billion-context-omp 0.3.3-pr.146.13 → 0.3.3-pr.147.16

package/dist/index.js

@@ -2679,14 +2679,14 @@ function renderUncompressedRanges(visible) {
   };
   const merged = [];
   for (const m of visible) {
-    const num = refNum2(m.ref);
+    const num2 = refNum2(m.ref);
     const last = merged[merged.length - 1];
-    if (last && num === last.startNum + last.count) {
+    if (last && num2 === last.startNum + last.count) {
       last.endRef = m.ref;
       last.count += 1;
       last.tokens += m.tokens;
     } else {
-      merged.push({ startRef: m.ref, endRef: m.ref, startNum: num, count: 1, tokens: m.tokens, tool: m.tool });
+      merged.push({ startRef: m.ref, endRef: m.ref, startNum: num2, count: 1, tokens: m.tokens, tool: m.tool });
     }
   }
   for (const r of merged.slice(0, 30)) {
@@ -5768,6 +5768,37 @@ function formatCompactTokens(count) {
   if (count < 1e7) return `${(count / 1e6).toFixed(1)}M`;
   return `${Math.round(count / 1e6)}M`;
 }
+function num(v) {
+  return typeof v === "number" && Number.isFinite(v) && v > 0 ? v : 0;
+}
+function cacheHitStats(usages) {
+  let cacheRead = 0;
+  let billedPrompt = 0;
+  let requests = 0;
+  let last;
+  for (const u of usages) {
+    const read = num(u?.cacheRead);
+    const write = num(u?.cacheWrite);
+    const input = num(u?.input);
+    if (read + write <= 0) continue;
+    const total = read + write + input;
+    if (total <= 0) continue;
+    cacheRead += read;
+    billedPrompt += total;
+    requests += 1;
+    last = read / total;
+  }
+  return {
+    session: requests > 0 ? cacheRead / billedPrompt : void 0,
+    last,
+    requests,
+    cacheRead,
+    billedPrompt
+  };
+}
+function formatHitRate(rate) {
+  return `${(Math.max(0, Math.min(1, rate)) * 100).toFixed(1)}%`;
+}
 function bar(value, total, width = 20) {
   if (total === 0) return "";
   const filled = Math.max(0, Math.min(width, Math.round(value / total * width)));
@@ -5820,6 +5851,15 @@ function buildStatusPanel(input) {
       lines.push(`  ${cat.label.padEnd(10)} ${b} ${String(pct2).padStart(3)}%  ${fmt2(cat.value)}`);
     }
   }
+  if (input.cacheUsages) {
+    const cache2 = cacheHitStats(input.cacheUsages);
+    if (cache2.requests > 0 && cache2.session !== void 0 && cache2.last !== void 0) {
+      lines.push("");
+      lines.push(
+        `Prompt cache (provider-reported): ${formatHitRate(cache2.last)} last \xB7 ${formatHitRate(cache2.session)} session avg \u2014 ${fmt2(cache2.cacheRead)} of ${fmt2(cache2.billedPrompt)} billed prompt tokens served from cache (${cache2.requests} req)`
+      );
+    }
+  }
   lines.push("");
   if (nudge) {
     if (nudge.shouldInject) {
@@ -5857,6 +5897,16 @@ function buildStatusPanel(input) {
 }
 
 // src/commands.ts
+function cacheUsageSamples(entries) {
+  const out = [];
+  for (const e of entries) {
+    if (e.type !== "message" || !e.message) continue;
+    const m = e.message;
+    if (m.role !== "assistant" || !m.usage) continue;
+    out.push({ input: m.usage.input ?? 0, cacheRead: m.usage.cacheRead ?? 0, cacheWrite: m.usage.cacheWrite ?? 0 });
+  }
+  return out;
+}
 function safeHandler(handler) {
   return async (args, ctx) => {
     try {
@@ -5951,7 +6001,8 @@ async function statusReport(runtime, ctx) {
     state: turn.state,
     nudge: turn.nudge,
     modelContextLimit: config.modelContextLimit,
-    unprunedTokens: coreMessages.reduce((sum, m) => sum + defaultCountTokens(m.text ?? ""), 0)
+    unprunedTokens: coreMessages.reduce((sum, m) => sum + defaultCountTokens(m.text ?? ""), 0),
+    cacheUsages: ctx.sessionManager?.getEntries ? cacheUsageSamples(ctx.sessionManager.getEntries()) : void 0
   });
 }