offgrid-ai 0.3.26 → 0.3.27

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "offgrid-ai",
-  "version": "0.3.26",
+  "version": "0.3.27",
   "description": "Privacy-first CLI for running local LLMs — discover, configure, run, benchmark",
   "author": "Eeshan Srivastava (https://eeshans.com)",
   "type": "module",

package/src/autodetect.mjs

@@ -6,6 +6,7 @@ import { readGgufMetadata } from "./gguf.mjs";
 
 export function detectCapabilities(modelPath, mmprojPath) {
   const meta = safeReadGgufMetadata(modelPath);
+  const mmprojMeta = mmprojPath ? safeReadGgufMetadata(mmprojPath) : {};
   const name = basename(modelPath).toLowerCase();
   const pathHints = String(modelPath).toLowerCase();
 
@@ -25,13 +26,14 @@ export function detectCapabilities(modelPath, mmprojPath) {
 
   // Vision — mmproj present
   const vision = Boolean(mmprojPath && existsSync(mmprojPath));
+  const mmprojProjectorType = stringMeta(mmprojMeta, "clip.vision.projector_type") ?? stringMeta(mmprojMeta, "clip.audio.projector_type") ?? null;
 
   // MTP (multi-token prediction) — detect speculative decoding.
   // Do not treat all Qwen models as MTP; require an explicit filename or metadata hint.
   const mtp = /\bmtp\b|draft-mtp|multi-token/i.test(pathHints) || Object.keys(meta).some((key) => /mtp|draft|speculative/i.test(key));
 
   // Quantization
-  const quant = name.match(/(Q\d_K_[A-Z]+|UD-[A-Z0-9_]+)/i)?.[1] ?? null;
+  const quant = name.match(/(Q\d_K_[A-Z]+|Q\d_[01]|UD-[A-Z0-9_]+)/i)?.[1] ?? null;
 
   // Context size from metadata, fallback to name hints
   const metaCtx = architecture
@@ -39,7 +41,7 @@ export function detectCapabilities(modelPath, mmprojPath) {
     : undefined;
   const ctxSize = metaCtx ?? (thinking ? 80000 : 32768);
 
-  return { architecture, thinking, vision, mtp, qat, imatrix, quant, metaCtx, ctxSize, meta };
+  return { architecture, thinking, vision, mtp, qat, imatrix, quant, metaCtx, ctxSize, meta, mmprojProjectorType };
 }
 
 // ── Compute llama-server flags from capabilities ───────────────────────────
@@ -127,4 +129,9 @@ function safeReadGgufMetadata(modelPath) {
 function numberMeta(meta, key) {
   const value = key ? meta[key] : undefined;
   return typeof value === "number" && Number.isFinite(value) ? value : undefined;
+}
+
+function stringMeta(meta, key) {
+  const value = key ? meta[key] : undefined;
+  return typeof value === "string" && value ? value : undefined;
 }

package/src/cli.mjs

@@ -456,6 +456,33 @@ function capabilitySummary(caps) {
   return parts.length > 0 ? parts.join(" · ") : "standard GGUF";
 }
 
+function isUnsupportedMmprojError(err, profile) {
+  const message = String(err?.message ?? "");
+  return Boolean(profile.mmprojPath && /unknown projector type|failed to load multimodal model|failed to load CLIP model/i.test(message));
+}
+
+function textOnlyProfile(profile) {
+  return normalizeProfile({
+    ...profile,
+    mmprojPath: null,
+    disabledMmprojPath: profile.disabledMmprojPath ?? profile.mmprojPath,
+    capabilities: { ...(profile.capabilities ?? {}), vision: false, visionDisabledReason: "unsupported-mmproj" },
+    commandArgv: removeCommandOption(profile.commandArgv ?? [], "--mmproj"),
+  });
+}
+
+function removeCommandOption(argv, flag) {
+  const next = [];
+  for (let i = 0; i < argv.length; i++) {
+    if (argv[i] === flag) {
+      if (argv[i + 1] && !argv[i + 1].startsWith("--")) i += 1;
+      continue;
+    }
+    next.push(argv[i]);
+  }
+  return next;
+}
+
 function createManagedProfile(model, backendId) {
   return normalizeProfile({
     id: model.id.replace(/[^a-z0-9._-]+/gi, "-").toLowerCase(),
@@ -510,6 +537,13 @@ async function runProfile(profile, options = {}) {
         if (state?.pid) {
           try { await stopProfile(profile); } catch { /* best effort */ }
         }
+        if (!options.textOnlyRetry && isUnsupportedMmprojError(err, profile)) {
+          console.log(pc.yellow("Vision projector is not supported by this llama.cpp build. Retrying text-only."));
+          console.log(pc.dim("Update llama.cpp later to re-enable vision for this model."));
+          const textOnly = textOnlyProfile(profile);
+          await saveProfile(textOnly);
+          return await runProfile(textOnly, { ...options, textOnlyRetry: true });
+        }
         throw err;
       }
     }

