omnius 1.0.626 → 1.0.627

package/dist/index.js

@@ -2843,21 +2843,54 @@ function classifyNvidiaArmEnvironmentSync(preferredPython = "python3", env2 = pr
   const l4tRelease = l4tMatch ? `${l4tMatch[1]}.${l4tMatch[2]}` : void 0;
   const l4tMajor = Number(l4tMatch?.[1] ?? 0);
   const l4tRevision = Number(l4tMatch?.[2] ?? 0);
-  const jetPackFamily = l4tMajor >= 36 ? l4tRevision >= 4 ? "6.2+" : "6.0/6.1" : l4tMajor === 35 ? "5.x" : void 0;
+  const jetPackMajor = l4tMajor >= 38 ? 7 : l4tMajor >= 36 ? 6 : l4tMajor >= 34 ? 5 : l4tMajor === 32 ? 4 : void 0;
+  const jetPackPackage = run("dpkg-query", ["-W", "-f=${Version}", "nvidia-jetpack"], 15e3, env2);
+  const jetPackVersion = jetPackPackage.ok ? jetPackPackage.output.match(/\d+\.\d+(?:\.\d+)?/)?.[0] : void 0;
+  const jetPackFamily = l4tMajor >= 38 ? "7.x" : l4tMajor >= 36 ? l4tRevision >= 4 ? "6.2+" : "6.0/6.1" : l4tMajor === 35 ? "5.x" : l4tMajor === 32 ? "4.x" : void 0;
   const cudaToolkitVersion = detectCudaToolkitVersion(env2);
   const vendor = resolveVendorCudaPythonSync(preferredPython, env2);
   const pythonInfo = run(vendor.python || preferredPython, ["-c", "import sys;print(f'{sys.version_info.major}.{sys.version_info.minor} cp{sys.version_info.major}{sys.version_info.minor}')"], 3e4, env2);
   const pythonTokens = pythonInfo.ok ? pythonInfo.output.trim().split(/\s+/).slice(-2) : [];
   const pythonVersion = pythonTokens[0];
   const pythonAbi = pythonTokens[1];
-  const yuyiCompatible = nvidiaArm && family === "orin" && l4tMajor === 36 && l4tRevision >= 4 && cudaToolkitVersion?.startsWith("12.6") === true && pythonVersion === "3.10" && vendor.computeCapability === 8.7;
+  const managedTorchProviders = [];
+  const cudaIndexMatch = cudaToolkitVersion?.match(/^(\d+)\.(\d+)/);
+  if (nvidiaArm && jetPackMajor && cudaIndexMatch) {
+    const cudaIndexTag = `cu${cudaIndexMatch[1]}${cudaIndexMatch[2]}`;
+    if (jetPackMajor >= 6) {
+      managedTorchProviders.push({
+        id: "jetson-ai-lab-index",
+        mode: "index",
+        location: `https://pypi.jetson-ai-lab.io/jp${jetPackMajor}/${cudaIndexTag}`
+      });
+    }
+    if (jetPackVersion) {
+      managedTorchProviders.push({
+        id: "nvidia-jetpack-index",
+        mode: "find-links",
+        location: `https://developer.download.nvidia.com/compute/redist/jp/v${jetPackVersion.replace(/\D/g, "")}/pytorch/`
+      });
+    }
+  }
+  const yuyiCompatible = nvidiaArm && !vendor.ok && family === "orin" && l4tMajor === 36 && l4tRevision >= 4 && cudaToolkitVersion?.startsWith("12.6") === true && pythonVersion === "3.10" && (vendor.computeCapability === 8.7 || !vendor.ok && family === "orin");
   if (yuyiCompatible) {
+    managedTorchProviders.unshift({
+      id: "yuyirobotlab-orin-cu126-vllm-0.19.0",
+      mode: "wheel",
+      location: "https://github.com/YuyiRobotLab/vllm-orin-release/releases/download/v0.19.0-orin/torch-2.10.0-cp310-cp310-linux_aarch64.whl"
+    });
     reasons.push("Exact Orin SM 8.7 / L4T 36.4+ / CUDA 12.6 / CPython 3.10 provider match");
+  }
+  if (vendor.ok) {
+    reasons.push("No exact binary provider matches every hardware/runtime dimension");
+    reasons.push("CUDA-enabled vendor Torch is available for a source build");
     return {
       nvidiaArm,
       family,
       boardModel,
       l4tRelease,
+      jetPackMajor,
+      jetPackVersion,
       jetPackFamily,
       cudaToolkitVersion,
       pythonVersion,
@@ -2866,28 +2899,30 @@ function classifyNvidiaArmEnvironmentSync(preferredPython = "python3", env2 = pr
       vendorPython: vendor.python,
       vendorTorchVersion: vendor.torchVersion,
       vendorTorchCudaVersion: vendor.torchCudaVersion,
