turbollm 1.13.2 → 1.13.4

package/README.md

@@ -443,8 +443,9 @@ connected); **Remove** undoes it.
 
 - A **persistent bottom strip** on every screen shows real-time CPU, RAM, GPU and VRAM — values
   and percentages — pulled from `nvidia-smi`, WDDM counters, `rocm-smi`, AMD sysfs, `ioreg`,
-  or `/proc` on Linux. Clicking the bar opens detailed gauges and sparklines in Settings → System.
-  Toggle it off entirely from Settings → General.
+  or `/proc` on Linux. Clicking the bar opens detailed gauges and sparklines in Settings → System,
+  including live **disk I/O** (read/write MB/s) and a **shared/spilled VRAM** figure that shows
+  when a discrete GPU has spilled into system memory. Toggle it off entirely from Settings → General.
 - **Engines → Monitor** puts the live engine log and those same hardware gauges on one
   always-visible split screen, plus a second, LM Studio-style **developer request log**: every
   completion the daemon serves — external API clients, Code sessions, in-app Chat — with

package/dist/{chunk-HQ76BEUS.js → chunk-FJDRILNH.js}

@@ -1162,11 +1162,11 @@ async function buildOpencodeConfigContent(base, apiKey, modelKey, modelName, mcp
 function piArgs(ctx) {
   return ["--provider", "turbollm", "--model", ctx.pinnedModel, "--api-key", ctx.authToken];
 }
-var realRunCommand = (bin, args) => new Promise((resolve3) => {
+var realRunCommand = (bin, args) => new Promise((resolve4) => {
   const resolved = resolveExecutable(bin);
   const child = requiresShell(resolved) ? spawn2(buildShellCommand(bin, args), { shell: true, stdio: "ignore" }) : spawn2(resolved ?? bin, args, { stdio: "ignore" });
-  child.on("error", () => resolve3(false));
-  child.on("exit", (code) => resolve3(code === 0));
+  child.on("error", () => resolve4(false));
+  child.on("exit", (code) => resolve4(code === 0));
 });
 async function prepareHermes(base, _apiKey, modelKey, _modelName, run = realRunCommand) {
   const sets = [
@@ -1285,14 +1285,14 @@ var WORKTREES_SUBDIR = join9(".turbollm", "worktrees");
 var EXCLUDE_LINE = "/.turbollm/";
 var MAX_BRANCH_SUFFIX = 20;
 function runGit(cwd, args, timeoutMs = 3e4) {
-  return new Promise((resolve3) => {
+  return new Promise((resolve4) => {
     let stdout = "";
     let stderr = "";
     let settled = false;
     const done = (r) => {
       if (!settled) {
         settled = true;
-        resolve3(r);
+        resolve4(r);
       }
     };
     try {
@@ -1519,6 +1519,14 @@ import { readFileSync as readFileSync6, writeFileSync as writeFileSync5, mkdirSy
 import { join as join10 } from "path";
 var BENCH_RATE_LIMIT_MS = 60 * 60 * 1e3;
 
+// src/models/scanner.ts
+import { existsSync as existsSync7, lstatSync, readdirSync as readdirSync5, readFileSync as readFileSync7, realpathSync as realpathSync2, rmSync as rmSync4, writeFileSync as writeFileSync6 } from "fs";
+import { readdir, realpath, stat } from "fs/promises";
+import { basename as basename2, dirname as dirname4, isAbsolute, join as join11, relative, resolve as resolve2, sep as sep2 } from "path";
+
+// src/gguf/gguf.ts
+import { closeSync, openSync, readSync } from "fs";
+
 // src/models/profile.ts
 var ROCM_APU_NOMMAP_MIN_BYTES = 30 * 1024 ** 3;
 
