omnius 1.0.333 → 1.0.334

package/dist/index.js

@@ -279541,19 +279541,26 @@ import { execFileSync as execFileSync6 } from "node:child_process";
 import { existsSync as existsSync42, readFileSync as readFileSync31, writeFileSync as writeFileSync21, mkdirSync as mkdirSync23, statfsSync as statfsSync3 } from "node:fs";
 import { join as join51 } from "node:path";
 import { cpus, totalmem as totalmem2 } from "node:os";
+function gpuComputeUsable(computeCap) {
+  if (!computeCap)
+    return true;
+  const v = parseFloat(computeCap);
+  return Number.isFinite(v) ? v >= MIN_COMPUTE_CAP : true;
+}
 function bytesToGiB(bytes) {
   return Math.round(bytes / 1024 ** 3 * 10) / 10;
 }
 function probeGpus() {
   try {
-    const out = execFileSync6("nvidia-smi", ["--query-gpu=index,name,memory.total,memory.free", "--format=csv,noheader,nounits"], { encoding: "utf8", timeout: 5e3, stdio: ["ignore", "pipe", "ignore"] });
+    const out = execFileSync6("nvidia-smi", ["--query-gpu=index,name,memory.total,memory.free,compute_cap", "--format=csv,noheader,nounits"], { encoding: "utf8", timeout: 5e3, stdio: ["ignore", "pipe", "ignore"] });
     return out.trim().split("\n").map((line) => {
-      const [index, name10, total, free] = line.split(",").map((s2) => s2.trim());
+      const [index, name10, total, free, cap] = line.split(",").map((s2) => s2.trim());
       return {
         index: parseInt(index || "0", 10),
         name: name10 || "GPU",
         memTotalMiB: parseInt(total || "0", 10),
-        memFreeMiB: parseInt(free || "0", 10)
+        memFreeMiB: parseInt(free || "0", 10),
+        computeCap: cap || ""
       };
     }).filter((g) => Number.isFinite(g.memTotalMiB) && g.memTotalMiB > 0);
   } catch {
@@ -279577,9 +279584,10 @@ function diskFreeGiB(path12) {
   }
 }
 function selectStack(facts) {
-  const usable = facts.gpus.filter((g) => g.memTotalMiB >= USABLE_GPU_MIN_MIB);
+  const usable = facts.gpus.filter((g) => g.memTotalMiB >= USABLE_GPU_MIN_MIB && gpuComputeUsable(g.computeCap));
   const maxVramGiB = usable.length > 0 ? Math.round(Math.max(...usable.map((g) => g.memTotalMiB)) / 1024) : 0;
   const gpuCount = usable.length;
+  const cudaVisibleDevices = usable.map((g) => g.index).join(",");
   if (gpuCount === 0) {
     return {
       stack: "edge",
@@ -279587,7 +279595,8 @@ function selectStack(facts) {
       precision: "fp32",
       vramGiB: 0,
       gpuCount: 0,
-      reason: `No usable GPU (>=8 GiB) detected; CPU edge stack on ${facts.cpuCores} cores / ${facts.ramGiB} GiB RAM.`
+      cudaVisibleDevices: "",
+      reason: `No usable GPU (>=8 GiB, compute>=7.0) detected; CPU edge stack on ${facts.cpuCores} cores / ${facts.ramGiB} GiB RAM.`
     };
   }
   const precision = maxVramGiB >= 24 ? "bf16" : "fp16";
@@ -279598,7 +279607,8 @@ function selectStack(facts) {
       precision,
       vramGiB: maxVramGiB,
       gpuCount,
-      reason: `Forensic stack: ${gpuCount} usable GPU(s), max ${maxVramGiB} GiB VRAM — multi-pass + thinking-variant LALM + high-res reinspection.`
+      cudaVisibleDevices,
+      reason: `Forensic stack: ${gpuCount} usable GPU(s) [${cudaVisibleDevices}], max ${maxVramGiB} GiB VRAM — multi-pass + thinking-variant LALM + high-res reinspection.`
     };
   }
   if (maxVramGiB >= 16) {
@@ -279608,7 +279618,8 @@ function selectStack(facts) {
       precision,
       vramGiB: maxVramGiB,
       gpuCount,
-      reason: `Baseline stack: one ${maxVramGiB} GiB GPU — V-JEPA2-L/VideoPrism + YOLOE+SAM2 + event-triggered 7B VLM/LALM.`
+      cudaVisibleDevices,
+      reason: `Baseline stack: one ${maxVramGiB} GiB GPU [${cudaVisibleDevices}] — V-JEPA2-L/VideoPrism + YOLOE+SAM2 + event-triggered 7B VLM/LALM.`
     };
   }
   return {
@@ -279617,7 +279628,8 @@ function selectStack(facts) {
     precision,
     vramGiB: maxVramGiB,
     gpuCount,
-    reason: `Edge stack on GPU: only ${maxVramGiB} GiB VRAM — keyframe detectors + lite encoders, small models on event windows.`
+    cudaVisibleDevices,
+    reason: `Edge stack on GPU [${cudaVisibleDevices}]: only ${maxVramGiB} GiB VRAM — keyframe detectors + lite encoders, small models on event windows.`
   };
 }
 function adapterPlacement(cap, role) {
@@ -279664,11 +279676,12 @@ function getMediaCapability(opts = {}) {
   }
   return cap;
 }
-var USABLE_GPU_MIN_MIB, CACHE_FILE2, CACHE_TTL_MS2;
+var MIN_COMPUTE_CAP, USABLE_GPU_MIN_MIB, CACHE_FILE2, CACHE_TTL_MS2;
 var init_media_capability = __esm({
   "packages/execution/dist/av/media-capability.js"() {
     "use strict";
     init_model_store();
+    MIN_COMPUTE_CAP = 7;
     USABLE_GPU_MIN_MIB = 8 * 1024;
     CACHE_FILE2 = "av-capability.json";
     CACHE_TTL_MS2 = 60 * 60 * 1e3;
@@ -553530,6 +553543,7 @@ __export(dist_exports2, {
   gigabytesToBytes: () => gigabytesToBytes,
   globalMediaDir: () => globalMediaDir,
   globalMediaRootDir: () => globalMediaRootDir,
+  gpuComputeUsable: () => gpuComputeUsable,
   hasMaterialAmbiguity: () => hasMaterialAmbiguity,
   hashGeneratedArtifactContent: () => hashGeneratedArtifactContent,
   hashProcessCommand: () => hashProcessCommand,

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.333",
+  "version": "1.0.334",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.333",
+      "version": "1.0.334",
       "bundleDependencies": [
         "image-to-ascii"
       ],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.333",
+  "version": "1.0.334",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",