-      recommendedStrategy: "matched-binary",
-      matchedProvider: "yuyirobotlab-orin-cu126-vllm-0.19.0",
+      recommendedStrategy: "source-existing-torch",
+      managedTorchProviders,
       reasons
     };
   }
-  if (vendor.ok) {
-    reasons.push("No exact binary provider matches every hardware/runtime dimension");
-    reasons.push("CUDA-enabled vendor Torch is available for a source build");
+  if (managedTorchProviders.length > 0) {
+    reasons.push("No CUDA-enabled vendor Torch interpreter was found");
+    reasons.push(`Managed CUDA Torch providers: ${managedTorchProviders.map((provider) => provider.id).join(", ")}`);
     return {
       nvidiaArm,
       family,
       boardModel,
       l4tRelease,
+      jetPackMajor,
+      jetPackVersion,
       jetPackFamily,
       cudaToolkitVersion,
       pythonVersion,
       pythonAbi,
-      computeCapability: vendor.computeCapability,
+      computeCapability: family === "orin" ? 8.7 : vendor.computeCapability,
       vendorPython: vendor.python,
-      vendorTorchVersion: vendor.torchVersion,
-      vendorTorchCudaVersion: vendor.torchCudaVersion,
-      recommendedStrategy: "source-existing-torch",
+      recommendedStrategy: "managed-torch",
+      matchedProvider: managedTorchProviders[0]?.id,
+      managedTorchProviders,
       reasons
     };
   }
@@ -2897,6 +2932,8 @@ function classifyNvidiaArmEnvironmentSync(preferredPython = "python3", env2 = pr
     family,
     boardModel,
     l4tRelease,
+    jetPackMajor,
+    jetPackVersion,
     jetPackFamily,
     cudaToolkitVersion,
     pythonVersion,
@@ -2913,9 +2950,11 @@ function ensureCudaPythonVenvSync(options2) {
   const requireHost = options2.requireHostCudaTorch ?? nvidiaArm;
   const label = options2.label?.trim() || "managed CUDA Python runtime";
   const host2 = nvidiaArm ? resolveVendorCudaPythonSync(preferredPython, env2) : probeCudaTorchSync(preferredPython, env2);
+  const classification = nvidiaArm ? classifyNvidiaArmEnvironmentSync(preferredPython, env2) : void 0;
+  const managedProviders = classification?.managedTorchProviders ?? [];
   const bootstrapPython2 = host2.ok ? host2.python : preferredPython;
-  if (requireHost && !host2.ok) {
-    throw new Error(`${label} requires the host's vendor CUDA Torch; Omnius will not install generic PyPI Torch: ${host2.output}`);
+  if (requireHost && !host2.ok && managedProviders.length === 0) {
+    throw new Error(`${label} requires compatible CUDA Torch, but no vendor installation or environment-matched provider was found: ${host2.output}`);
   }
   let repaired = false;
   if (existsSync6(managedPython)) {
@@ -2947,14 +2986,62 @@ function ensureCudaPythonVenvSync(options2) {
     } catch {
     }
   }
+  let torchProvider;
+  if (!managed.ok && managedProviders.length > 0) {
+    const providerEnv = { ...env2 };
+    delete providerEnv["PIP_CONSTRAINT"];
+    const failures = [];
+    for (const provider of managedProviders) {
+      const providerArgs = provider.mode === "wheel" ? ["-m", "pip", "install", "--disable-pip-version-check", "--no-deps", "--force-reinstall", provider.location] : provider.mode === "index" ? ["-m", "pip", "install", "--disable-pip-version-check", "--upgrade", "--force-reinstall", "--index-url", provider.location, "torch"] : ["-m", "pip", "install", "--disable-pip-version-check", "--upgrade", "--force-reinstall", "--no-index", "--find-links", provider.location, "--no-deps", "torch"];
+      const installed = run(managedPython, providerArgs, 12e5, providerEnv);
+      if (!installed.ok) {
+        failures.push(`${provider.id}: ${installed.output}`);
+        continue;
+      }
+      const candidate = probeCudaTorchSync(managedPython, providerEnv);