@@ -1588,7 +1596,7 @@ async function autoTitleFromConversation(d, convId) {
 }
 
 // src/code/code-session.ts
-import { join as join11, resolve as resolve2 } from "path";
+import { join as join12, resolve as resolve3 } from "path";
 import { readFile as readFile3 } from "fs/promises";
 import { Type, Unsafe } from "typebox";
 import {
@@ -1610,7 +1618,7 @@ import { spawn as spawn4 } from "child_process";
 import { getShellConfig } from "@earendil-works/pi-coding-agent";
 var EXIT_STDIO_GRACE_MS = 100;
 function waitForChildProcess(child) {
-  return new Promise((resolve3, reject) => {
+  return new Promise((resolve4, reject) => {
     let settled = false;
     let exited = false;
     let exitCode = null;
@@ -1636,7 +1644,7 @@ function waitForChildProcess(child) {
       cleanup();
       child.stdout?.destroy();
       child.stderr?.destroy();
-      resolve3(code);
+      resolve4(code);
     };
     const maybeFinalizeAfterExit = () => {
       if (!exited || settled) return;
@@ -1688,11 +1696,11 @@ function isAlive(pid) {
   }
 }
 function sleep(ms) {
-  return new Promise((resolve3) => setTimeout(resolve3, ms));
+  return new Promise((resolve4) => setTimeout(resolve4, ms));
 }
 async function killDescendantsViaWmi(rootPid) {
   const psScript = `$all = Get-CimInstance Win32_Process | Select-Object ProcessId,ParentProcessId; $ids = @(${rootPid}); $frontier = @(${rootPid}); for ($i = 0; $i -lt 8; $i++) {   $children = $all | Where-Object { $frontier -contains $_.ParentProcessId } | Select-Object -ExpandProperty ProcessId;   if (-not $children) { break }; $ids += $children; $frontier = $children }; $ids | Select-Object -Unique | ForEach-Object { Write-Output $_ }`;
-  const pids = await new Promise((resolve3) => {
+  const pids = await new Promise((resolve4) => {
     const proc = spawn4("powershell", ["-NoProfile", "-NonInteractive", "-Command", psScript], {
       stdio: ["ignore", "pipe", "ignore"],
       windowsHide: true
@@ -1703,9 +1711,9 @@ async function killDescendantsViaWmi(rootPid) {
     });
     proc.on("close", () => {
       const found = out.split(/\r?\n/).map((l) => Number(l.trim())).filter((n) => Number.isInteger(n) && n > 0);
-      resolve3([...new Set(found)]);
+      resolve4([...new Set(found)]);
     });
-    proc.on("error", () => resolve3([]));
+    proc.on("error", () => resolve4([]));
   });
   for (const pid of pids) {
     try {
@@ -2038,10 +2046,10 @@ var LspClient = class {
       const npxCmd = process.platform === "win32" ? "npx.cmd" : "npx";
       const proc = spawn5(npxCmd, this.spec.npxArgs, npxSpawnOptions(this.rootDir));
       this.proc = proc;
-      const spawnFailure = new Promise((resolve3) => {
-        proc.once("error", (e) => resolve3(e instanceof Error ? e.message : String(e)));
+      const spawnFailure = new Promise((resolve4) => {
+        proc.once("error", (e) => resolve4(e instanceof Error ? e.message : String(e)));
         proc.once("exit", (code) => {
-          if (!this.ready) resolve3(`server exited early (code ${code})`);
+          if (!this.ready) resolve4(`server exited early (code ${code})`);
         });
       });
       const conn = createMessageConnection(new StreamMessageReader(proc.stdout), new StreamMessageWriter(proc.stdin));
@@ -2073,7 +2081,7 @@ var LspClient = class {
         // repo doesn't have typescript installed yet, e.g. a brand-new project.
         ...fallbackTsserverPath ? { initializationOptions: { tsserver: { path: fallbackTsserverPath } } } : {}
       });
-      const timeout = new Promise((resolve3) => setTimeout(() => resolve3("timeout"), INITIALIZE_TIMEOUT_MS));
+      const timeout = new Promise((resolve4) => setTimeout(() => resolve4("timeout"), INITIALIZE_TIMEOUT_MS));
       const result = await Promise.race([initialize.then(() => "ok"), spawnFailure.then((e) => `error:${e}`), timeout]);
