B = 1024 ** 3
const bud = brain?.budget
const weightsGb = memberModels.reduce((a, m) => a + (m.size_bytes || 0), 0) / B
const gttTotal = sysStatus?.gpu?.gtt_total || sysStatus?.gpu?.vram_total || 0
const totalGb = bud?.gtt_gb || (gttTotal ? gttTotal / B : 0)
// Ehrlicher Fußabdruck inkl. Arbeitsspeicher (KV-Cache) aus dem Budget; sonst nur Gewichte.
const reservedGb = bud?.warm_projected_gb ?? weightsGb
const pct = totalGb > 0 ? Math.min(100, (reservedGb / totalGb) * 100) : 0
const eligible = models.filter((m) => !members.includes(m.name) && !m.incomplete)
async function save(next: string[]) {
setBusy(true)
try {
await setGroup("brains", next, group?.swap ?? false, group?.persist ?? true)
qc.invalidateQueries({ queryKey: qk.groups })
qc.invalidateQueries({ queryKey: qk.models })
qc.invalidateQueries({ queryKey: qk.hermesBrain })
} catch (e: any) {
showAlert("Fehler", `Immer-bereit-Set konnte nicht geändert werden: ${e?.message || e}`)
} finally {
setBusy(false)
}
}
function remove(name: string) {
const m = byName(name)
if (m && roleMeta(m.role).protected) {
showAlert("Geschützt", `${roleMeta(m.role).label}" ist lebenswichtig und bleibt immer bereit.`)
return
}
void save(members.filter((x) => x !== name))
}
function add(name: string) {
setPicking(false)
void save([...members, name])
}
if (!group) {
return (
<div className="rounded-2xl border border-border/60 bg-card/45 backdrop-blur-md p-5 shadow-lg shadow-black/10">
<Header />
<p className="mt-3 text-xs text-muted-foreground">
Es gibt noch kein Immer-bereit-Set. Es entsteht automatisch, sobald Lucys Hirn zugewiesen ist.
</p>
{dialogElement}
</div>
)
}
return (
<div className="rounded-2xl border border-border/60 bg-card/45 backdrop-blur-md p-5 shadow-lg shadow-black/10 space-y-4">
<Header />
{/* Mitglieder */}
<div className="space-y-1.5">
{memberModels.length === 0 && (
<div className="text-xs text-muted-foreground">Noch keine Modelle im Set.</div>
)}
{memberModels.map((m) => {
const isWarm = running.includes(m.name)
const prot = roleMeta(m.role).protected
return (
<div key={m.name} className="flex items-center justify-between gap-2 rounded-xl border border-border/30 bg-background/20 p-2.5">
<div className="flex min-w-0 items-center gap-2.5">
<span className={cn("h-2 w-2 shrink-0 rounded-full ring-2 ring-black/40", isWarm ? "bg-emerald-500 animate-pulse" : "bg-muted-foreground/40")}
title={isWarm ? "Gerade warm (bereit)" : "Reserviert — lädt bei Bedarf sofort"} />
{m.role && <RoleLabel role={m.role} dense className="shrink-0" />}
<span className="truncate font-mono text-xs font-semibold text-foreground" title={m.name}>{shortName(m.name)}</span>
<span className="shrink-0 font-mono text-[10px] text-muted-foreground/70">{fmtSize(m.size_bytes)}</span>
</div>
{prot ? (
<span className="shrink-0 text-[10px] font-semibold text-muted-foreground" title="Lebenswichtig — bleibt immer im Set.">🔒</span>
) : (
<button
onClick={() => remove(m.name)}
disabled={busy}
className="flex h-7 w-7 shrink-0 items-center justify-center rounded-lg border border-border/40 text-muted-foreground transition-colors hover:bg-red-500/5 hover:text-red-400 cursor-pointer disabled:opacity-50"
title="Aus dem Immer-bereit-Set nehmen (lädt dann nur noch bei Bedarf)"
>
<X className="h-3.5 w-3.5" />
</button>
)}
</div>
)
})}
</div>
{/* Hinzufügen */}
<button
onClick={() => setPicking(true)}
disabled={busy || eligible.length === 0}
className="flex h-8 w-full items-center justify-center gap-1.5 rounded-lg border border-dashed border-border/50 text-[11px] font-semibold uppercase tracking-wide text-muted-foreground transition-colors hover:border-primary/40 hover:text-primary cursor-pointer disabled:opacity-50"
