omnius 1.0.643 → 1.0.646

package/dist/postinstall-daemon.cjs

@@ -657,7 +657,10 @@ function installSystemd(nodeBin, omniusScript, user) {
     "Environment=OMNIUS_DAEMON=1",
     "Environment=OMNIUS_PORT=" + PORT,
     "Environment=NODE_ENV=production",
-    "EnvironmentFile=-" + envPath,
+    // Keep this user-relative so a unit written during a root/global install
+    // still loads the owning user's overrides after the existing unit is
+    // rewritten on update.
+    "EnvironmentFile=%h/.config/omnius/daemon.env",
     "ExecStart=" + nodeBin + " " + omniusScript + " serve --daemon --quiet",
     // Restart=always (was on-failure) — also relaunch on clean exit.
     // Some upgrade flows trigger process.exit(0) (e.g. /update reload,

package/dist/scripts/audio-clap-semantic-worker.py

@@ -1,9 +1,10 @@
 #!/usr/bin/env python3
 """Persistent, offline CLAP semantic-audio embedding worker.
 
-This process deliberately has no setup behavior.  The Node runtime has already
-created the system-site venv, installed its pinned non-Torch requirements, and
-verified the immutable model artifact before this process is started.
+This process deliberately has no setup behavior. The Node runtime has already
+created an isolated venv with an explicit vendor-Torch link, installed its
+pinned non-Torch requirements, and verified the immutable model artifact
+before this process is started.
 """
 
 import argparse
@@ -22,7 +23,14 @@ if os.environ.get("PYTHONNOUSERSITE") != "1":
 import numpy as np
 import torch
 import torch.nn.functional as F
-from transformers import ClapModel, ClapProcessor
+from PIL import Image
+from transformers import (
+    ClapAudioModelWithProjection,
+    ClapFeatureExtractor,
+    ClapModel,
+    ClapProcessor,
+    ClapTextModelWithProjection,
+)
 
 MODEL_ID = "laion/clap-htsat-unfused"
 # The worker receives exactly Egg's caller-conditioned audio window.  It does

package/dist/update-worker.js

@@ -245098,6 +245098,7 @@ init_venv_paths();
 var SETUP_TIMEOUT_MS2 = 30 * 6e4;
 var CLAP_RUNTIME_WHEELS = [
   { distribution: "numpy", version: "1.26.4", importName: "numpy", filename: "numpy-1.26.4-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", source: "https://files.pythonhosted.org/packages/fc/a5/4beee6488160798683eed5bdb7eead455892c3b4e1f78d79d8d3f3b084ac/numpy-1.26.4-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", sha256: "d209d8969599b27ad20994c8e41936ee0964e6da07478d6c35016bc386b66ad4" },
+  { distribution: "Pillow", version: "10.4.0", importName: "PIL", filename: "pillow-10.4.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", source: "https://files.pythonhosted.org/packages/8a/25/1fc45761955f9359b1169aa75e241551e74ac01a09f487adaaf4c3472d11/pillow-10.4.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", sha256: "7928ecbf1ece13956b95d9cbcfc77137652b02763ba384d9ab508099a2eca856" },
   { distribution: "transformers", version: "4.57.3", importName: "transformers", filename: "transformers-4.57.3-py3-none-any.whl", source: "https://files.pythonhosted.org/packages/6a/6b/2f416568b3c4c91c96e5a365d164f8a4a4a88030aa8ab4644181fdadce97/transformers-4.57.3-py3-none-any.whl", sha256: "c77d353a4851b1880191603d36acb313411d3577f6e2897814f333841f7003f4" },
   { distribution: "huggingface-hub", version: "0.36.0", importName: "huggingface_hub", filename: "huggingface_hub-0.36.0-py3-none-any.whl", source: "https://files.pythonhosted.org/packages/cb/bd/1a875e0d592d447cbc02805fd3fe0f497714d6a2583f59d14fa9ebad96eb/huggingface_hub-0.36.0-py3-none-any.whl", sha256: "7bcc9ad17d5b3f07b57c78e79d527102d08313caa278a641993acddcb894548d" },
   { distribution: "hf-xet", version: "1.1.5", importName: "hf_xet", filename: "hf_xet-1.1.5-cp37-abi3-manylinux_2_28_aarch64.whl", source: "https://files.pythonhosted.org/packages/d0/54/0fcf2b619720a26fbb6cc941e89f2472a522cd963a776c089b189559447f/hf_xet-1.1.5-cp37-abi3-manylinux_2_28_aarch64.whl", sha256: "dbba1660e5d810bd0ea77c511a99e9242d920790d0e63c0e4673ed36c4022d18" },
@@ -245118,6 +245119,9 @@ var CLAP_RUNTIME_WHEELS = [
 ];
 var CLAP_PYTHON_REQUIREMENTS = CLAP_RUNTIME_WHEELS.map((wheel) => `${wheel.distribution}==${wheel.version}`);
 