       if (result === "timeout") throw new Error(`${this.spec.language} language server did not initialize within ${INITIALIZE_TIMEOUT_MS}ms`);
       if (result.startsWith("error:")) throw new Error(result.slice("error:".length));
@@ -2098,9 +2106,9 @@ var LspClient = class {
     const uri = toFileUri(path);
     const version = (this.docVersionByUri.get(uri) ?? 0) + 1;
     this.docVersionByUri.set(uri, version);
-    const waitForPush = new Promise((resolve3) => {
+    const waitForPush = new Promise((resolve4) => {
       const list = this.waitersByUri.get(uri) ?? [];
-      list.push(resolve3);
+      list.push(resolve4);
       this.waitersByUri.set(uri, list);
     });
     if (version === 1) {
@@ -2110,7 +2118,7 @@ var LspClient = class {
       this.conn.sendNotification("textDocument/didChange", { textDocument: { uri, version }, contentChanges: [{ text }] }).catch(() => {
       });
     }
-    await Promise.race([waitForPush, new Promise((resolve3) => setTimeout(resolve3, DIAGNOSTICS_WAIT_MS))]);
+    await Promise.race([waitForPush, new Promise((resolve4) => setTimeout(resolve4, DIAGNOSTICS_WAIT_MS))]);
     return this.diagnosticsByUri.get(uri) ?? [];
   }
   dispose() {
@@ -2410,7 +2418,7 @@ async function runCodeSession(params) {
     console.warn(`[code-session] history seed for session ${sessionId} produced ${sessionManager.getEntries().length}/${expectedEntries} expected entries \u2014 falling back to a plain-text history prefix on this turn's prompt`);
     task = `${HISTORY_FALLBACK_PREFIX}${renderHistoryFallbackText(summaryText, priorMessages)}${HISTORY_FALLBACK_SUFFIX}${rawTask}`;
   }
-  const agentDir = join11(d.store.dir(), "pi-agent");
+  const agentDir = join12(d.store.dir(), "pi-agent");
   modelRegistry.registerProvider("local", {
     ...upstream.provider,
     api: "openai-completions",
@@ -2537,7 +2545,7 @@ async function runCodeSession(params) {
         await sink({ event: "tool_call", data });
       });
     };
-    const skillAgentDir = join11(d.store.dir(), "pi-agent");
+    const skillAgentDir = join12(d.store.dir(), "pi-agent");
     const skillResourceLoader = new DefaultResourceLoader({
       cwd: repoRoot,
       agentDir: skillAgentDir,
@@ -2678,7 +2686,7 @@ async function runCodeSession(params) {
         await sink({ event: "tool_call", data });
       });
     };
-    const subAgentDir = join11(d.store.dir(), "pi-agent");
+    const subAgentDir = join12(d.store.dir(), "pi-agent");
     const subResourceLoader = new DefaultResourceLoader({
       cwd: repoRoot,
       agentDir: subAgentDir,
@@ -2893,7 +2901,7 @@ async function runCodeSession(params) {
         const spec = relPath ? lspSpecForPath(relPath) : null;
         if (spec) {
           try {
-            const absPath = resolve2(repoRoot, relPath);
+            const absPath = resolve3(repoRoot, relPath);
             const content = await readFile3(absPath, "utf8");
             const diagnostics = await getLspClient(spec).getDiagnostics(absPath, content);
             if (diagnostics.length > 0) {
@@ -3333,7 +3341,7 @@ async function compactCodeSession(params) {
       return payload;
     });
   };
-  const compactAgentDir = join11(d.store.dir(), "pi-agent");
+  const compactAgentDir = join12(d.store.dir(), "pi-agent");
   const compactResourceLoader = new DefaultResourceLoader({
     cwd: repoRoot,
     agentDir: compactAgentDir,
@@ -3432,7 +3440,7 @@ async function lookbackPreCompactionHistory(params) {
       return payload;
     });
   };
-  const lookbackAgentDir = join11(d.store.dir(), "pi-agent");
+  const lookbackAgentDir = join12(d.store.dir(), "pi-agent");
   const lookbackResourceLoader = new DefaultResourceLoader({
     cwd: repoRoot,
     agentDir: lookbackAgentDir,

