billion-context 0.1.75-pr.489.309 → 0.1.75-pr.490.310

package/dist/index.js

@@ -47856,7 +47856,7 @@ function parseCompressSettings(v2) {
     }
     out[key] = typeof obj[key] === "string" ? obj[key].trim() : obj[key];
   }
-  for (const key of ["nudgeGrowthTokens", "preserveRecentMessages", "preserveRecentTokens", "minCompressRange"]) {
+  for (const key of ["nudgeGrowthTokens", "preserveRecentMessages", "preserveRecentTokens", "minCompressRange", "minCompressRangeChars"]) {
     takeNumber(key);
   }
   if ("tiers" in obj) {
@@ -47922,6 +47922,7 @@ function resolveContextLimitValue(raw, nativeLimit) {
 function mergeCompress(global2, provider, model) {
   const pick = (k2) => model?.[k2] ?? provider?.[k2] ?? global2?.[k2];
   const promptLevels = [global2?.prompts, provider?.prompts, model?.prompts].filter(Boolean);
+  const rangeOf = (s3) => s3?.minCompressRangeChars ?? s3?.minCompressRange;
   return {
     modelContextLimit: pick("modelContextLimit"),
     maxContextLimit: pick("maxContextLimit"),
@@ -47929,7 +47930,7 @@ function mergeCompress(global2, provider, model) {
     nudgeGrowthTokens: pick("nudgeGrowthTokens"),
     preserveRecentMessages: pick("preserveRecentMessages"),
     preserveRecentTokens: pick("preserveRecentTokens"),
-    minCompressRange: pick("minCompressRange"),
+    minCompressRangeChars: rangeOf(model) ?? rangeOf(provider) ?? rangeOf(global2),
     tiers: pick("tiers"),
     prompts: promptLevels.length > 0 ? Object.assign({}, ...promptLevels) : void 0,
     acknowledgePromptsRisk: pick("acknowledgePromptsRisk")
@@ -47983,7 +47984,7 @@ function applyCompressSettings(base, limit, s3) {
     preserveRecentTokens: s3.preserveRecentTokens ?? base.preserveRecentTokens,
     compress: {
       ...base.compress,
-      minCompressRange: s3.minCompressRange ?? base.compress.minCompressRange
+      minCompressRange: s3.minCompressRangeChars ?? s3.minCompressRange ?? base.compress.minCompressRange
     }
   };
 }
@@ -48165,6 +48166,7 @@ var CACHE_FILE = path3.join(cacheDir(), "models-dev.json");
 var TTL_MS = 24 * 60 * 60 * 1e3;
 var cache2 = null;
 var loading = null;
+var warnedConflicts = /* @__PURE__ */ new Set();
 var snapshotReg = null;
 var snapshotMs = 0;
 function bundledSnapshotRegistry() {
@@ -48337,15 +48339,22 @@ function registryLookup(reg, model, host) {
   }
   if (provider === void 0) {
     const suffix = `/${model}`;
-    let agreed;
+    let max;
+    const distinct = /* @__PURE__ */ new Set();
+    const parts = [];
     for (const key of Object.keys(reg)) {
       if (!key.endsWith(suffix)) continue;
       const ctx = reg[key].limit?.context;
       if (typeof ctx !== "number" || ctx <= 0) continue;
-      if (agreed === void 0) agreed = ctx;
-      else if (agreed !== ctx) return void 0;
+      if (max === void 0 || ctx > max) max = ctx;
+      distinct.add(ctx);
+      parts.push(`${key}=${ctx}`);
+    }
+    if (max !== void 0 && distinct.size > 1 && !warnedConflicts.has(model)) {
+      warnedConflicts.add(model);
+      log("warn", `[acp-registry] conflicting context windows for "${model}" (${parts.join(", ")}) \u2014 using max ${max}`);
     }
-    return agreed;
+    return max;
   }
   return void 0;
 }
@@ -50705,7 +50714,7 @@ function summaryPayload(protocol, model, system, content) {
   if (protocol === "openai") {
