Add per-model GPU and context profiles

frontend/src/api/client.ts

@@ -62,38 +62,12 @@ export interface AgentConfig {
   top_k: number;
   max_tokens: number;
   num_ctx: number;
+  num_gpu: number;
+  num_batch: number;
   tools: Record<string, boolean>;
   confirm_shell: boolean;
 }
 
-export interface AutoTuneResult {
-  detection: {
-    gpu: { available: boolean; name: string; vram_total_mb: number; source: string };
-    model_profile: {
-      context_length: number | null;
-      parameter_size: string | null;
-      quantization: string | null;
-      size_mb: number | null;
-    };
-    recommended: { num_ctx: number; max_tokens: number };
-    rationale: string;
-  };
-  placement: { loaded: boolean; where?: string; gpu_percent?: number };
-  config: AgentConfig;
-}
-
-export async function autoTune(
-  model: string,
-  apply = true
-): Promise<AutoTuneResult> {
-  const res = await fetch("/api/config/auto", {
-    method: "POST",
-    headers: { "Content-Type": "application/json" },
-    body: JSON.stringify({ model, apply }),
-  });
-  return res.json();
-}
-
 export async function runShell(
   command: string
 ): Promise<{ command: string; exit_code: number; output: string }> {
@@ -105,18 +79,21 @@ export async function runShell(
   return res.json();
 }
 
-export async function getConfig(): Promise<{
+export async function getConfig(model?: string): Promise<{
   config: AgentConfig;
   available_tools: string[];
 }> {
-  const res = await fetch("/api/config");
+  const query = model ? `?model=${encodeURIComponent(model)}` : "";
+  const res = await fetch(`/api/config${query}`);
   return res.json();
 }
 
 export async function saveConfig(
-  patch: Partial<AgentConfig>
+  patch: Partial<AgentConfig>,
+  model?: string
 ): Promise<AgentConfig> {
-  const res = await fetch("/api/config", {
+  const query = model ? `?model=${encodeURIComponent(model)}` : "";
+  const res = await fetch(`/api/config${query}`, {
     method: "PUT",
     headers: { "Content-Type": "application/json" },
     body: JSON.stringify(patch),

frontend/src/panels/SettingsView.tsx

@@ -1,7 +1,7 @@
 import { useEffect, useState } from "react";
 import { useStore } from "../store/useStore";
 import { pullModel } from "../api/client";
-import type { AgentConfig, AutoTuneResult } from "../api/client";
+import type { AgentConfig } from "../api/client";
 import { DownloadIcon, RefreshIcon } from "../components/Icon";
 
 const TOOL_DESC: Record<string, string> = {
@@ -25,9 +25,6 @@ export function SettingsView() {
     availableTools,
     refreshConfig,
     updateConfig,
-    tuning,
-    tuneResult,
-    runAutoTune,
   } = useStore();
 
   // Brouillon local édité, synchronisé depuis la config serveur.
@@ -36,7 +33,7 @@ export function SettingsView() {
 
   useEffect(() => {
     refreshConfig();
-  }, [refreshConfig]);
+  }, [refreshConfig, selectedModel]);
   useEffect(() => {
     if (config) setDraft(config);
   }, [config]);
@@ -224,25 +221,12 @@ export function SettingsView() {
               {/* Génération + Outils */}
               <div className="flex flex-col gap-5">
                 <div className="rounded-card border border-line bg-card p-[18px]">
-                  <div className="mb-4 flex items-center justify-between">
-                    <div className="text-sm font-bold">Génération</div>
-                    <button
-                      onClick={runAutoTune}
-                      disabled={tuning || !selectedModel}
-                      className="flex h-[26px] items-center gap-1.5 rounded-[7px] border border-[rgba(240,161,92,.4)] bg-[rgba(240,161,92,.10)] px-2.5 text-[11.5px] font-semibold text-accent disabled:opacity-40"
-                      title="Détecte le GPU et le modèle, puis optimise le contexte et les jetons"
-                    >
-                      {tuning ? (
-                        <span className="h-3 w-3 animate-spin rounded-full border-2 border-accent/40 border-t-accent" />
-                      ) : (
-                        <span>⚡</span>
-                      )}
-                      {tuning ? "Détection…" : "Réglage auto"}
-                    </button>
+                  <div className="mb-1 text-sm font-bold">Génération</div>
+                  <div className="mb-4 text-[11.5px] text-muted-2">
+                    Profil enregistré pour <b>{selectedModel || "ce modèle"}</b>.
+                    Le contexte détermine directement la taille du cache KV.
                   </div>
 
-                  {tuneResult && <TuneBanner result={tuneResult} />}
-
                   <Slider
                     label="Température"
                     value={draft.temperature}
@@ -280,17 +264,38 @@ export function SettingsView() {
                     onChange={(v) => set("max_tokens", Math.round(v))}
                   />
                   <Slider
-                    label="Contexte (num_ctx)"
+                    label="Cache KV / contexte"
                     value={draft.num_ctx}
-                    min={0}
+                    min={2048}
                     max={32768}
                     step={1024}
-                    fmt={(v) =>
-                      v === 0 ? "auto" : v >= 1024 ? `${v / 1024}K` : String(v)
-                    }
+                    fmt={(v) => `${Math.round(v / 1024)}K`}
                     onChange={(v) => set("num_ctx", Math.round(v))}
+                  />
+                  <Slider
+                    label="Couches GPU"
+                    value={draft.num_gpu}
+                    min={-1}
+                    max={120}
+                    step={1}
+                    fmt={(v) => (v < 0 ? "auto" : String(Math.round(v)))}
+                    onChange={(v) => set("num_gpu", Math.round(v))}
+                  />
+                  <Slider
+                    label="Batch"
+                    value={draft.num_batch}
+                    min={64}
+                    max={1024}
+                    step={64}
+                    fmt={(v) => String(Math.round(v))}
+                    onChange={(v) => set("num_batch", Math.round(v))}
                     last
                   />
+                  <div className="mt-3 rounded-lg border border-line-soft bg-base px-3 py-2 text-[11px] leading-relaxed text-muted-2">
+                    La précision KV (<code>f16</code>/<code>q8_0</code>) est un
+                    réglage global du serveur Ollama et nécessite son redémarrage.
+                    Les valeurs ci-dessus sont, elles, propres à chaque modèle.
+                  </div>
                 </div>
 
