@vtstech/pi-shared 1.0.4-1 → 1.0.5

package/format.js

@@ -63,8 +63,57 @@ function sanitizeForReport(s, maxLines = 40) {
 function padRight(s, n) {
   return s + " ".repeat(Math.max(0, n - s.length));
 }
+function estimateVram(parameterSize, quantizationLevel) {
+  const params = parseParamCount(parameterSize);
+  if (params === void 0) return void 0;
+  const bitsPerParam = bitsPerParamForQuant(quantizationLevel);
+  const modelBytes = params * bitsPerParam / 8;
+  return Math.ceil(modelBytes * 1.1);
+}
+function parseParamCount(s) {
+  if (!s || typeof s !== "string") return void 0;
+  const str = s.trim().toLowerCase();
+  const match = str.match(/^([\d.]+)\s*([bmt]?|a(?:pple)?)$/);
+  if (!match) return void 0;
+  const num = parseFloat(match[1]);
+  if (isNaN(num) || num <= 0) return void 0;
+  const suffix = match[2];
+  switch (suffix) {
+    case "b":
+      return num * 1e9;
+    case "m":
+      return num * 1e6;
+    case "t":
+      return num * 1e12;
+    case "a":
+      return num * 1e9;
+    // Apple-style (e.g., "3a" = 3B parameters)
+    case "":
+      return num * 1e9;
+    // Bare number assumed to be billions
+    default:
+      return void 0;
+  }
+}
+function bitsPerParamForQuant(quant) {
+  const q = quant.toUpperCase().replace(/[-_.]/g, "");
+  if (q.startsWith("FP32") || q === "FP32") return 32;
+  if (q.startsWith("F16") || q === "F16" || q.startsWith("BF16")) return 16;
+  if (q.startsWith("Q8")) return 8;
+  if (q.startsWith("IQ4")) return 4.5;
+  if (q.startsWith("IQ3")) return 3.5;
+  if (q.startsWith("IQ2")) return 2.5;
+  if (q.startsWith("IQ1")) return 1.75;
+  if (q.startsWith("Q5") || q.startsWith("Q6")) return 5.5;
+  if (q.startsWith("Q4")) return 4.5;
+  if (q.startsWith("Q3")) return 3.5;
+  if (q.startsWith("Q2")) return 2.5;
+  if (q.startsWith("Q1")) return 1.75;
+  return 5;
+}
 export {
   bytesHuman,
+  estimateVram,
   fail,
   fmtBytes,
   fmtDur,

package/ollama.js

@@ -45,9 +45,44 @@ async function fetchOllamaModels(baseUrl) {
   const data = await res.json();
   return data.models ?? [];
 }
+async function fetchModelContextLength(baseUrl, modelName) {
+  try {
+    const res = await fetch(`${baseUrl}/api/show`, {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({ name: modelName }),
+      signal: AbortSignal.timeout(3e4)
+    });
+    if (!res.ok) return void 0;
+    const data = await res.json();
+    for (const key of Object.keys(data?.model_info ?? {})) {
+      if (key.endsWith(".context_length")) {
+        const val = data.model_info[key];
+        if (typeof val === "number") return val;
+      }
+    }
+    const numCtx = data?.model_info?.["num_ctx"];
+    if (typeof numCtx === "number") return numCtx;
+  } catch {
+    return void 0;
+  }
+}
+async function fetchContextLengthsBatched(baseUrl, modelNames, batchSize = 3) {
+  const result = /* @__PURE__ */ new Map();
+  for (let i = 0; i < modelNames.length; i += batchSize) {
+    const batch = modelNames.slice(i, i + batchSize);
+    const results = await Promise.allSettled(
+      batch.map((name) => fetchModelContextLength(baseUrl, name))
+    );
+    results.forEach((r, idx) => {
+      result.set(batch[idx], r.status === "fulfilled" ? r.value : void 0);
+    });
+  }
+  return result;
+}
 function isReasoningModel(name) {
   const lower = name.toLowerCase();
-  return lower.includes("deepseek-r1") || lower.includes("qwq") || lower.includes("o1") || lower.includes("o3") || lower.includes("think") || lower.includes("reason");
+  return lower.includes("deepseek-r1") || lower.includes("qwq") || lower.includes("o1") || lower.includes("o3") || lower.includes("qwen3") || lower.includes("think") || lower.includes("reason");
 }
 function detectModelFamily(modelName) {
   const name = modelName.toLowerCase();
@@ -82,6 +117,8 @@ function detectModelFamily(modelName) {
 export {
   MODELS_JSON_PATH,
   detectModelFamily,
+  fetchContextLengthsBatched,
+  fetchModelContextLength,
   fetchOllamaModels,
   getOllamaBaseUrl,
   isReasoningModel,

package/package.json

@@ -1,10 +1,10 @@
 {
   "name": "@vtstech/pi-shared",
-  "version": "1.0.4-1",
+  "version": "1.0.5",
   "description": "Shared utilities for Pi Coding Agent extensions",
   "main": "index.js",
   "types": "index.d.ts",
-  "keywords": ["pi-package", "pi-extensions"],
+  "keywords": ["pi-extensions"],
   "license": "MIT",
   "access": "public",
   "type": "module",