fix(cookbook): treat local Windows as Windows for serve commands (odysseus-dev#3975)

* fix(cookbook): prefer native llama-server on local Windows

* fix(cookbook): harden local llama-server launch commands

* fix(cookbook): build serve commands for selected target

Author: zoomdbz

routes/cookbook_routes.py

@@ -74,6 +74,9 @@ def _mask_secret(value: str) -> str:
             return "stored"
         return f"{value[:4]}...{value[-4:]}"
 
+    def _client_host_platform() -> str:
+        return "windows" if IS_WINDOWS else ""
+
     def _decrypt_secret(value: str | None) -> str:
         if not value:
             return ""
@@ -246,11 +249,15 @@ def _state_for_client(state):
         """Return cookbook state without raw secrets for browser clients."""
         _strip_task_secrets(state)
         env = state.get("env") if isinstance(state, dict) else None
+        if isinstance(state, dict) and not isinstance(env, dict):
+            env = {}
+            state["env"] = env
         if isinstance(env, dict):
             token = _decrypt_secret(env.get("hfToken"))
             env.pop("hfToken", None)
             env["hfTokenConfigured"] = bool(token)
             env["hfTokenMasked"] = _mask_secret(token)
+            env["hostPlatform"] = _client_host_platform()
         return state
 
     def _state_for_storage(state, on_disk=None):
@@ -269,6 +276,7 @@ def _state_for_storage(state, on_disk=None):
                 env.pop("hfToken", None)
             env.pop("hfTokenMasked", None)
             env.pop("hfTokenConfigured", None)
+            env.pop("hostPlatform", None)
         return state
 
     def _load_stored_hf_token() -> str:
@@ -1528,6 +1536,10 @@ async def model_serve(request: Request, req: ServeRequest):
             # shell resolves the bundled python3/hf, mirroring the download flow.
             if not remote:
                 runner_lines.append(_local_tooling_path_export(sys.executable))
+                if local_windows:
+                    # Detached Git Bash runs do not always inherit recently edited
+                    # user PATH entries from the already-running Odysseus process.
+                    runner_lines.append('export PATH="$HOME/bin:$HOME/llama.cpp/build-cuda/bin/Release:$HOME/llama.cpp/build/bin/Release:$HOME/llama.cpp/build/bin/Debug:$HOME/llama.cpp/build/bin:$PATH"')
             runner_lines.append("export FLASHINFER_DISABLE_VERSION_CHECK=1")
             if req.hf_token:
                 runner_lines.append(f"export HF_TOKEN='{_bash_squote(req.hf_token)}'")
@@ -1542,7 +1554,8 @@ async def model_serve(request: Request, req: ServeRequest):
             runner_lines.append(_HF_TOKEN_STATUS_SNIPPET)
             handled_ollama_serve = False
             # Auto-install inference engine if missing
-            if "llama_cpp" in req.cmd or "llama-server" in req.cmd:
+            local_windows_llama_cmd = local_windows and ("llama_cpp" in req.cmd or "llama-server" in req.cmd)
+            if ("llama_cpp" in req.cmd or "llama-server" in req.cmd) and not local_windows_llama_cmd:
                 # Prefer the NATIVE llama-server binary — its minja templating
                 # renders modern GGUF chat templates that the Python bindings'
                 # Jinja2 rejects (do_tojson ensure_ascii). Build it once from
@@ -2445,8 +2458,8 @@ async def get_cookbook_state(request: Request):
             try:
                 return _state_for_client(json.loads(_cookbook_state_path.read_text(encoding="utf-8")))
             except Exception:
-                return {}
-        return {}
+                return _state_for_client({})
+        return _state_for_client({})
 