+// packages/execution/dist/openclip-memory-admission.js
+init_jetson_monitor();
+
 // packages/execution/dist/speaker-embedding-runtime.js
 init_jetson_monitor();
 init_process_async();

package/docs/DISCOVERY.json

@@ -4631,7 +4631,7 @@
           "tags": [
             "Audio"
           ],
-          "description": "Non-mutating readiness for NVIDIA diar_streaming_sortformer_4spk-v2. It never downloads, installs, creates environments, or loads a model. Managed JetPack setup pins NVIDIA's Q8 GGUF plus NeMo-Speech.cpp source revision and keeps one persistent worker/controller serialized; operator-provided .nemo/Python snapshots remain supported. 200 requires the verified runtime worker to be active. Session-local labels are never durable identities.",
+          "description": "Non-mutating readiness for NVIDIA diar_streaming_sortformer_4spk-v2. It never downloads, installs, creates environments, or loads a model. Managed JetPack setup pins NVIDIA's Q8 GGUF plus NeMo-Speech.cpp source revision and keeps one persistent worker/controller serialized; operator-provided .nemo/Python snapshots remain supported. Readiness persists setup.stage, setup.failed_stage, complete setup stderr, and the terminal error across daemon restart. 200 requires the verified runtime worker to be active. Session-local labels are never durable identities.",
           "responses": {
             "200": {
               "description": "Verified local snapshot and warm managed live worker."
@@ -4707,7 +4707,7 @@
           "tags": [
             "Audio"
           ],
-          "description": "Admin-only and asynchronous. An empty JSON object provisions the pinned public Q8 Sortformer artifact and builds an immutable-revision CUDA NeMo-Speech.cpp runtime under ~/.omnius without modifying JetPack Torch. Poll readiness after HTTP 202. Advanced operators may instead provide snapshot_path/manifest_path plus python_path for a checksum-pinned .nemo runtime. Setup may download/build; inference never does.",
+          "description": "Admin-only and asynchronous. An empty JSON object provisions the pinned public Q8 Sortformer artifact and builds an immutable-revision CUDA NeMo-Speech.cpp runtime under ~/.omnius without modifying JetPack Torch. nvcc discovery checks OMNIUS_DIAR_NVCC, /usr/local/cuda-12.2/bin/nvcc, then /usr/local/cuda/bin/nvcc and reports exact absence before downloading a model. Poll readiness after HTTP 202. Advanced operators may instead provide snapshot_path/manifest_path plus python_path for a checksum-pinned .nemo runtime. Setup may download/build; inference never does.",
           "requestBody": {
             "required": true,
             "content": {
@@ -5401,7 +5401,7 @@
           "tags": [
             "Audio"
           ],
-          "description": "Admin-only. Requires kind=acoustic|speaker|semantic in the query (canonical) or JSON body. acoustic reuses pinned JetPack YAMNet/TensorRT. speaker creates a private CPU-only WeSpeaker CAM++ venv with checksum-pinned NumPy/ONNX Runtime wheels and no Torch/Torchaudio dependency. semantic installs a checksum-locked CPython 3.10/aarch64 dependency closure plus the immutable CLAP revision while leaving vendor Torch untouched. Omnius never imports, links, replaces, or resolves a generic Torch package over JetPack Torch for either isolated runtime. JetPack daemon bootstrap provisions all three roles by default; OMNIUS_AUDIO_AUTO_SETUP=0 disables all and OMNIUS_SEMANTIC_AUDIO_AUTO_SETUP=0 disables CLAP. Semantic provisioning may finish under memory pressure, while activation/inference still requires the 8 GiB admission threshold and never evicts another workload. This is the only REST operation allowed to provision, and no role is substituted for another.",
+          "description": "Admin-only. Requires kind=acoustic|speaker|semantic in the query (canonical) or JSON body. acoustic reuses pinned JetPack YAMNet/TensorRT. speaker creates a private CPU-only WeSpeaker CAM++ venv with checksum-pinned NumPy/ONNX Runtime wheels and no Torch/Torchaudio dependency. semantic creates a private include-system-site-packages=false venv, explicitly links only the validated vendor CUDA Torch provider, and installs a checksum-locked CPython 3.10/aarch64 dependency closure (including Pillow and probed Transformers CLAP imports) plus the immutable CLAP revision. Omnius never replaces or resolves a generic Torch package over JetPack Torch. JetPack daemon bootstrap provisions all three roles by default; OMNIUS_AUDIO_AUTO_SETUP=0 disables all and OMNIUS_SEMANTIC_AUDIO_AUTO_SETUP=0 disables CLAP. Semantic provisioning may finish under memory pressure, while activation/inference still requires the 8 GiB admission threshold and never evicts another workload. CLAP readiness persists setup stage, failed_stage, complete subprocess stderr, and error. This is the only REST operation allowed to provision, and no role is substituted for another.",
           "parameters": [
             {
               "name": "kind",
@@ -16896,7 +16896,7 @@
           "tags": [
             "Vision"
           ],
-          "description": "Admin-only single-flight setup, also daemon-bootstrapped by default on JetPack (OMNIUS_VISION_AUTO_SETUP=0 disables). It creates a --system-site-packages venv, inherits existing CUDA Torch and vendor torchvision, installs a fully pinned non-Torch wheel closure with --no-deps, and then validates the managed runtime with user-site packages disabled. It downloads the OpenCLIP checkpoint from immutable revision 1a25a446712ba5ee05982a381eed697ef9b435cf with a fixed SHA-256. It never resolves Torch/torchvision from generic PyPI. A checksum-pinned local torchvision wheel override remains supported. Provisioning completes despite transient memory pressure; only model loading/inference applies the 8 GiB admission gate.",
+          "description": "Admin-only single-flight setup, also daemon-bootstrapped by default on JetPack (OMNIUS_VISION_AUTO_SETUP=0 disables). It creates an include-system-site-packages=false venv, explicitly links only the validated vendor CUDA Torch provider, installs a fully pinned non-Torch wheel closure including wcwidth with --no-deps, and then validates Torch, vendor torchvision, and OpenCLIP with user-site packages disabled. It downloads the OpenCLIP checkpoint from immutable revision 1a25a446712ba5ee05982a381eed697ef9b435cf with a fixed SHA-256. It never resolves Torch/torchvision from generic PyPI. A checksum-pinned local torchvision wheel override remains supported. Provisioning completes despite transient memory pressure; only model loading/inference applies the OpenCLIP-specific OMNIUS_VISION_MIN_AVAILABLE_MB admission gate.",
           "responses": {
             "200": {
               "description": "Provisioned; readiness may still report memory-blocked until model loading is admitted"