>
<Plus className="h-3.5 w-3.5" /> Modell dauerhaft bereithalten
</button>
{/* Speicher-Budget */}
<div className="rounded-xl border border-border/30 bg-background/20 p-3">
<div className="mb-1.5 flex items-center justify-between text-[11px]">
<span className="flex items-center gap-1.5 font-semibold uppercase tracking-wider text-muted-foreground" title="Modellgewichte + KV-Cache, berechnet aus den echten Architektur-Daten jedes Modells (Layer × KV-Köpfe × Kontext) und seiner KV-Quantisierung. So viel legt llama.cpp beim Laden wirklich für den eingestellten Kontext an — kein Aufschlag mehr.">
<HardDrive className="h-3.5 w-3.5" /> Reserviert fürs Set
</span>
<span className="font-mono font-bold text-foreground">
{reservedGb.toFixed(1)}{totalGb > 0 ? ` / ${totalGb.toFixed(0)}` : ""} GB
</span>
</div>
<div className="h-2 w-full overflow-hidden rounded-full border border-border/30 bg-background/50">
<div className={cn("h-full rounded-full transition-all duration-500", pct >= 85 ? "bg-red-500" : pct >= 65 ? "bg-amber-500" : "bg-teal-500")} style={{ width: `${pct}%` }} />
</div>
{bud && !bud.fits && (
<p className="mt-2 flex items-start gap-1.5 text-[11px] text-amber-400">
<AlertTriangle className="mt-0.5 h-3.5 w-3.5 shrink-0" />
<span>
Zusammen mit dem größten Gelegenheits-Modell (~{bud.largest_ondemand_gb} GB) wird der Speicher knapp. Das Set
bleibt dabei immer geladen wird es wirklich eng, schlägt das Laden des großen Modells fehl (es wartet dann,
statt das Set zu verdrängen).
</span>
</p>
)}
{bud && bud.fits && (
<p className="mt-2 flex items-center gap-1.5 text-[11px] text-emerald-400">
<Check className="h-3.5 w-3.5 shrink-0" /> Passt auch neben dem größten Gelegenheits-Modell nichts wird verdrängt.
</p>
)}
</div>
{/* Auswahl-Modal */}
{picking && (
<div className="fixed inset-0 z-50 flex items-center justify-center bg-black/60 p-4 backdrop-blur-sm" role="dialog" aria-modal="true" aria-label="Modell zum Immer-bereit-Set hinzufügen">
<div className="w-full max-w-md space-y-3 rounded-2xl border border-border/80 bg-card p-6 shadow-2xl animate-in fade-in zoom-in-95 duration-200">
<div className="flex items-center justify-between border-b border-border/20 pb-2">
<h3 className="text-xs font-bold uppercase tracking-wider text-primary">Dauerhaft bereithalten</h3>
<button onClick={() => setPicking(false)} className="rounded p-1 text-muted-foreground hover:bg-accent hover:text-foreground cursor-pointer"><X className="h-4 w-4" /></button>
</div>
<p className="text-xs text-muted-foreground">Wähle ein Modell, das ohne Ladezeit bereitstehen soll:</p>
<div className="max-h-60 space-y-1.5 overflow-y-auto pr-1">
{eligible.map((m) => (
<button
key={m.name}
onClick={() => add(m.name)}
className="flex w-full items-center justify-between gap-2 rounded-lg border border-border/30 bg-background/20 px-3 py-2.5 text-left transition-colors hover:bg-accent cursor-pointer"
>
<span className="flex min-w-0 items-center gap-2">
{m.role && <RoleLabel role={m.role} dense className="shrink-0" />}
<span className="truncate font-mono text-xs font-semibold text-foreground">{shortName(m.name)}</span>
</span>
<span className="shrink-0 font-mono text-[10px] text-muted-foreground/70">{fmtSize(m.size_bytes)}</span>
</button>
))}
</div>
</div>
</div>
)}
{dialogElement}
</div>
)
}
function Header() {
return (
<div className="flex items-center gap-2">
<Zap className="h-4.5 w-4.5 text-primary" />
<div>
<h2 className="text-sm font-bold uppercase tracking-wide text-foreground">Immer-bereit-Set</h2>
<p className="text-[11px] text-muted-foreground">Diese Modelle hält Lucy immer bereit sie antworten ohne Ladezeit.</p>
</div>
</div>
)
}