+      if (!candidate.ok) {
+        failures.push(`${provider.id}: installed Torch failed CUDA validation: ${candidate.output}`);
+        continue;
+      }
+      managed = candidate;
+      torchProvider = provider.id;
+      break;
+    }
+    if (!managed.ok && requireHost) {
+      throw new Error(`${label} could not install CUDA Torch for ${classification?.jetPackFamily ?? "detected JetPack"} / CUDA ${classification?.cudaToolkitVersion ?? "unknown"} / ${classification?.pythonAbi ?? "unknown ABI"}: ${failures.join("; ")}`);
+    }
+  }
+  const companions = [...new Set(options2.cudaCompanionPackages ?? [])].filter((name10) => name10.trim() && name10.trim().toLowerCase() !== "torch");
+  if (managed.ok && nvidiaArm && companions.length > 0) {
+    const companionImports = companions.map((name10) => name10.replace(/-/g, "_"));
+    const importProbe = () => run(managedPython, ["-c", `import ${companionImports.join(",")}`], 6e4, env2);
+    if (!importProbe().ok) {
+      const failures = [];
+      for (const provider of managedProviders.filter((candidate) => candidate.mode !== "wheel")) {
+        const providerArgs = provider.mode === "index" ? ["-m", "pip", "install", "--disable-pip-version-check", "--no-deps", "--upgrade", "--index-url", provider.location, ...companions] : ["-m", "pip", "install", "--disable-pip-version-check", "--no-deps", "--upgrade", "--no-index", "--find-links", provider.location, ...companions];
+        const installed = run(managedPython, providerArgs, 12e5, env2);
+        if (installed.ok && importProbe().ok)
+          break;
+        failures.push(`${provider.id}: ${installed.output}`);
+      }
+      const verified = importProbe();
+      if (!verified.ok && requireHost) {
+        throw new Error(`${label} could not resolve CUDA companion packages (${companions.join(", ")}) for ${classification?.jetPackFamily ?? "detected JetPack"}: ${failures.join("; ") || verified.output}`);
+      }
+    }
+  }
   if (requireHost && !managed.ok) {
     throw new Error(`${label} cannot inherit CUDA Torch from ${bootstrapPython2}: ${managed.output}`);
   }
   const pipEnv = { ...env2 };
-  if (host2.ok && host2.torchVersion) {
+  const protectedTorch = managed.ok ? managed : host2;
+  if (protectedTorch.ok && protectedTorch.torchVersion) {
     const constraintPath = join6(options2.venvDir, "omnius-cuda-constraints.txt");
-    writeFileSync5(constraintPath, `torch===${host2.torchVersion}
-# Omnius protects vendor CUDA Torch from pip replacement.
+    writeFileSync5(constraintPath, `torch===${protectedTorch.torchVersion}
+# Omnius protects environment-matched CUDA Torch from pip replacement.
+`, "utf8");
+    writeFileSync5(join6(options2.venvDir, process.platform === "win32" ? "pip.ini" : "pip.conf"), `[global]
+constraint = ${constraintPath}
 `, "utf8");
     pipEnv["PIP_CONSTRAINT"] = constraintPath;
   }
@@ -2966,7 +3053,8 @@ function ensureCudaPythonVenvSync(options2) {
     repaired,
     host: host2,
     managed,
-    pipEnv
+    pipEnv,
+    torchProvider
   };
 }
 function packageName(spec) {
@@ -92845,6 +92933,7 @@ var init_transcribe_tool = __esm({
     init_cuda_device_filter();
     init_process_async();
     init_venv_paths();
+    init_python_cuda_runtime();
     init_transcribe_python_runtime();
     AUDIO_EXTS = /* @__PURE__ */ new Set([
       ".mp3",
@@ -93138,20 +93227,19 @@ var init_transcribe_tool = __esm({
         }
       }
       async ensureManagedWhisperRuntime() {
-        const python2 = venvPython(MANAGED_ASR_VENV);
-        if (!existsSync43(python2)) {
-          mkdirSync25(join44(homedir11(), ".omnius", "runtimes", "asr"), { recursive: true });