package/docs/reference/configuration.md

@@ -27,7 +27,9 @@ Configuration is layered from environment, global user settings, project setting
 | `OMNIUS_SEMANTIC_AUDIO_AUTO_SETUP` | Set to `0` to disable CLAP setup specifically; JetPack default is enabled |
 | `OMNIUS_VISION_AUTO_SETUP` | Set to `0` to disable managed OpenCLIP provisioning on JetPack |
 | `OMNIUS_VISION_PYTHON` | Explicit vendor CUDA Python for the isolated OpenCLIP runtime |
+| `OMNIUS_VISION_MIN_AVAILABLE_MB` | OpenCLIP-specific Jetson unified-memory admission threshold (default `8192`) |
 | `OMNIUS_NEMO_SPEECH_BIN` | Optional existing CUDA-enabled NeMo-Speech.cpp binary; otherwise live diarization builds the pinned source revision |
+| `OMNIUS_DIAR_NVCC` | Optional absolute JetPack CUDA compiler; discovery otherwise checks `/usr/local/cuda-12.2/bin/nvcc` then `/usr/local/cuda/bin/nvcc` |
 | `OMNIUS_DIARIZATION_AUTO_SETUP` | Set to `0` to disable managed diarizer provisioning; live Sortformer defaults on for JetPack |
 | `OMNIUS_HF_TOKEN` | Setup-only Hugging Face token for gated Community-1 download; never persisted or passed to inference |
 | `OMNIUS_PYANNOTE_TERMS_ACCEPTED` | Set to `1` only after accepting Community-1 terms; permits gated daemon bootstrap when a token is present |
@@ -52,7 +54,8 @@ upstream provider credentials.
 