     return { model, max_tokens: MAX_SUMMARY_OUTPUT_TOKENS, messages: [{ role: "system", content: system }, { role: "user", content }], stream: false };
   }
-  return { model, max_output_tokens: MAX_SUMMARY_OUTPUT_TOKENS, instructions: system, input: [{ role: "user", content }], stream: false, store: false };
+  return { model, max_output_tokens: MAX_SUMMARY_OUTPUT_TOKENS, instructions: system, input: [{ role: "user", content }], stream: false };
 }
 function extractSummaryText(protocol, json) {
   if (protocol === "anthropic") {
@@ -50771,7 +50780,7 @@ TASK: The conversation segment below (messages ${startRef}\u2013${endRef}) must
 var ABORTED_FAILURE = { kind: "aborted", detail: "the client disconnected during preflight compression" };
 async function preflightCompress(deps, messages) {
   const limit = deps.config.modelContextLimit;
-  const result = { compressedRanges: 0, savedTokens: 0, payloadEstimate: estimateCoreMessages(messages) + (deps.imageFloor ?? 0) };
+  const result = { compressedRanges: 0, savedTokens: 0, payloadEstimate: estimateCoreMessages(messages) };
   if (limit <= 0) return result;
   const budget = Math.max(MIN_CHUNK_TOKENS, Math.floor(limit * CHUNK_FRACTION));
   const minChars = deps.config.compress.minCompressRange;
@@ -50791,8 +50800,8 @@ async function preflightCompress(deps, messages) {
       renderTags: "text-only"
     });
     deps.session.state = turn.state;
-    currentTokens = Math.max(deps.session.stats.lastInputTokens, estimateCoreMessages(turn.messages) + (deps.imageFloor ?? 0));
-    result.payloadEstimate = estimateCoreMessages(turn.messages) + (deps.imageFloor ?? 0);
+    currentTokens = Math.max(deps.session.stats.lastInputTokens, estimateCoreMessages(turn.messages));
+    result.payloadEstimate = estimateCoreMessages(turn.messages);
     if (startTokens < 0) startTokens = currentTokens;
     if (currentTokens < limit) break;
     const ranges = viableRanges(turn.nudge?.compressibleRanges ?? []);
@@ -50878,78 +50887,6 @@ async function preflightCompress(deps, messages) {
   return result;
 }
 
-// src/image-tokens.ts
-var REMOTE_IMAGE_TOKENS = 4096;
-function imageTokenCap() {
-  const v2 = Number(process.env.BILI_IMAGE_TOKEN_CAP ?? "");
-  return Number.isInteger(v2) && v2 > 0 ? v2 : 0;
-}
-function applyCap(cost) {
-  const cap = imageTokenCap();
-  return cap > 0 ? Math.min(cost, cap) : cost;
-}
-function costForUrl(url) {
-  if (url.startsWith("data:")) {
-    const i = url.indexOf("base64,");
-    if (i >= 0) return applyCap(Math.ceil((url.length - i - "base64,".length) / 4));
-  }
-  return applyCap(REMOTE_IMAGE_TOKENS);
-}
-function isObj(v2) {
-  return typeof v2 === "object" && v2 !== null;
-}
-function urlOf(v2) {
-  if (typeof v2 === "string") return v2;
-  if (isObj(v2) && typeof v2.url === "string") return v2.url;
-  return void 0;
-}
-function imageTokensInParsedBody(protocol, body) {
-  if (!isObj(body)) return 0;
-  let total = 0;
-  if (protocol === "responses") {
-    const input = body.input;
-    if (!Array.isArray(input)) return 0;
-    for (const item of input) {
-      if (!isObj(item) || !Array.isArray(item.content)) continue;
-      for (const part of item.content) {
-        if (!isObj(part) || part.type !== "input_image") continue;
-        const url = urlOf(part.image_url);
-        if (url) total += costForUrl(url);
-      }
-    }
-    return total;
-  }