     @router.post("/api/cookbook/state")
     async def save_cookbook_state(request: Request):

static/js/cookbookServe.js

@@ -46,30 +46,6 @@ const SERVE_STATE_KEY = 'cookbook-serve-state';
 const SERVE_FAVORITES_KEY = 'cookbook-serve-favorite-models';
 
 let _cachedAllModels = [];
-const _CACHED_MODELS_SCAN_KEY = 'cookbook_cached_models_scan_v1';
-const _CACHED_MODELS_SCAN_TTL = 6 * 3600 * 1000;
-
-function _readCachedModelScan(sig) {
-  try {
-    const all = JSON.parse(localStorage.getItem(_CACHED_MODELS_SCAN_KEY) || '{}');
-    const entry = all[sig];
-    if (entry && Date.now() - (entry.ts || 0) < _CACHED_MODELS_SCAN_TTL) return entry.data || null;
-  } catch {}
-  return null;
-}
-
-function _writeCachedModelScan(sig, data) {
-  try {
-    const all = JSON.parse(localStorage.getItem(_CACHED_MODELS_SCAN_KEY) || '{}');
-    all[sig] = { ts: Date.now(), data };
-    const keys = Object.keys(all);
-    if (keys.length > 12) {
-      keys.sort((a, b) => (all[a].ts || 0) - (all[b].ts || 0));
-      for (const k of keys.slice(0, keys.length - 12)) delete all[k];
-    }
-    localStorage.setItem(_CACHED_MODELS_SCAN_KEY, JSON.stringify(all));
-  } catch {}
-}
 
 function _loadServeFavorites() {
   try {
@@ -551,7 +527,7 @@ function _selectedServeTarget(panel) {
     env: server?.env || '',
     port: host ? (server?.port || _getPort(host) || '') : '',
     venv,
-    platform: server?.platform || _envState.platform || '',
+    platform: host ? (server?.platform || '') : (_envState.hostPlatform || ''),
     label,
   };
 }
@@ -682,6 +658,12 @@ function _selectedGgufSizeGb(model, relPath) {
   return bytes / (1024 ** 3);
 }
 