 Linux installations load optional daemon-only overrides from
 `~/.config/omnius/daemon.env`. The postinstall creates this file with mode
-`0600` and the systemd unit references it with an optional `EnvironmentFile`.
+`0600` and creates or rewrites the systemd unit with
+`EnvironmentFile=%h/.config/omnius/daemon.env`.
 This is the appropriate place for `OMNIUS_AUDIO_PYTHON` or a setup-only gated
 model token; credentials do not need to be embedded in the unit or sent in a
 REST request.

package/docs/rest/endpoints/voice-vision.md

@@ -81,9 +81,10 @@ descriptive claims or invoke an LLM.
 First poll `GET /v1/vision/embed/readiness`. It is non-mutating and returns
 HTTP 503 until the isolated runtime and checksum-manifested weights are ready.
 Use the admin-scoped `POST /v1/vision/embed/setup` to create the
-`--system-site-packages` runtime under
-`~/.omnius/runtimes/vision/open-clip`, install pinned non-Torch dependencies,
-and fetch/verify the checkpoint from immutable revision
+private runtime under `~/.omnius/runtimes/vision/open-clip` with
+`include-system-site-packages = false`, link only the already-validated vendor
+Torch provider, install the pinned non-Torch closure (including `wcwidth`), and
+fetch/verify the checkpoint from immutable revision
 `1a25a446712ba5ee05982a381eed697ef9b435cf`. Setup is single-flight and
 daemon-bootstrapped by default on JetPack; set `OMNIUS_VISION_AUTO_SETUP=0`
 to disable it. Inference never installs packages or downloads artifacts, and
@@ -95,8 +96,10 @@ present in the compatible vendor stack or be supplied explicitly with the local
 `OMNIUS_VISION_TORCHVISION_WHEEL_SHA256`; generic PyPI Torch/torchvision is
 blocked. Bootstrap discovery may inspect the selected vendor interpreter, but
 the managed runtime installs its own pinned support closure and proves Torch,
-torchvision, and OpenCLIP with user-site packages disabled. On memory-constrained
-Jetson systems model loading and embedding fail with a typed 503 rather than
+torchvision, and OpenCLIP with user-site packages disabled. OpenCLIP reports
+its own `component: openclip` memory-admission result, governed by
+`OMNIUS_VISION_MIN_AVAILABLE_MB`, rather than reusing CLAP telemetry. On
+memory-constrained Jetson systems model loading and embedding fail with a typed 503 rather than
 evicting a resident ASR/Ollama model. Dependency and
 weight provisioning itself is allowed to finish under transient pressure;
 readiness separately reports `installed`, `weightsReady`,
@@ -280,7 +283,10 @@ startup provisions acoustic, speaker, and semantic roles by default; set
 `OMNIUS_SEMANTIC_AUDIO_AUTO_SETUP=0` to disable CLAP specifically.
 
 CLAP provisioning installs a checksum-locked CPython 3.10/aarch64 wheel
-closure and the immutable model revision without loading the model. It is
+closure—including Pillow and the complete probed Transformers CLAP import
+surface—inside a private venv with `include-system-site-packages = false`.
+Only the validated vendor CUDA Torch site is linked explicitly. The immutable
+model revision is provisioned without loading the model. It is
 allowed to finish while unified memory is busy. Worker activation and
 inference retain the 8 GiB admission gate and idle eviction, so setup cannot
 silently evict ASR, Ollama, or another CUDA workload.
@@ -344,8 +350,11 @@ exact remediation. Inference is likewise non-provisioning and returns typed
 `speaker_diarization_runtime_unavailable` until admin setup succeeds.
 
 Admin setup is asynchronous and single-flight. It returns HTTP 202 while it
-provisions, and readiness exposes the exact phase or terminal error. Existing
-ready workers return HTTP 200. Inference never performs these setup actions.
+provisions, and readiness exposes `setup.stage`, `setup.failed_stage`, the
+complete captured setup `stderr`, and the terminal error. Those diagnostics
+are persisted under the managed runtime so they survive daemon restart.
+Existing ready workers return HTTP 200. Inference never performs these setup
+actions.
 
 On JetPack, live setup with an empty object downloads the immutable,
 checksum-pinned Q8 Sortformer artifact and builds NVIDIA NeMo-Speech.cpp at a
@@ -360,7 +369,9 @@ POST /v1/audio/diarization/live/setup
 The native build requires the JetPack CUDA compiler plus `git` and a C++17
 compiler. Omnius installs pinned CMake/Ninja only in its private build-tools
 venv and never invokes `sudo`; missing native prerequisites are reported by
-readiness for operator installation outside inference. Set
+readiness for operator installation outside inference. CUDA compiler discovery
+checks `OMNIUS_DIAR_NVCC`, `/usr/local/cuda-12.2/bin/nvcc`, then
+`/usr/local/cuda/bin/nvcc`, and reports those exact locations when none exists. Set
 `OMNIUS_NEMO_SPEECH_BIN` to reuse a prebuilt CUDA-enabled binary.
 
 Managed Community-1 setup creates a separate CPU-only CPython environment, so