-  const messages = body.messages;
-  if (!Array.isArray(messages)) return 0;
-  for (const m2 of messages) {
-    if (!isObj(m2) || !Array.isArray(m2.content)) continue;
-    for (const part of m2.content) {
-      if (!isObj(part)) continue;
-      if (protocol === "openai") {
-        if (part.type !== "image_url") continue;
-        const url = urlOf(part.image_url);
-        if (url) total += costForUrl(url);
-      } else {
-        if (part.type !== "image") continue;
-        const src = part.source;
-        if (isObj(src) && src.type === "base64" && typeof src.data === "string") total += applyCap(Math.ceil(src.data.length / 4));
-        else if (isObj(src) && src.type === "url" && typeof src.url === "string") total += costForUrl(src.url);
-      }
-    }
-  }
-  return total;
-}
-function imageTokensInRawBody(protocol, raw) {
-  const s3 = typeof raw === "string" ? raw : raw.toString("utf8");
-  const probe = protocol === "responses" ? s3.includes("input_image") : protocol === "openai" ? s3.includes("image_url") : s3.includes('"type":"image"') || s3.includes('"type": "image"');
-  if (!probe) return 0;
-  try {
-    return imageTokensInParsedBody(protocol, JSON.parse(s3));
-  } catch {
-    return 0;
-  }
-}
-
 // src/web/index.ts
 import { readFileSync as readFileSync3 } from "fs";
 import { dirname as dirname4, join as join3 } from "path";
@@ -51158,7 +51095,7 @@ function renderPage(origin, version2) {
 <div class="card"><div class="card-head"><h2>ZCode\uFF08\u7F16\u7A0B\u5957\u9910\uFF09</h2><span class="badge">MITM</span><button class="btn small copy-btn" data-copy="${caPathEsc}">\u590D\u5236 CA \u8DEF\u5F84</button></div><dl class="kv"><dt>\u65B9\u5F0F</dt><dd>MITM \u900F\u660E\u4EE3\u7406</dd><dt>\u8BBE\u7F6E</dt><dd>Settings \u2192 Network \u2192 HTTP Proxy = <span class="mono">${origin}</span></dd><dt>CA \u8BC1\u4E66</dt><dd class="mono">${caPathEsc}${caReady ? "" : "\uFF08\u5C1A\u672A\u751F\u6210\uFF0C\u542F\u52A8\u4EE3\u7406\u540E\u81EA\u52A8\u521B\u5EFA\uFF09"}</dd></dl><p class="hint">CA \u8DEF\u5F84\u662F\u672C\u673A bili \u5B9E\u9645\u751F\u6210\u7684\u7EDD\u5BF9\u8DEF\u5F84\uFF0C\u70B9"\u590D\u5236 CA \u8DEF\u5F84"\u540E\u76F4\u63A5\u7C98\u8D34\u8FDB ZCode\u3002\u26A0\uFE0F Windows \u4E0A ZCode \u4E0D\u4F1A\u5C55\u5F00 <span class="mono">~</span>\uFF0C\u8BF7\u52FF\u624B\u586B <span class="mono">~/\u2026</span> \u5F62\u5F0F\u7684\u8DEF\u5F84\u3002</p></div></section>
 <section id="page-upstream" class="page"><h1>\u4E0A\u6E38\u7F51\u7EDC</h1><p class="lead">\u63A7\u5236 bili \u5982\u4F55\u8BBF\u95EE\u771F\u5B9E Provider\uFF0C\u4E0D\u4F1A\u6539\u53D8\u5BA2\u6237\u7AEF\u7684\u672C\u5730\u8DEF\u7531\u5730\u5740\u3002</p><div class="card"><div class="card-head"><h2>\u5168\u5C40\u4EE3\u7406</h2></div><div class="modes"><label><input type="radio" name="proxy-mode" value="direct" checked>\u76F4\u8FDE\uFF08\u9ED8\u8BA4\uFF09</label><label><input type="radio" name="proxy-mode" value="manual">\u624B\u52A8\u4EE3\u7406</label><label><input type="radio" name="proxy-mode" value="auto">\u81EA\u52A8\uFF08\u8DDF\u968F\u7CFB\u7EDF\uFF09</label></div><div class="field"><label>HTTP / HTTPS Proxy</label><input id="proxy-url" class="mono" placeholder="http://127.0.0.1:7897"></div><dl class="kv"><dt>\u5F53\u524D\u6765\u6E90</dt><dd id="upstream-source">\u76F4\u8FDE</dd><dt>\u6709\u6548\u4EE3\u7406</dt><dd