package/src/profile-setup.mjs

@@ -50,10 +50,22 @@ export async function configureLocalProfile(prompt, profile) {
     console.log("");
     console.log(renderSection("QAT detected", renderRows([
       ["Meaning", "quantization-aware trained"],
-      ["Runtime flags", "none required"],
+      ["Runtime flags", "none QAT-specific"],
     ])));
   }
 
+  if (caps.vision && profile.mmprojPath) {
+    console.log("");
+    const gemma4Unified = isGemma4UnifiedProjector(caps.mmprojProjectorType);
+    console.log(renderSection("Vision projector detected", renderRows([
+      ["Projector", caps.mmprojProjectorType ?? "unknown"],
+      ["Flag", `--mmproj ${profile.mmprojPath}`],
+      ...(gemma4Unified ? [["Note", pc.yellow("Gemma 4 unified projectors need newer llama.cpp than current Homebrew stable.")]] : []),
+    ])));
+    const useVision = await prompt.yesNo("Enable vision with --mmproj?", !gemma4Unified);
+    configured = useVision ? applyVisionDefaults(configured) : removeVisionDefaults(configured, gemma4Unified ? "gemma4-unified-unsupported" : "user-disabled");
+  }
+
   if (caps.thinking) {
     console.log("");
     console.log(renderSection("Thinking model detected", renderRows([
@@ -106,6 +118,23 @@ function removeMtpDefaults(profile) {
   });
 }
 
+function applyVisionDefaults(profile) {
+  if (!profile.mmprojPath) return profile;
+  return applyProfileFlags({
+    ...profile,
+    capabilities: { ...(profile.capabilities ?? {}), vision: true, visionDisabledReason: undefined },
+  }, profile.flags, { values: { "--mmproj": profile.mmprojPath } });
+}
+
+function removeVisionDefaults(profile, reason) {
+  return applyProfileFlags({
+    ...profile,
+    disabledMmprojPath: profile.mmprojPath,
+    mmprojPath: null,
+    capabilities: { ...(profile.capabilities ?? {}), vision: false, visionDisabledReason: reason },
+  }, profile.flags, { remove: ["--mmproj"] });
+}
+
 function applyThinkingDefaults(profile) {
   const flags = { ...profile.flags, ...THINKING_DEFAULTS };
   return applyProfileFlags(profile, flags);
@@ -179,6 +208,10 @@ function renderMemoryEstimate(profile) {
   }
 }
 
+function isGemma4UnifiedProjector(projectorType) {
+  return /gemma4u[va]/i.test(String(projectorType ?? ""));
+}
+
 function detectionSummary(caps) {
   const parts = [];
   if (caps.architecture) parts.push(caps.architecture);

package/src/profiles.mjs

@@ -168,6 +168,7 @@ function summarizeCapabilities(caps) {
     imatrix: caps.imatrix,
     quant: caps.quant,
     metaCtx: caps.metaCtx,
+    mmprojProjectorType: caps.mmprojProjectorType,
     ctxSize: caps.ctxSize,
   };
 }

package/src/scan.mjs

@@ -74,5 +74,5 @@ function aliasFromName(name) {
 }
 
 function quantFromName(name) {
-  return name.match(/(Q\d_K_[A-Z]+|UD-[A-Z0-9_]+)/)?.[1];
+  return name.match(/(Q\d_K_[A-Z]+|Q\d_[01]|UD-[A-Z0-9_]+)/)?.[1];
 }