Support Ollama use: chat|memory and parallel embed beside VL.

Pull nomic-embed-text for Assistent memory, write ollama-roles.json, CPU Modelfile, and raise MAX_LOADED_MODELS/NUM_PARALLEL to 2.

Co-authored-by: Cursor <cursoragent@cursor.com>

src/gpu_rent/llm_runtime.py

@@ -18,9 +18,12 @@ from gpu_rent.paths import (
 
 VALID_RUNTIMES = frozenset({"none", "ollama"})
 
+MEMORY_EMBED_MODEL = "nomic-embed-text"
+
 OLLAMA_PRESETS: dict[str, list[str]] = {
     # Requirement: uncensored (abliterated) + solid Russian. Qwen2.5 family.
     # Vision tags preferred for SwarmUI prompt help with images.
+    # Memory embed (nomic) is appended separately with use: memory.
     "recommended": ["huihui_ai/qwen2.5-vl-abliterated:7b"],  # ~6 GB
     "light": ["huihui_ai/qwen2.5-vl-abliterated:3b"],  # ~3 GB
     "text": ["huihui_ai/qwen2.5-abliterate:7b"],  # ~5 GB, no vision
@@ -79,6 +82,14 @@ def ollama_preset_menu(*, include_keep: bool = False) -> list:
 class OllamaModelEntry:
     name: str
     default: bool = False
+    use: str = "chat"  # chat | memory
+
+
+def _normalize_use(raw: object) -> str:
+    s = str(raw or "chat").strip().lower()
+    if s in {"memory", "embed", "embedding"}:
+        return "memory"
+    return "chat"
 
 
 def normalize_runtime(value: str | None) -> str:
@@ -119,17 +130,37 @@ def parse_ollama_models(path: Path) -> list[OllamaModelEntry]:
         if isinstance(item, str):
             name = item.strip()
             if name:
-                out.append(OllamaModelEntry(name=name))
+                out.append(OllamaModelEntry(name=name, use="chat"))
             continue
         if not isinstance(item, dict):
             continue
         name = str(item.get("name") or "").strip()
         if not name:
             continue
-        out.append(OllamaModelEntry(name=name, default=bool(item.get("default"))))
+        use = _normalize_use(item.get("use") or item.get("role"))
+        out.append(
+            OllamaModelEntry(
+                name=name,
+                default=bool(item.get("default")) and use == "chat",
+                use=use,
+            )
+        )
     return out
 
 
+def ensure_memory_model_entries(entries: list[OllamaModelEntry]) -> list[OllamaModelEntry]:
+    """Append default memory embed if the manifest has chat models but no memory."""
+    if not entries:
+        return entries
+    if any(e.use == "memory" for e in entries):
+        return entries
+    if all(e.use == "memory" for e in entries):
+        return entries
+    return list(entries) + [
+        OllamaModelEntry(name=MEMORY_EMBED_MODEL, default=False, use="memory")
+    ]
+
+
 def already_have_ollama_tag(have: set[str], wanted: str) -> bool:
     """Exact tag match only — qwen2.5:3b must not satisfy qwen2.5:7b."""
     if wanted in have:
@@ -142,14 +173,20 @@ def already_have_ollama_tag(have: set[str], wanted: str) -> bool:
 
 
 def preferred_ollama_model(path: Path) -> str | None:
-    """Manifest default, else first tag."""
-    entries = parse_ollama_models(path)
+    """Manifest default chat model, else first chat tag (never memory/embed)."""
+    entries = [e for e in parse_ollama_models(path) if e.use == "chat"]
     for entry in entries:
         if entry.default:
             return entry.name
     return entries[0].name if entries else None
 
 
+def ollama_roles_payload(entries: list[OllamaModelEntry]) -> dict[str, list[str]]:
+    chat = [e.name for e in entries if e.use == "chat"]
+    memory = [e.name for e in entries if e.use == "memory"]
+    return {"chat": chat, "memory": memory}
+
+
 def _ollama_ps_names(payload: object) -> set[str]:
     names: set[str] = set()
     if not isinstance(payload, dict):
@@ -233,7 +270,9 @@ def write_ollama_models_preset(path: Path, preset: str) -> None:
     names = OLLAMA_PRESETS[key]
     lines = [
         "# Локальный манифест Ollama (не коммить). Пример: ollama-models.example.yaml",
-        "# name = точный тег для `ollama pull`. Пустой models: [] — без pull.",
+        "# name = точный тег для `ollama pull`.",
+        "# use: chat — селект Assistent; use: memory — модель памяти (⚙).",
+        "# Пустой models: [] — без pull.",
         "models:",
     ]
     if not names:
@@ -241,8 +280,11 @@ def write_ollama_models_preset(path: Path, preset: str) -> None:
     else:
         for i, name in enumerate(names):
             lines.append(f"  - name: {name}")
+            lines.append("    use: chat")
             if i == 0:
                 lines.append("    default: true")
+        lines.append(f"  - name: {MEMORY_EMBED_MODEL}")
+        lines.append("    use: memory")
     path.write_text("\n".join(lines) + "\n", encoding="utf-8")