id="upstream-effective" class="mono">direct</dd><dt>\u7CFB\u7EDF PAC</dt><dd id="upstream-pac" class="mono">\u2014</dd><dt>\u72B6\u6001</dt><dd id="upstream-state" class="status">\u5C1A\u672A\u6D4B\u8BD5</dd></dl><div class="actions"><button id="save-upstream" class="btn primary">\u4FDD\u5B58\u5E76\u70ED\u66F4\u65B0</button><button id="test-upstream" class="btn">\u6D4B\u8BD5\u8FDE\u63A5</button></div></div><div class="card"><div class="card-head"><h2>\u6309 URL \u8986\u76D6</h2></div><p class="status">\u4E3A\u7279\u5B9A Provider \u5355\u72EC\u8BBE\u7F6E\u4E0A\u6E38\u4EE3\u7406\uFF0C\u8986\u76D6\u5168\u5C40\u8BBE\u7F6E\u3002\u7559\u7A7A = \u7EE7\u627F\u5168\u5C40\u3002</p><div id="upstream-overrides" class="status">\u52A0\u8F7D\u4E2D</div><div class="actions"><button id="save-overrides" class="btn primary">\u4FDD\u5B58\u8986\u76D6\u5E76\u70ED\u66F4\u65B0</button></div></div></section>
 <section id="page-sessions" class="page"><div class="card-head"><div><h1>\u4F1A\u8BDD</h1><p class="lead">ACP \u538B\u7F29\u72B6\u6001\u4E0E\u4E0A\u6E38\u7528\u91CF\u3002</p></div><button id="refresh-sessions" class="btn">\u5237\u65B0</button></div><div class="card"><table><thead><tr><th>\u6807\u9898</th><th>\u534F\u8BAE</th><th>\u6807\u8BC6</th><th>\u8BF7\u6C42</th><th>\u4E0A\u4E0B\u6587</th><th>\u6700\u540E\u6D3B\u52A8</th></tr></thead><tbody id="sessions-body"></tbody></table></div></section>
-<section id="page-settings" class="page"><h1>\u9AD8\u7EA7\u8BBE\u7F6E</h1><p class="lead">\u76F4\u63A5\u7F16\u8F91 providers JSON\uFF08\u6DFB\u52A0\u65B0 Provider\u3001\u914D\u7F6E\u6A21\u578B\u4E0A\u4E0B\u6587\u7A97\u53E3\u3001\u6309 URL \u4EE3\u7406\u7B49\uFF09\uFF1B\u4FDD\u5B58\u540E\u7ACB\u5373\u70ED\u66F4\u65B0\u3002</p><div class="card"><div class="field"><label>providers JSON</label><textarea id="providers-json">{}</textarea></div><div class="actions"><button id="save-providers" class="btn primary">\u4FDD\u5B58\u5E76\u5E94\u7528</button></div></div><div class="card"><div class="card-head"><h2>\u538B\u7F29\u53C2\u6570\uFF08\u5168\u5C40\uFF09</h2></div><p class="status">\u5168\u5C40 compress \u914D\u7F6E\uFF08modelContextLimit / maxContextLimit / emergencyThresholdPercent / nudgeGrowthTokens / preserveRecentMessages / preserveRecentTokens / minCompressRange / tiers / injectTool / injectNudge\uFF09\u3002\u4FDD\u5B58\u540E\u70ED\u66F4\u65B0\uFF0C\u65E0\u9700\u91CD\u542F\u3002\u7F6E\u7A7A = \u6062\u590D\u9ED8\u8BA4\u3002</p><div class="field"><label>compress JSON</label><textarea id="compress-json" placeholder='{"modelContextLimit": 200000}'></textarea></div><div class="actions"><button id="save-compress" class="btn primary">\u4FDD\u5B58\u5E76\u70ED\u66F4\u65B0</button></div></div></section></main></div><div id="toast" class="toast"></div><script>${WEB_CLIENT}</script></body></html>`;
+<section id="page-settings" class="page"><h1>\u9AD8\u7EA7\u8BBE\u7F6E</h1><p class="lead">\u76F4\u63A5\u7F16\u8F91 providers JSON\uFF08\u6DFB\u52A0\u65B0 Provider\u3001\u914D\u7F6E\u6A21\u578B\u4E0A\u4E0B\u6587\u7A97\u53E3\u3001\u6309 URL \u4EE3\u7406\u7B49\uFF09\uFF1B\u4FDD\u5B58\u540E\u7ACB\u5373\u70ED\u66F4\u65B0\u3002</p><div class="card"><div class="field"><label>providers JSON</label><textarea id="providers-json">{}</textarea></div><div class="actions"><button id="save-providers" class="btn primary">\u4FDD\u5B58\u5E76\u5E94\u7528</button></div></div><div