+function _projectorGgufFiles(model) {
+  return _ggufFilesForModel(model)
+    .filter(f => (f.role || '') === 'projector' || /(^|\/)mmproj[^/]*\.gguf$/i.test(f.rel_path || f.name || ''))
+    .sort((a, b) => String(a.rel_path || a.name || '').localeCompare(String(b.rel_path || b.name || '')));
+}
+
 function _ggufFileLabel(file) {
   const base = (file.name || file.rel_path || '').split('/').pop();
   const size = _formatGgufSize(file.size_bytes);
@@ -1222,6 +1204,7 @@ function _rerenderCachedModels() {
         panelHtml += `<div class="hwfit-serve-warn" style="margin:0 0 8px;padding:6px 10px;border-radius:5px;font-size:11px;background:color-mix(in srgb, var(--color-warning, #f0ad4e) 14%, transparent);border:1px solid color-mix(in srgb, var(--color-warning, #f0ad4e) 40%, transparent);color:var(--color-warning, #f0ad4e);display:flex;gap:6px;align-items:flex-start;line-height:1.4;"><span aria-hidden="true">⚠</span><span>${_warnText}</span></div>`;
       }
       panelHtml += `<div class="hwfit-serve-preset-row">${_slotsHtml}</div>`;
+      panelHtml += `<div class="hwfit-serve-vision-warn" style="display:none;margin:0 0 8px;padding:6px 10px;border-radius:5px;font-size:11px;background:color-mix(in srgb, var(--color-warning, #f0ad4e) 14%, transparent);border:1px solid color-mix(in srgb, var(--color-warning, #f0ad4e) 40%, transparent);color:var(--color-warning, #f0ad4e);gap:6px;align-items:flex-start;line-height:1.4;"><span aria-hidden="true">⚠</span><span>Vision is enabled, but no mmproj GGUF projector was found in the cached model scan. Download an mmproj-*.gguf for this model, then refresh the cached model list before launching.</span></div>`;
       // Row 1: Engine + Server + Env
       panelHtml += `<div class="hwfit-serve-row">`;
       const backendOpts = _backendChoices.map(([v,l]) => `<option value="${v}"${defaultBackend===v?' selected':''}>${l}</option>`).join('');
@@ -1548,6 +1531,11 @@ function _rerenderCachedModels() {
           if (el.type === 'checkbox') f[el.dataset.field] = el.checked;
           else f[el.dataset.field] = el.value;
         });
+        const buildTarget = _selectedServeTarget(panel);
+        f.host = buildTarget.host || '';
+        f.platform = buildTarget.platform || '';
+        const hostField = panel.querySelector('[data-field="host"]');
+        if (hostField) hostField.value = f.host;
         const backend = f.backend || 'vllm';
         const serveModel = (f.model_path || '').trim() || (m.is_local_dir && m.path ? `${m.path}/${repo}` : repo);
         if (backend === 'llamacpp') {
@@ -1567,11 +1555,11 @@ function _rerenderCachedModels() {
             : m.is_local_dir && m.path
             ? `$({ find ${_ldir} -name '*-00001-of-*.gguf' 2>/dev/null | sort; find ${_ldir} -name '*.gguf' 2>/dev/null | sort; } | head -1)`
             : `$({ find ${dir} -name '*-00001-of-*.gguf' 2>/dev/null | sort; find ${dir} -name '*.gguf' 2>/dev/null | sort; } | head -1)`;
