fix: spec santizer for glm, outetts, miotts, minimax.

1 files changed, 45 insertions, 17 deletions

diff --git a/app/backends/audiocpp/wizard.py b/app/backends/audiocpp/wizard.py
index 97f50a5..a1a2a6a 100644
--- a/app/backends/audiocpp/wizard.py
+++ b/app/backends/audiocpp/wizard.py
@@ -26,10 +26,12 @@ from . import configsync as _configsync
 from . import models as _models
 from . import prebuilt as _prebuilt
 from . import voices as _voices
-from .catalog import (_backend_options, build_model_entry,
-                      build_server_config, detect_backend, hosting_task,
+from .catalog import (_backend_options, build_model_entry, build_server_config,
+                      detect_backend, entry_model_path, hosting_task,
                       load_model_catalog, load_server_config,
-                      package_dir_options, server_config_selections)
+                      MIOTTS_CODEC_DISPLAY_NAME, MIOTTS_CODEC_INSTALL_ID,
+                      package_dir_options, server_config_selections,
+                      apply_entry_session_options)
 from .constants import (AUDIOCPP_DIR_NAME, AUDIOCPP_GIT_URL, BACKENDS,
                         DEFAULT_HOST, TASK_TTS, TASK_VDES)
 
@@ -80,7 +82,7 @@ def _build_entries(family_keys: List[str], chosen: Dict[str, List[dict]],
                    task_picker: Callable[[str], str],
                    known_tasks: Optional[Dict[Tuple[str, str], str]] = None
                    ) -> Tuple[List[dict], List[str], List[Tuple[str, str]],
-                              List[str], bool]:
+                              List[Tuple[str, str]], List[str], bool]:
     """Build server.json model entries from the selected families/packages.
 
     TASK_PICKER is called for each design package to choose vdes/tts.
@@ -89,12 +91,18 @@ def _build_entries(family_keys: List[str], chosen: Dict[str, List[dict]],
     was hosted instead of re-asking. Each entry's server id is its package
     ``target_directory`` (flattened to a token), so packages from the same
     family never collide; an id that does collide (across families) is
-    auto-suffixed without prompting. Returns (model_entries, entry_ids,
-    install_guidance, design_entry_ids, include_clone).
+    auto-suffixed without prompting. Entry paths come from the catalog
+    (``entry_model_path``): normally ``models/<target_directory>``, or the
+    package's first GGUF file when the package ships several GGUFs into one
+    directory (audio.cpp refuses multi-GGUF directories). COMPANION_GUIDANCE
+    carries the companion packages hosted models require but the TTS catalog
+    never offers (MioCodec for MioTTS). Returns (model_entries, entry_ids,
+    install_guidance, companion_guidance, design_entry_ids, include_clone).
     """
     model_entries: List[dict] = []
     entry_ids: List[str] = []
     install_guidance: List[Tuple[str, str]] = []
+    companion_guidance: List[Tuple[str, str]] = []
     design_entry_ids: List[str] = []
     include_clone = False
     for family in family_keys:
@@ -120,12 +128,16 @@ def _build_entries(family_keys: List[str], chosen: Dict[str, List[dict]],
                 model_id = f"{base_id}-{n}"
             entry_ids.append(model_id)
             model_entries.append(build_model_entry(
-                family, model_id, f"models/{opt['target_directory']}",
+                family, model_id,
+                entry_model_path(entry, opt["target_directory"]),
                 task=task))
             install_guidance.append((entry["display_name"], opt["install_id"]))
             if task == TASK_VDES:
                 design_entry_ids.append(model_id)
-    return (model_entries, entry_ids, install_guidance,
+    if any(str(e.get("family")) == "miotts" for e in model_entries):
+        companion_guidance.append(
+            (MIOTTS_CODEC_DISPLAY_NAME, MIOTTS_CODEC_INSTALL_ID))
+    return (model_entries, entry_ids, install_guidance, companion_guidance,
             design_entry_ids, include_clone)
 
 
@@ -139,7 +151,11 @@ def _write_and_advise(audiocpp_dir: Path, wav_dir: Optional[Path],
     After a successful run the console output is the path of the written
     server.json. The model install commands (and optional automatic
     download) are handled separately by _install_models, called by both
-    UI modes once the user has decided whether to download.
+    UI modes once the user has decided whether to download. Before the
+    document is written, apply_entry_session_options bakes in the
+    per-entry session options heavy families need (MioTTS's codec path;
+    a VoxCPM AudioVAE encoder capacity sized to the voice directory's
+    longest reference), reported as one summary line.
     """
     voice_dir: Optional[str] = None
     if transcripts:
@@ -149,6 +165,11 @@ def _write_and_advise(audiocpp_dir: Path, wav_dir: Optional[Path],
             print(f"[OK] Wrote {prompt_path}")
         voice_dir = str(wav_dir.resolve())
 
