offgrid-ai 0.11.0 → 0.11.3

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "offgrid-ai",
-  "version": "0.11.0",
+  "version": "0.11.3",
   "description": "Privacy-first CLI for running local LLMs — discover, configure, run, benchmark",
   "author": "Eeshan Srivastava (https://eeshans.com)",
   "type": "module",

package/src/backends.mjs

@@ -1,7 +1,7 @@
 import { findLlamaServer } from "./config.mjs";
 import { scanGgufModels } from "./scan.mjs";
 import { parseModelName } from "./model-name.mjs";
-import { scanMlxModels, scanOmlxModelSizes, lookupOmlxModelSize } from "./mlx-discovery.mjs";
+import { scanMlxModels, scanOmlxModelSizes, lookupOmlxModelInfo } from "./mlx-discovery.mjs";
 import { DEFAULT_PORT as MLX_VLM_PORT } from "./mlx-flags.mjs";
 
 // ── Backend definitions ────────────────────────────────────────────────────
@@ -96,19 +96,23 @@ async function scanOmlxModels() {
   const body = await response.json();
   if (!Array.isArray(body?.data)) return [];
 
-  // The oMLX API doesn't return model sizes — look them up from disk.
-  const sizeMap = await scanOmlxModelSizes();
+  // The oMLX API doesn't return model sizes or publishers — look them up from disk.
+  const infoMap = await scanOmlxModelSizes();
 
   return body.data
     .filter((model) => isChatOmlxModel(model))
     .map((model) => {
-      const sizeFromDisk = lookupOmlxModelSize(model.id, sizeMap);
-      const parsed = parseModelName(model.id, "omlx");
+      const info = lookupOmlxModelInfo(model.id, infoMap);
+      // If the API ID doesn't already include a publisher (no / or --),
+      // prepend the publisher found on disk.
+      const hasPublisher = model.id.includes("/") || model.id.includes("--");
+      const fullName = (!hasPublisher && info?.publisher) ? `${info.publisher}/${model.id}` : model.id;
+      const parsed = parseModelName(fullName, "omlx");
       return {
         id: model.id,
         label: parsed.display,
         aliasSuggestion: model.id,
-        sizeBytes: sizeFromDisk ?? (model.size ?? 0),
+        sizeBytes: info?.sizeBytes ?? (model.size ?? 0),
         contextLength: model.max_model_len ?? null,
         quant: parsed.quant,
         family: null,

package/src/mlx-discovery.mjs

@@ -308,14 +308,14 @@ export function defaultMlxContextLength(trainedCtx, ramGb) {
 
 /**
  * Scan the oMLX models directory (~/.omlx/models/) for MLX model directories
- * and return a Map of basename → sizeBytes.  The oMLX API doesn't return model
- * sizes, so we compute them from the safetensors files on disk.
+ * and return a Map of basename → { sizeBytes, publisher }.  The oMLX API
+ * doesn't return model sizes or publishers, so we compute them from disk.
  */
 export async function scanOmlxModelSizes() {
   if (!existsSync(OMLX_MODELS_DIR)) return new Map();
-  const sizeByBasename = new Map();
+  const infoByBasename = new Map();
 
-  async function walk(dir) {
+  async function walk(dir, publisher) {
     let entries;
     try {
       entries = await readdir(dir, { withFileTypes: true });
@@ -327,26 +327,27 @@ export async function scanOmlxModelSizes() {
       const fullPath = join(dir, entry.name);
       if (await isMlxModelDir(fullPath)) {
         const sizeBytes = await getMlxDirSizeBytes(fullPath);
-        if (sizeBytes > 0) sizeByBasename.set(entry.name, sizeBytes);
+        if (sizeBytes > 0) infoByBasename.set(entry.name, { sizeBytes, publisher });
       } else {
-        await walk(fullPath);
+        // First-level directories under ~/.omlx/models/ are publishers
+        await walk(fullPath, publisher ?? entry.name);
       }
     }
   }
 
-  await walk(OMLX_MODELS_DIR);
-  return sizeByBasename;
+  await walk(OMLX_MODELS_DIR, null);
+  return infoByBasename;
 }
 
 /**
- * Look up a model's size by its oMLX API id.  Tries exact match, then the
+ * Look up a model's info by its oMLX API id.  Tries exact match, then the
  * segment after `--` (oMLX org--name format), then after `/` (HF format).
  */
-export function lookupOmlxModelSize(modelId, sizeMap) {
-  if (sizeMap.has(modelId)) return sizeMap.get(modelId);
+export function lookupOmlxModelInfo(modelId, infoMap) {
+  if (infoMap.has(modelId)) return infoMap.get(modelId);
   const dashIdx = modelId.indexOf("--");
-  if (dashIdx >= 0 && sizeMap.has(modelId.slice(dashIdx + 2))) return sizeMap.get(modelId.slice(dashIdx + 2));
+  if (dashIdx >= 0 && infoMap.has(modelId.slice(dashIdx + 2))) return infoMap.get(modelId.slice(dashIdx + 2));
   const slashIdx = modelId.indexOf("/");
-  if (slashIdx >= 0 && sizeMap.has(modelId.slice(slashIdx + 1))) return sizeMap.get(modelId.slice(slashIdx + 1));
+  if (slashIdx >= 0 && infoMap.has(modelId.slice(slashIdx + 1))) return infoMap.get(modelId.slice(slashIdx + 1));
   return null;
 }

package/src/model-name.mjs

@@ -195,6 +195,9 @@ function titleCaseModel(name) {
   // Title-case standalone aXb patterns (A3B, A12B — active parameters)
   result = result.replace(/\ba(\d+)\s*b\b/gi, (_, num) => `A${num}B`);
 
+  // Title-case eXb patterns (E2B, E12B — effective parameters)
+  result = result.replace(/\be(\d+)\s*b\b/gi, (_, num) => `E${num}B`);
+
   // Clean up extra spaces
   result = result.replace(/\s{2,}/g, " ").trim();
 

package/src/runtime.mjs

@@ -26,7 +26,7 @@ export async function offerManagedLlamaRuntimeUpdate(prompt, { fetchImpl = globa
     ["Source", "official GitHub release binary"],
   ]), { formatBorder: pc.cyan }));
 
-  const shouldInstall = await prompt.yesNo(installed ? "Update llama.cpp runtime?" : "Install llama.cpp runtime?", true);
+  const shouldInstall = await prompt.yesNo(installed ? "Update llama.cpp runtime?" : "Install llama.cpp runtime?", false);
   if (!shouldInstall) return false;
 
   await installLlamaRelease(latest, { fetchImpl });