                 <div className="rounded-card border border-line bg-card p-[18px]">
@@ -384,51 +389,6 @@ export function SettingsView() {
   );
 }
 
-function TuneBanner({ result }: { result: AutoTuneResult }) {
-  const { gpu, model_profile, recommended, rationale } = result.detection;
-  const place = result.placement;
-
-  const placeLabel =
-    place.loaded && place.where
-      ? place.where === "gpu"
-        ? `chargé GPU (${place.gpu_percent}%)`
-        : place.where === "cpu"
-        ? "chargé CPU ⚠️"
-        : `mixte GPU ${place.gpu_percent}%`
-      : "non chargé";
-  const placeColor =
-    place.where === "gpu" ? "text-ok" : place.where === "cpu" ? "text-warn" : "text-muted";
-
-  return (
-    <div className="mb-4 rounded-[10px] border border-[rgba(240,161,92,.35)] bg-[rgba(240,161,92,.06)] p-3 text-[11.5px]">
-      <div className="mb-1.5 flex items-center justify-between">
-        <span className="font-semibold text-accent">⚡ Optimisé</span>
-        <span className={`font-mono ${placeColor}`}>{placeLabel}</span>
-      </div>
-      <div className="font-mono text-muted">
-        {gpu.available
-          ? `${gpu.name} · ${(gpu.vram_total_mb / 1024).toFixed(1)} Go VRAM`
-          : "Aucun GPU détecté"}
-        {model_profile.context_length
-          ? ` · ctx modèle ${Math.round(model_profile.context_length / 1024)}K`
-          : ""}
-      </div>
-      <div className="mt-1 text-muted-2">
-        → contexte <b className="text-ink-2">{recommended.num_ctx}</b> · jetons max{" "}
-        <b className="text-ink-2">{recommended.max_tokens}</b>
-      </div>
-      {rationale && <div className="mt-1 text-muted-3">{rationale}</div>}
-      {!gpu.available && (
-        <div className="mt-1.5 text-muted-3">
-          GPU non détecté dans le conteneur. Si Ollama tourne sur une autre
-          machine, déclare la VRAM via <code>GPU_VRAM_MB</code> (ex. 12000) dans
-          le compose pour un réglage précis.
-        </div>
-      )}
-    </div>
-  );
-}
-
 function Slider({
   label,
   value,

frontend/src/store/useStore.ts

@@ -1,6 +1,5 @@
 import { create } from "zustand";
 import {
-  autoTune,
   createSession,
   deleteSession,
   getConfig,
@@ -14,7 +13,6 @@ import {
   saveConfig,
   streamChat,
   type AgentConfig,
-  type AutoTuneResult,
   type FileNode,
   type Message,
   type OllamaModel,
@@ -47,10 +45,6 @@ interface LokiState {
   refreshConfig: () => Promise<void>;
   updateConfig: (patch: Partial<AgentConfig>) => Promise<void>;
 
-  tuning: boolean;
-  tuneResult: AutoTuneResult | null;
-  runAutoTune: () => Promise<void>;
-
   pendingShell: string | null; // commande shell en attente de validation
   approveShell: () => Promise<void>;
   rejectShell: () => Promise<void>;
@@ -115,36 +109,26 @@ export const useStore = create<LokiState>((set, get) => ({
   },
 
   refreshConfig: async () => {
-    const { config, available_tools } = await getConfig();
+    const { config, available_tools } = await getConfig(
+      get().selectedModel || undefined
+    );
     set({ config, availableTools: available_tools });
   },
 
   updateConfig: async (patch) => {
-    const config = await saveConfig(patch);
+    const config = await saveConfig(patch, get().selectedModel || undefined);
     set({ config });
   },
 
-  tuning: false,
-  tuneResult: null,
-
-  runAutoTune: async () => {
-    const model = get().selectedModel;
-    if (!model || get().tuning) return;
-    set({ tuning: true });
-    try {
-      const result = await autoTune(model, true);
-      set({ tuneResult: result, config: result.config });
-    } finally {
-      set({ tuning: false });
-    }
-  },
-
   openPreview: async (path) => {
     const content = await fileContent(path);
     set({ previewPath: path, previewContent: content });
   },
 
-  setSelectedModel: (name) => set({ selectedModel: name }),
+  setSelectedModel: (name) => {
+    set({ selectedModel: name });
+    void get().refreshConfig();
+  },
 
   refreshFiles: async () => {
     try {
@@ -175,6 +159,7 @@ export const useStore = create<LokiState>((set, get) => ({
           ? def
           : models[0]?.name ?? "";
       set({ models, selectedModel });
+      await get().refreshConfig();
     } finally {
       set({ loadingModels: false });
     }