class="card"><div class="card-head"><h2>\u538B\u7F29\u53C2\u6570\uFF08\u5168\u5C40\uFF09</h2></div><p class="status">\u5168\u5C40 compress \u914D\u7F6E\uFF08modelContextLimit / maxContextLimit / emergencyThresholdPercent / nudgeGrowthTokens / preserveRecentMessages / preserveRecentTokens / minCompressRangeChars\uFF08\u5355\u4F4D\uFF1A\u5B57\u7B26\uFF0C\u65E7\u540D minCompressRange \u517C\u5BB9\uFF09 / tiers / injectTool / injectNudge\uFF09\u3002\u4FDD\u5B58\u540E\u70ED\u66F4\u65B0\uFF0C\u65E0\u9700\u91CD\u542F\u3002\u7F6E\u7A7A = \u6062\u590D\u9ED8\u8BA4\u3002</p><div class="field"><label>compress JSON</label><textarea id="compress-json" placeholder='{"modelContextLimit": 200000}'></textarea></div><div class="actions"><button id="save-compress" class="btn primary">\u4FDD\u5B58\u5E76\u70ED\u66F4\u65B0</button></div></div></section></main></div><div id="toast" class="toast"></div><script>${WEB_CLIENT}</script></body></html>`;
 }
 
 // src/web/api.ts
@@ -57307,7 +57244,7 @@ function emergencyNudge(nudge, escalationPct = EMERGENCY_NUDGE_ESCALATION_PCT) {
 function clampOutgoingOutput(rebuilt, field, ctx, sessionId, log2) {
   const raw = rebuilt[field];
   if (typeof raw !== "number") return;
-  const inputEstimate = estimateInputTokens(ctx.processedMessages, ctx.systemText, ctx.tools, ctx.lastInputTokens) + ctx.imageTokens;
+  const inputEstimate = estimateInputTokens(ctx.processedMessages, ctx.systemText, ctx.tools, ctx.lastInputTokens);
   const capped = clampOutputBudget(raw, inputEstimate, ctx.nativeWindow);
   if (capped !== void 0) {
     rebuilt[field] = capped;
@@ -57372,7 +57309,7 @@ function prepareOpenai(parsed, req, opts, core, config, prompts, log2, session,
     processedMessages = [];
   }
   const rebuilt = { ...parsed, messages: rebuiltMessages, tools: toolsOut };
-  clampOutgoingOutput(rebuilt, typeof parsed.max_completion_tokens === "number" ? "max_completion_tokens" : "max_tokens", { systemText: openaiSystemText, tools: toolsOut, processedMessages, lastInputTokens: session.stats.lastInputTokens, nativeWindow, imageTokens: imageTokensInParsedBody("openai", rebuilt) }, sessionId, log2);
+  clampOutgoingOutput(rebuilt, typeof parsed.max_completion_tokens === "number" ? "max_completion_tokens" : "max_tokens", { systemText: openaiSystemText, tools: toolsOut, processedMessages, lastInputTokens: session.stats.lastInputTokens, nativeWindow }, sessionId, log2);
   delete rebuilt.prompt_cache_retention;
   if (stream2 && rebuilt.stream_options === void 0) {
     rebuilt.stream_options = { include_usage: true };
@@ -57490,7 +57427,7 @@ function prepareResponses(parsed, req, opts, core, config, prompts, log2, sessio
   }
   const rebuilt = { ...parsed, input: rebuiltInput, tools: toolsOut };
   if (!isCompactionTrigger) {
-    clampOutgoingOutput(rebuilt, "max_output_tokens", { systemText: (responsesProjection?.systemParts ?? []).join("\n"), tools: toolsOut, processedMessages, lastInputTokens: session.stats.lastInputTokens, nativeWindow, imageTokens: imageTokensInParsedBody("responses", rebuilt) }, sessionId, log2);
+    clampOutgoingOutput(rebuilt, "max_output_tokens", { systemText: (responsesProjection?.systemParts ?? []).join("\n"), tools: toolsOut, processedMessages, lastInputTokens: session.stats.lastInputTokens, nativeWindow }, sessionId, log2);
   }
   const promptCacheKey = resolvePromptCacheKey(