-          // Vision: auto-find the mmproj (CLIP/projector) file in the same dir.
-          // Resolved at runtime so the toggle just works if an mmproj-*.gguf is
-          // present (downloaded alongside the model). Empty if none → cmd omits it.
-          const _vsearchdir = (m.is_local_dir && m.path) ? _ldir : dir;
-          f._mmproj_path = `$(find ${_vsearchdir} -iname 'mmproj*.gguf' 2>/dev/null | sort | head -1)`;
+          // Vision: use the scanned projector (CLIP/mmproj) file when present.
+          // Keeping this as a printf path avoids generating a command substitution
+          // that the backend serve-command validator must reject as unsafe.
+          const selectedProjector = _projectorGgufFiles(m)[0];
+          f._mmproj_path = selectedProjector ? _selectedGgufExpr(m, repo, selectedProjector.rel_path) : '';
         }
         if (f.reasoning_parser) {
           const _rpEl2 = panel.querySelector('[data-field="reasoning_parser"]');
@@ -1587,6 +1575,10 @@ function _rerenderCachedModels() {
         }
         let cmd = _buildServeCmd(f, serveModel, backend);
         if (f.extra && f.extra.trim()) cmd += ' ' + f.extra.trim();
+        const missingVisionProjector = backend === 'llamacpp' && !!f.vision && !f._mmproj_path;
+        panel._visionMissingProjector = missingVisionProjector;
+        const _visionWarn = panel.querySelector('.hwfit-serve-vision-warn');
+        if (_visionWarn) _visionWarn.style.display = missingVisionProjector ? 'flex' : 'none';
         const _ce2 = panel.querySelector('.hwfit-serve-cmd'); _ce2.value = _formatServeCmdPreview(cmd); _ce2.style.height = 'auto'; _ce2.style.height = _ce2.scrollHeight + 'px';
         panel._cmd = cmd;
         panel._host = f.host || '';
@@ -2962,12 +2954,16 @@ function _rerenderCachedModels() {
         });
         serveState.backend = serveState.backend || (_detectBackend(m).backend) || 'vllm';
         const launchTarget = _selectedServeTarget(panel);
+        if (serveState.backend === 'llamacpp' && serveState.vision && !/(?:^|\s)(?:--mmproj|--clip_model_path)\b/.test(launchCmd)) {
+          _restoreLaunchBtn();
+          uiModule.showToast('Vision is checked, but no mmproj projector is in the launch command. Refresh cached models after downloading mmproj, or add --mmproj manually.', 8000);
+          return;
+        }
         if (serveState.backend === 'diffusers' && _remoteWindowsDiffusersUnsupported(launchTarget)) {
           _restoreLaunchBtn();
           uiModule.showToast('Diffusers serving is not supported on remote Windows servers yet. Use local Windows or a Linux server.', 9000);
           return;
         }
-
         // Pre-launch: check our own task list for a serve already running
         // on this host. Offer to stop+launch as the default action — the
         // SSH-based port probe below is more thorough but it can miss
@@ -3596,91 +3592,12 @@ export async function openServePanelForRepo(repo, fields) {
 