+    configured = apply_entry_session_options(model_entries, wav_dir,
+                                             audiocpp_dir)
+    if configured:
+        print(f"[OK] Added family session options to: {', '.join(configured)}")
+
     server_config = build_server_config(
         host=host, port=port, backend=backend, lazy_load=lazy_load,
         model_entries=model_entries, voice_dir=voice_dir)
@@ -269,7 +290,7 @@ def _wizard(stdscr, args: argparse.Namespace, parser: argparse.ArgumentParser
                 raise _GoBack()
             return result
 
-        model_entries, entry_ids, install_guidance, \
+        model_entries, entry_ids, install_guidance, companion_guidance, \
             design_entry_ids, include_clone = _build_entries(
                 s["family_keys"], s["chosen"], s["catalog_by_family"],
                 task_picker, known_tasks=s["existing_tasks"])
@@ -277,6 +298,7 @@ def _wizard(stdscr, args: argparse.Namespace, parser: argparse.ArgumentParser
             "model_entries": model_entries,
             "entry_ids": entry_ids,
             "install_guidance": install_guidance,
+            "companion_guidance": companion_guidance,
             "design_entry_ids": design_entry_ids,
             "include_clone": include_clone,
         })
@@ -299,6 +321,7 @@ def _wizard(stdscr, args: argparse.Namespace, parser: argparse.ArgumentParser
             "model_entries": s["model_entries"],
             "entry_ids": s["entry_ids"],
             "install_guidance": s["install_guidance"],
+            "companion_guidance": s["companion_guidance"],
             "design_entry_ids": s["design_entry_ids"],
             "include_clone": s["include_clone"],
             "host": host,
@@ -419,8 +442,9 @@ def _wizard(stdscr, args: argparse.Namespace, parser: argparse.ArgumentParser
             s["plan"] = _plan_from_mode(mode, wav_files, existing)
 
         s["download"] = bool(result.get("download")) and (
-            _models.download_applicable(s["audiocpp_dir"],
-                                        s["model_entries"]))
+            _models.download_applicable(
+                s["audiocpp_dir"], s["model_entries"],
+                companions=s.get("companion_guidance")))
         s["delete_unused"] = bool(result.get("delete_unused")) \
             and bool(s["unused_entries"])
         return _finalize()
@@ -540,7 +564,9 @@ def _wizard(stdscr, args: argparse.Namespace, parser: argparse.ArgumentParser
             "visible": lambda fs: bool(s["include_clone"]),
         })
 
-        if _models.download_applicable(s["audiocpp_dir"], s["model_entries"]):
+        if _models.download_applicable(
+                s["audiocpp_dir"], s["model_entries"],
+                companions=s.get("companion_guidance")):
             fields.append({
                 "key": "download",
                 "label": "Download the selected models automatically?",
@@ -718,7 +744,8 @@ def _execute_lanes(settings: dict,
     def install(emit, cancel):
         _models._install_models(audiocpp_dir, settings["install_guidance"],
                         settings["download"], emit=emit, cancel=cancel,
-                        model_entries=settings["model_entries"])
+                        model_entries=settings["model_entries"],
+                        companions=settings.get("companion_guidance"))
         _build._print_launch_hint(audiocpp_dir, settings["output_path"])
         return 0
     # Everything already on disk: the install step just reports it, so the
@@ -992,9 +1019,9 @@ def _collect_from_flags(args: argparse.Namespace,
     def task_picker(install_id: str) -> str:
         return TASK_VDES
 
-    model_entries, entry_ids, install_guidance, design_entry_ids, include_clone = \
-        _build_entries(family_keys, chosen, catalog_by_family,
-                       task_picker)
+    model_entries, entry_ids, install_guidance, companion_guidance, \
+        design_entry_ids, include_clone = _build_entries(
+            family_keys, chosen, catalog_by_family, task_picker)
 
     # Server settings. Host is always 127.0.0.1 and the port comes from
     # AUDIOCPP_API_URL in app/converter/config.py (the Settings screen) —
@@ -1056,6 +1083,7 @@ def _collect_from_flags(args: argparse.Namespace,
         "model_entries": model_entries,
         "entry_ids": entry_ids,
         "install_guidance": install_guidance,
+        "companion_guidance": companion_guidance,
         "design_entry_ids": design_entry_ids,
         "include_clone": include_clone,
         "host": host,