     rebuilt.prompt_cache_key,
@@ -57715,13 +57652,11 @@ function isPreflightFailFast(outcome) {
 async function preflightCompressIfNeeded(prepared, runPrepare, req, res, opts, core, config, model, route, affinity, log2, instanceId) {
   const session = prepared.session;
   const limit = config.modelContextLimit;
-  const imageTokens = imageTokensInRawBody(prepared.protocol, prepared.body);
-  const payloadEstimate = estimateCoreMessages(prepared.processedMessages) + imageTokens;
+  const payloadEstimate = estimateCoreMessages(prepared.processedMessages);
   const tokenCount = Math.max(session.stats.lastInputTokens, payloadEstimate);
   if (limit <= 0 || !model || tokenCount < limit) return prepared;
   const failFast = (status2, detail, retryable) => {
-    const imageNote = imageTokens >= limit ? ` Images alone account for ~${imageTokens} tokens (\u2265 window ${limit}); compression cannot remove them \u2014 shrink or remove the images, or raise the window.` : "";
-    const message = `context ~${tokenCount} tokens exceeds the model window ${limit} (model=${model}) and preflight compression could not bring it under: ${detail}.` + imageNote + ` The over-window payload was NOT forwarded.`;
+    const message = `context ~${tokenCount} tokens exceeds the model window ${limit} (model=${model}) and preflight compression could not bring it under: ${detail}. The over-window payload was NOT forwarded.`;
     log2("error", `[${session.id}] preflight fail-fast ${status2} (retryable=${retryable}): ${message}`);
     return { failFast: true, status: status2, message, retryable, respond: !res.writableEnded };
   };
@@ -57751,8 +57686,7 @@ async function preflightCompressIfNeeded(prepared, runPrepare, req, res, opts, c
       model,
       proxyUrl,
       signal: clientAbort.signal,
-      log: log2,
-      imageFloor: imageTokens
+      log: log2
     },
     prepared.originalMessages
   );
@@ -57921,12 +57855,9 @@ ${hdrText}
       const info = inspectContextOverflow(upstream.status, errBody.toString("utf8"));
       if (info.isOverflow) {
         let reqModel;
-        let rejectedImageTokens = 0;
         try {
           const rawBody = typeof prepared.body === "string" ? prepared.body : prepared.body.toString("utf8");
-          const parsedBody = JSON.parse(rawBody);
-          reqModel = typeof parsedBody.model === "string" ? parsedBody.model : void 0;
-          rejectedImageTokens = imageTokensInParsedBody(prepared.protocol, parsedBody);
+          reqModel = JSON.parse(rawBody).model;
         } catch {
           reqModel = void 0;
         }
@@ -57938,7 +57869,7 @@ ${hdrText}
           s3.metadata.learnedContextLimits = learnedMap;
           log2("warn", `[${s3.id}] upstream context overflow \u2014 learned real window ${info.window} for ${reqModel ?? "(unknown model)"} (was ${prev ?? "unset"}); arming emergency shrink`);
         } else {
-          const payloadEstimate = estimateCoreMessages(prepared.processedMessages) + rejectedImageTokens;
+          const payloadEstimate = estimateCoreMessages(prepared.processedMessages);
           const prev = (reqModel ? learnedMap[reqModel] : void 0) ?? s3.metadata.learnedContextLimit;
           if (payloadEstimate >= 1e3 && (prev === void 0 || payloadEstimate < prev)) {
             if (reqModel) learnedMap[reqModel] = payloadEstimate;