 // ── Fetch cached models from server ──
 
-function _renderCachedModelsData(list, data, host) {
-  // CHANGELOG: 'ready' already excludes partial downloads;
-  // show every complete model regardless of size/backend.
-  const ready = (data.models || []).filter(m => m.status === 'ready');
-
-  const downloading = (data.models || []).filter(m => m.status === 'downloading');
-  const allModels = [...ready, ...downloading];
-  _cachedAllModels = allModels;
-
-  if (!allModels.length) {
-    if (!host) {
-      list.innerHTML = '<div class="hwfit-loading" style="flex-direction:column;gap:6px;text-align:center;"><div>No cached models found</div><div style="font-size:11px;opacity:0.55;max-width:420px;line-height:1.4;">Docker Local uses Odysseus’s cache in <code>data/huggingface</code>. Download a model here, or copy an existing host HuggingFace cache into that folder once.</div></div>';
-    } else {
-      list.innerHTML = '<div class="hwfit-loading">No cached models found</div>';
-    }
-    const tagContainer = document.getElementById('serve-tags');
-    if (tagContainer) tagContainer.innerHTML = '';
-    return;
-  }
-
-  // Auto-detect type + family tags
-  const _tagMap = {};
-  const _familyMap = {};
-  const _families = [
-    [/qwen/i, 'qwen'], [/llama/i, 'llama'], [/mistral|mixtral/i, 'mistral'],
-    [/deepseek/i, 'deepseek'], [/gemma/i, 'gemma'], [/phi/i, 'phi'],
-    [/minimax/i, 'minimax'], [/glm/i, 'glm'], [/flux/i, 'flux'],
-    [/stable.?diffusion|sdxl/i, 'sd'], [/z-image/i, 'z-image'],
-    [/whisper/i, 'whisper'], [/command|cohere/i, 'cohere'],
-    [/yi-/i, 'yi'], [/intern/i, 'intern'], [/falcon/i, 'falcon'],
-  ];
-  for (const m of allModels) {
-    const n = (m.repo_id || '').toLowerCase();
-    let tag = 'other';
-    if (m.backend === 'ollama' || m.is_ollama) tag = 'llm';
-    else if (m.is_diffusion || /flux|sdxl|stable-diffusion|z-image|qwen-image|diffusion|dreamshar/i.test(n)) tag = 'image';
-    else if (/whisper|stt|asr/i.test(n)) tag = 'stt';
-    else if (/tts|cosyvoice|parler/i.test(n)) tag = 'tts';
-    else if (/embed|bge|minilm|e5-/i.test(n)) tag = 'embedding';
-    else if (/lora|adapter/i.test(n)) tag = 'lora';
-    else tag = 'llm';
-    m._tag = tag;
-    _tagMap[tag] = (_tagMap[tag] || 0) + 1;
-    m._family = '';
-    for (const [re, fam] of _families) {
-      if (re.test(n)) { m._family = fam; _familyMap[fam] = (_familyMap[fam] || 0) + 1; break; }
-    }
-    if ((m.backend === 'ollama' || m.is_ollama) && !m._family) {
-      m._family = 'ollama';
-      _familyMap.ollama = (_familyMap.ollama || 0) + 1;
-    }
-  }
-
-  // Render tag chips
-  const tagContainer = document.getElementById('serve-tags');
-  if (tagContainer) {
-    const tagOrder = ['llm', 'image', 'lora', 'embedding', 'tts', 'stt', 'other'];
-    let tagHtml = `<button class="memory-cat-chip active" data-serve-tag="">All (${allModels.length})</button>`;
-    for (const t of tagOrder) {
-      if (!_tagMap[t]) continue;
-      tagHtml += `<button class="memory-cat-chip" data-serve-tag="${t}">${t} (${_tagMap[t]})</button>`;
-    }
-    const sortedFamilies = Object.entries(_familyMap).sort((a, b) => b[1] - a[1]);
-    if (sortedFamilies.length) {
-      for (const [fam, count] of sortedFamilies) {
-        const logo = providerLogo(fam);
-        const logoHtml = logo ? `<span style="width:12px;height:12px;display:inline-flex;align-items:center;vertical-align:-2px;margin-right:2px;opacity:0.6;">${logo}</span>` : '';
-        tagHtml += `<button class="memory-cat-chip" data-serve-tag="fam:${fam}">${logoHtml}${fam} (${count})</button>`;
-      }
-    }
-    tagContainer.innerHTML = tagHtml;
-  }
-
-  _rerenderCachedModels();
-}
-
-export async function _fetchCachedModels(fresh = false) {
+export async function _fetchCachedModels() {
   const list = document.getElementById('hwfit-cached-list');
   if (!list) return;
 