-          await execFileText2("python3", ["-m", "venv", MANAGED_ASR_VENV], {
-            timeout: 6e4,
-            env: transcriptionPythonEnv()
-          });
-        }
+        const runtime2 = ensureCudaPythonVenvSync({
+          venvDir: MANAGED_ASR_VENV,
+          bootstrapPython: process.env["OMNIUS_ASR_PYTHON"]?.trim() || "python3",
+          env: transcriptionPythonEnv(),
+          label: "managed Whisper ASR runtime"
+        });
+        const python2 = runtime2.python;
         if (_managedAsrReady)
           return python2;
         try {
           await execFileText2(python2, ["-c", "import whisper, numpy"], {
             timeout: 1e4,
-            env: transcriptionPythonEnv()
+            env: runtime2.pipEnv
           });
           _managedAsrReady = true;
           return python2;
@@ -93168,7 +93256,7 @@ var init_transcribe_tool = __esm({
             "openai-whisper"
           ], {
             timeout: 9e5,
-            env: transcriptionPythonEnv()
+            env: runtime2.pipEnv
           });
           _managedAsrReady = true;
           return python2;
@@ -602966,7 +603054,8 @@ async function ensureVibeVoiceAsrSetup() {
     bootstrapPython: bootstrapPython2,
     env: managedEnv(),
     label: "VibeVoice ASR runtime",
-    requireHostCudaTorch: true
+    requireHostCudaTorch: true,
+    cudaCompanionPackages: ["torchaudio"]
   });
   protectedPipEnv = cudaRuntime.pipEnv;
   protectedBootstrapPython = cudaRuntime.bootstrapPython;
@@ -730127,12 +730216,14 @@ function markerCurrent() {
 }
 async function ensurePythonRuntime() {
   const python2 = bootstrapPython();
-  if (!existsSync153(venvPython2())) {
-    const args = ["-m", "venv"];
-    if (isJetsonHost()) args.push("--system-site-packages");
-    args.push(venvDir());
-    await runCommand(python2, args, { timeoutMs: 18e4 });
-  }
+  const cudaRuntime = ensureCudaPythonVenvSync({
+    venvDir: venvDir(),
+    bootstrapPython: python2,
+    env: managedEnv2(),
+    label: "Voicebox TTS runtime",
+    requireHostCudaTorch: isNvidiaArmCudaHost(),
+    cudaCompanionPackages: isNvidiaArmCudaHost() ? ["torchaudio"] : []
+  });
   if (markerCurrent()) {
     await runCommand(
       venvPython2(),
@@ -730146,14 +730237,14 @@ async function ensurePythonRuntime() {
           "print('voicebox imports ok')"
         ].join("; ")
       ],
-      { cwd: repoDir(), timeoutMs: 12e4 }
+      { cwd: repoDir(), timeoutMs: 12e4, env: cudaRuntime.pipEnv }
     );
     return;
   }
   await runCommand(
     venvPython2(),
     ["-m", "pip", "install", "--quiet", "--upgrade", "pip", "wheel", "setuptools"],
-    { timeoutMs: 3e5 }
+    { timeoutMs: 3e5, env: cudaRuntime.pipEnv }
   );
   if (isJetsonHost()) {
     await runCommand(
@@ -730162,7 +730253,7 @@ async function ensurePythonRuntime() {
         "-c",
         "import torch; assert torch.cuda.is_available(), 'JetPack CUDA PyTorch is required'; print(torch.__version__)"
       ],
-      { timeoutMs: 12e4 }
+      { timeoutMs: 12e4, env: cudaRuntime.pipEnv }
     );
   } else {
     const torchArgs = ["-m", "pip", "install", "--quiet"];
@@ -730187,13 +730278,13 @@ async function ensurePythonRuntime() {
   await runCommand(
     venvPython2(),
     ["-m", "pip", "install", "--quiet", "-r", requirementsPath],
-    { cwd: repoDir(), timeoutMs: 12e5 }
+    { cwd: repoDir(), timeoutMs: 12e5, env: cudaRuntime.pipEnv }
   );
   for (const dependency of ["chatterbox-tts", "hume-tada"]) {
     await runCommand(
       venvPython2(),
       ["-m", "pip", "install", "--quiet", "--no-deps", dependency],
-      { cwd: repoDir(), timeoutMs: 6e5 }
+      { cwd: repoDir(), timeoutMs: 6e5, env: cudaRuntime.pipEnv }
     );
   }
   await runCommand(
