omnius 1.0.511 → 1.0.512

package/dist/index.js

@@ -31906,7 +31906,7 @@ async function _collectSysMetrics() {
 	    } catch {}
 	    try {
       var cp = await import('node:child_process');
-      var smiOut = cp.execSync('nvidia-smi --query-gpu=utilization.gpu,memory.used,memory.total,name --format=csv,noheader,nounits 2>/dev/null', { encoding: 'utf8', timeout: 3000 });
+      var smiOut = cp.execSync('nvidia-smi --query-gpu=utilization.gpu,memory.used,memory.total,name --format=csv,noheader,nounits 2>/dev/null', { encoding: 'utf8', timeout: 8000 });
       var smiLine = smiOut.trim().split('\\n')[0];
       if (smiLine) {
         var sp = smiLine.split(',').map(function(s) { return s.trim(); });
@@ -31917,7 +31917,21 @@ async function _collectSysMetrics() {
         gpuInfo.name = sp[3] || '';
         gpuInfo.vramUtilization = gpuInfo.vramTotalMB > 0 ? Math.round((gpuInfo.vramUsedMB / gpuInfo.vramTotalMB) * 100) : 0;
       }
-    } catch {}
+    } catch {
+      // nvidia-smi can be killed or time out under tight memory pressure even
+      // when a GPU is present (e.g. Ollama actively holding VRAM). Fall back to a
+      // cheap /dev/nvidia* node probe so we don't falsely report the GPU as
+      // unavailable on systems that clearly have an NVIDIA device.
+      try {
+        var fsProbe = await import('node:fs');
+        var devEntries = fsProbe.readdirSync('/dev');
+        var nvidiaPresent = devEntries.some(function(d) { return /^nvidiad+$/.test(d); });
+        if (nvidiaPresent) {
+          gpuInfo.available = true;
+          if (!gpuInfo.name) gpuInfo.name = 'NVIDIA GPU (nvidia-smi probe failed)';
+        }
+      } catch {}
+    }
     _sysMetricsCache = {
 	      cpu: { utilization: Math.min(100, Math.round((loads[0] / cores) * 100)), cores: cores, model: cpuModel },
 	      memory: { utilization: Math.round((usedMem / totalMem) * 100), totalGB: Math.round((totalMem / (1024*1024*1024)) * 10) / 10, usedGB: Math.round((usedMem / (1024*1024*1024)) * 10) / 10 },
@@ -34279,7 +34293,7 @@ async function handleCmd(cmd) {
 	              } catch {}
               try {
                 var cp = await import('node:child_process');
-                var smiOut = cp.execSync('nvidia-smi --query-gpu=utilization.gpu,memory.used,memory.total,name --format=csv,noheader,nounits 2>/dev/null', { encoding: 'utf8', timeout: 3000 });
+                var smiOut = cp.execSync('nvidia-smi --query-gpu=utilization.gpu,memory.used,memory.total,name --format=csv,noheader,nounits 2>/dev/null', { encoding: 'utf8', timeout: 8000 });
                 var smiLine = smiOut.trim().split('\\n')[0];
                 if (smiLine) {
                   var sp = smiLine.split(',').map(function(s) { return s.trim(); });
@@ -34290,7 +34304,27 @@ async function handleCmd(cmd) {
                   gpuInfo.name = sp[3] || '';
                   gpuInfo.vramUtilization = gpuInfo.vramTotalMB > 0 ? Math.round((gpuInfo.vramUsedMB / gpuInfo.vramTotalMB) * 100) : 0;
                 }
-              } catch (ge) { /* no GPU */ }
+              } catch {
+                // nvidia-smi can be killed or time out under tight memory pressure
+                // even when a GPU is present (e.g. Ollama holding VRAM). Fall back to
+                // a cheap /dev/nvidia* node probe so we don't falsely report the GPU
+                // as unavailable on systems that clearly have an NVIDIA device.
+                try {
+                  var fsProbe = await import('node:fs');
+                  var devEntries = fsProbe.readdirSync('/dev');
+                  var nvidiaPresent = devEntries.some(function(d) { return /^nvidiad+$/.test(d); });
+                  // Also accept the nvidia-smi binary existing on PATH as a
+                  // permissive signal — on tight-memory systems the driver is
+                  // loaded (so the binary is present) even when a live query is
+                  // killed. Either signal means a GPU is genuinely available.
+                  var smiBin = '';
+                  try { smiBin = cp.execSync('command -v nvidia-smi 2>/dev/null', { encoding: 'utf8', timeout: 3000 }).trim(); } catch {}
+                  if (nvidiaPresent || !!smiBin) {
+                    gpuInfo.available = true;
+                    if (!gpuInfo.name) gpuInfo.name = 'NVIDIA GPU (nvidia-smi probe failed)';
+                  }
+                } catch {}
+              }
               var cpuModel = '';
               try { var cpuInfoArr = os.cpus(); if (cpuInfoArr.length > 0) cpuModel = cpuInfoArr[0].model || ''; } catch {}
 	              var metricsPayload = {
@@ -602900,19 +602934,22 @@ ${description}`
         }
         const requestMessages = effectiveThink ? cleanedMessages : injectNoThinkDirective(cleanedMessages);
         const responseFormat = request.responseFormat ?? request.response_format;
+        const isOllama = shouldUseOllamaPoolForBaseUrl(this.baseUrl);
         const body = {
           model: this.model,
           messages: requestMessages,
           tools: request.tools,
           temperature: request.temperature,
-          max_tokens: effectiveMaxTokens,
-          think: effectiveThink
+          max_tokens: effectiveMaxTokens
         };
+        if (isOllama) {
+          body["think"] = effectiveThink;
+        }
         if (responseFormat !== void 0) {
           body["response_format"] = responseFormat;
         }
         const reqNumCtx = request.numCtx ?? request.num_ctx;
-        if (Number.isFinite(reqNumCtx) && (reqNumCtx ?? 0) > 0) {
+        if (isOllama && Number.isFinite(reqNumCtx) && (reqNumCtx ?? 0) > 0) {
           const opts = body["options"] ?? {};
           opts["num_ctx"] = reqNumCtx;
           body["options"] = opts;

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.511",
+  "version": "1.0.512",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.511",
+      "version": "1.0.512",
       "bundleDependencies": [
         "image-to-ascii"
       ],
@@ -2579,9 +2579,9 @@
       }
     },
     "node_modules/bare-path": {
-      "version": "3.1.0",
-      "resolved": "https://registry.npmjs.org/bare-path/-/bare-path-3.1.0.tgz",
-      "integrity": "sha512-322oSBHOhMhOm2CVwn7b3+gWQqwzSgIY4gNpKKb+Ha5jx2dEl9ClOgz/SK57DwuDql8YQg8JRpb9U0o8K3pW9g==",
+      "version": "3.1.1",
+      "resolved": "https://registry.npmjs.org/bare-path/-/bare-path-3.1.1.tgz",
+      "integrity": "sha512-JprUlveX3QjApC1cTpsUOiscADftCGVWkzitbHsRqv84hzYwYHw2mbluddsq5TvI8mH/8Ov1f4BiMAdcB0oYnQ==",
       "license": "Apache-2.0",
       "optional": true
     },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.511",
+  "version": "1.0.512",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",