   list.innerHTML = '';
-  const _dlWp = spinnerModule.createWhirlpool(22);
-  _dlWp.element.classList.add('cookbook-section-loading-wp');
-  _dlWp.element.style.width = '22px';
-  _dlWp.element.style.height = '22px';
+  const _dlWp = spinnerModule.createWhirlpool(18);
   const _dlWrap = document.createElement('div');
   _dlWrap.className = 'hwfit-loading';
   _dlWrap.style.cssText = 'flex-direction:column;gap:6px;';
@@ -3742,13 +3659,6 @@ export async function _fetchCachedModels(fresh = false) {
     if (host) { qp.set('host', host); const _sp4 = _getPort(host); if (_sp4) qp.set('ssh_port', _sp4); const _plat = _getPlatform(host); if (_plat) qp.set('platform', _plat); }
     if (modelDirs.length) qp.set('model_dir', modelDirs.join(','));
     const params = qp.toString() ? `?${qp}` : '';
-    const scanSig = params || 'local';
-    const cached = fresh ? null : _readCachedModelScan(scanSig);
-    if (cached) {
-      _dlWp.destroy();
-      _renderCachedModelsData(list, cached, host);
-      return;
-    }
     const res = await fetch(`/api/model/cached${params}`);
     if (!res.ok) {
       const body = await res.text().catch(() => '');
@@ -3763,9 +3673,80 @@ export async function _fetchCachedModels(fresh = false) {
       throw new Error(`HTTP ${res.status} ${res.statusText}${msg ? `: ${msg}` : ''}`);
     }
     const data = await res.json();
-    _writeCachedModelScan(scanSig, data);
     _dlWp.destroy();
-    _renderCachedModelsData(list, data, host);
+
+    // CHANGELOG: 'ready' already excludes partial downloads; 
+    // show every complete model regardless of size/backend.
+    const ready = data.models.filter(m => m.status === 'ready');
+
+    const downloading = data.models.filter(m => m.status === 'downloading');
+    const allModels = [...ready, ...downloading];
+    _cachedAllModels = allModels;
+
+    if (!allModels.length) {
+      if (!host) {
+        list.innerHTML = '<div class="hwfit-loading" style="flex-direction:column;gap:6px;text-align:center;"><div>No cached models found</div><div style="font-size:11px;opacity:0.55;max-width:420px;line-height:1.4;">Docker Local uses Odysseus’s cache in <code>data/huggingface</code>. Download a model here, or copy an existing host HuggingFace cache into that folder once.</div></div>';
+      } else {
+        list.innerHTML = '<div class="hwfit-loading">No cached models found</div>';
+      }
+      document.getElementById('serve-tags').innerHTML = '';
+      return;
+    }
+
+    // Auto-detect type + family tags
+    const _tagMap = {};
+    const _familyMap = {};
+    const _families = [
+      [/qwen/i, 'qwen'], [/llama/i, 'llama'], [/mistral|mixtral/i, 'mistral'],
+      [/deepseek/i, 'deepseek'], [/gemma/i, 'gemma'], [/phi/i, 'phi'],
+      [/minimax/i, 'minimax'], [/glm/i, 'glm'], [/flux/i, 'flux'],
+      [/stable.?diffusion|sdxl/i, 'sd'], [/z-image/i, 'z-image'],
+      [/whisper/i, 'whisper'], [/command|cohere/i, 'cohere'],
+      [/yi-/i, 'yi'], [/intern/i, 'intern'], [/falcon/i, 'falcon'],
+    ];
+    for (const m of allModels) {
+      const n = (m.repo_id || '').toLowerCase();
+      let tag = 'other';
+      if (m.backend === 'ollama' || m.is_ollama) tag = 'llm';
+      else if (m.is_diffusion || /flux|sdxl|stable-diffusion|z-image|qwen-image|diffusion|dreamshar/i.test(n)) tag = 'image';
+      else if (/whisper|stt|asr/i.test(n)) tag = 'stt';
+      else if (/tts|cosyvoice|parler/i.test(n)) tag = 'tts';
+      else if (/embed|bge|minilm|e5-/i.test(n)) tag = 'embedding';
+      else if (/lora|adapter/i.test(n)) tag = 'lora';
+      else tag = 'llm';
+      m._tag = tag;
+      _tagMap[tag] = (_tagMap[tag] || 0) + 1;
+      m._family = '';
+      for (const [re, fam] of _families) {
+        if (re.test(n)) { m._family = fam; _familyMap[fam] = (_familyMap[fam] || 0) + 1; break; }
+      }
+      if ((m.backend === 'ollama' || m.is_ollama) && !m._family) {
+        m._family = 'ollama';
+        _familyMap.ollama = (_familyMap.ollama || 0) + 1;
+      }
+    }
+
+    // Render tag chips
+    const tagContainer = document.getElementById('serve-tags');
+    if (tagContainer) {
+      const tagOrder = ['llm', 'image', 'lora', 'embedding', 'tts', 'stt', 'other'];
+      let tagHtml = `<button class="memory-cat-chip active" data-serve-tag="">All (${allModels.length})</button>`;
+      for (const t of tagOrder) {
+        if (!_tagMap[t]) continue;
+        tagHtml += `<button class="memory-cat-chip" data-serve-tag="${t}">${t} (${_tagMap[t]})</button>`;
+      }
+      const sortedFamilies = Object.entries(_familyMap).sort((a, b) => b[1] - a[1]);
+      if (sortedFamilies.length) {
+        for (const [fam, count] of sortedFamilies) {
+          const logo = providerLogo(fam);
+          const logoHtml = logo ? `<span style="width:12px;height:12px;display:inline-flex;align-items:center;vertical-align:-2px;margin-right:2px;opacity:0.6;">${logo}</span>` : '';
+          tagHtml += `<button class="memory-cat-chip" data-serve-tag="fam:${fam}">${logoHtml}${fam} (${count})</button>`;
+        }
+      }
+      tagContainer.innerHTML = tagHtml;
+    }
+
+    _rerenderCachedModels();
   } catch (e) {
     _dlWp.destroy();
     list.innerHTML = `<div class="hwfit-loading">Failed: ${esc(e.message)}</div>`;