@@ -730206,7 +730297,7 @@ async function ensurePythonRuntime() {
       "--no-deps",
       "git+https://github.com/QwenLM/Qwen3-TTS.git"
     ],
-    { cwd: repoDir(), timeoutMs: 6e5 }
+    { cwd: repoDir(), timeoutMs: 6e5, env: cudaRuntime.pipEnv }
   );
   await runCommand(
     venvPython2(),
@@ -730220,7 +730311,7 @@ async function ensurePythonRuntime() {
         "print('voicebox imports ok')"
       ].join("; ")
     ],
-    { cwd: repoDir(), timeoutMs: 12e4 }
+    { cwd: repoDir(), timeoutMs: 12e4, env: cudaRuntime.pipEnv }
   );
   const patch = packageAssetPath(
     "voicebox",
@@ -802282,22 +802373,25 @@ function tryInstallOrinRelease(hostTorch) {
   }
   const providerEnv = { ...process.env };
   delete providerEnv["PIP_CONSTRAINT"];
-  const torchInstall = run3(
-    python,
-    [
-      "-m",
-      "pip",
-      "install",
-      "--disable-pip-version-check",
-      "--no-deps",
-      "--force-reinstall",
-      ORIN_TORCH_WHEEL
-    ],
-    9e5,
-    { env: providerEnv }
-  );
-  if (!torchInstall.ok) return false;
-  const providerTorch = probeCudaTorchSync(python, providerEnv);
+  let providerTorch = probeCudaTorchSync(python, providerEnv);
+  if (!providerTorch.ok || providerTorch.computeCapability !== 8.7 || !String(providerTorch.torchCudaVersion || "").startsWith("12.6")) {
+    const torchInstall = run3(
+      python,
+      [
+        "-m",
+        "pip",
+        "install",
+        "--disable-pip-version-check",
+        "--no-deps",
+        "--force-reinstall",
+        ORIN_TORCH_WHEEL
+      ],
+      9e5,
+      { env: providerEnv }
+    );
+    if (!torchInstall.ok) return false;
+    providerTorch = probeCudaTorchSync(python, providerEnv);
+  }
   if (!providerTorch.ok || providerTorch.computeCapability !== 8.7 || !String(providerTorch.torchCudaVersion || "").startsWith("12.6")) {
     rmSync20(venv, { recursive: true, force: true });
     protectedPipEnv2 = {};
@@ -802344,9 +802438,9 @@ function ensureArm64Server() {
   const systemPython = managedCudaRuntime?.bootstrapPython || process.env["OMNIUS_PYTHON"]?.trim() || "python3";
   const hostTorch = managedCudaRuntime?.host ?? probeCudaTorchSync(systemPython);
   const managedTorch = probeCudaTorchSync(python, { ...process.env, ...protectedPipEnv2 });
-  if (!hostTorch.ok || !managedTorch.ok) {
+  if (!managedTorch.ok) {
     throw new Error(
-      `Jetson CUDA Torch preflight failed. Host: ${hostTorch.output}; managed: ${managedTorch.output}`
+      `Jetson CUDA Torch preflight failed after environment-specific provider resolution. Host: ${hostTorch.output}; managed: ${managedTorch.output}`
     );
   }
   if (tryInstallOrinRelease(hostTorch)) return;

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.626",
+  "version": "1.0.627",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.626",
+      "version": "1.0.627",
       "bundleDependencies": [
         "image-to-ascii"
       ],
@@ -4220,9 +4220,9 @@
       "license": "Apache-2.0 OR MIT"
     },
     "node_modules/ip-address": {
-      "version": "10.4.0",
-      "resolved": "https://registry.npmjs.org/ip-address/-/ip-address-10.4.0.tgz",
-      "integrity": "sha512-oSK96Grm3aP6OrS263xVxbNDGVL7rzBtYdpGqlDG8iQdoenDoTs/nkki+DflYbAEE8Xl6o5YxhxlrKvI3nqKXQ==",
+      "version": "10.5.0",
+      "resolved": "https://registry.npmjs.org/ip-address/-/ip-address-10.5.0.tgz",
+      "integrity": "sha512-R5SnVLJmgYYvf2F2ZgwSBnelz5G4q5AxIC277GDfUaNbrZKNANcBC7RHqYYePlszf4kBolVkJauG0ZjHHFh55g==",
       "license": "MIT",
       "engines": {
         "node": ">= 12"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.626",
+  "version": "1.0.627",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/library.js",