package/dist/{chunk-HOKZAC3U.js → chunk-Y3TMLJQD.js}

@@ -65,7 +65,17 @@ function availableBackends(tag = LLAMA_BUILD) {
     if (a === "arm64") return [def("cpu", "CPU", `llama-${tag}-bin-win-cpu-arm64.zip`)];
     return [
       def("cuda", "CUDA (NVIDIA)", `llama-${tag}-bin-win-cuda-${CUDA_VER}-x64.zip`, `cudart-llama-bin-win-cuda-${CUDA_VER}-x64.zip`),
-      def("rocm", "ROCm / HIP (AMD Radeon)", `llama-${tag}-bin-win-hip-radeon-x64.zip`),
+      // Upstream has renamed this archive at least once (`*-bin-win-hip-radeon-x64.zip` →
+      // `*-bin-win-rocm-<toolkit-ver>-x64.zip`, GitHub #220) without any deprecation window,
+      // so the literal name below is only a fallback label — assetPatterns resolves the
+      // real filename from the release's own asset list at download time and survives the
+      // next rename too.
+      {
+        id: "rocm",
+        label: "ROCm / HIP (AMD Radeon)",
+        assets: [`llama-${tag}-bin-win-hip-radeon-x64.zip`],
+        assetPatterns: [/^llama-.+-bin-win-(?:hip-radeon|rocm-[\d.]+)-x64\.zip$/i]
+      },
       def("sycl", "SYCL (Intel)", `llama-${tag}-bin-win-sycl-x64.zip`),
       def("vulkan", "Vulkan (any GPU)", `llama-${tag}-bin-win-vulkan-x64.zip`),
       def("cpu", "CPU", `llama-${tag}-bin-win-cpu-x64.zip`)
@@ -236,12 +246,30 @@ async function provisionBackend(enginesRoot, backend, tag, onProgress, signal) {
   }
   mkdirSync(destDir, { recursive: true });
   const parts = backend.assets.length;
+  const tmpPaths = [];
   try {
     for (let i = 0; i < parts; i++) {
-      const asset = backend.assets[i];
       const part = i + 1;
-      const tmp = join(enginesRoot, asset);
-      const url = backend.downloadBaseUrl ? `${backend.downloadBaseUrl}/${asset}` : `https://github.com/${backend.repo ?? REPO}/releases/download/${backend.releaseTag ?? tag}/${asset}`;
+      const pattern = backend.assetPatterns?.[i];
+      let assetName = backend.assets[i];
+      let url;
+      if (backend.downloadBaseUrl) {
+        url = `${backend.downloadBaseUrl}/${assetName}`;
+      } else {
+        const repo = backend.repo ?? REPO;
+        const releaseTag = backend.releaseTag ?? tag;
+        if (pattern) {
+          const releaseAssets = await releaseAssetsForTag(repo, releaseTag, signal);
+          const found = releaseAssets.find((a) => pattern.test(a.name));
+          if (!found) throw new Error(`no release asset matching ${pattern} in ${repo}@${releaseTag}`);
+          assetName = found.name;
+          url = found.browser_download_url;
+        } else {
+          url = `https://github.com/${repo}/releases/download/${releaseTag}/${assetName}`;
+        }
+      }
+      const tmp = join(enginesRoot, assetName);
+      tmpPaths.push(tmp);
       onProgress?.({ phase: "downloading", pct: 0, part, parts });
       await downloadFile(url, tmp, (p) => onProgress?.({ ...p, part, parts }), signal);
       onProgress?.({ phase: "extracting", pct: -1, part, parts });
@@ -249,7 +277,7 @@ async function provisionBackend(enginesRoot, backend, tag, onProgress, signal) {
       rmSync(tmp, { force: true });
     }
   } catch (e) {
-    for (const asset of backend.assets) rmSync(join(enginesRoot, asset), { force: true });
+    for (const tmp of tmpPaths) rmSync(tmp, { force: true });
     rmSync(destDir, { recursive: true, force: true });
     throw e;
   }
@@ -347,6 +375,16 @@ async function latestBuildTagRelease(repo, signal) {
   }
   return best;
 }
+async function releaseAssetsForTag(repo, tag, signal) {
+  const res = await fetch(`https://api.github.com/repos/${repo}/releases/tags/${encodeURIComponent(tag)}`, {
+    headers: githubHeaders(),
+    signal
+  });
+  if (isRateLimited(res)) throw new GithubRateLimitError(repo);
+  if (!res.ok) throw new Error(`could not query ${repo} release ${tag}: HTTP ${res.status}`);
+  const data = await res.json();
+  return data.assets ?? [];
+}
 async function latestCommitSha(repo, branch = "", signal) {
   const ref = branch.trim() ? encodeURIComponent(branch.trim()) : "HEAD";
   const res = await fetch(`https://api.github.com/repos/${repo}/commits/${ref}`, {