billion-context 0.1.58 → 0.1.59-pr.298.151

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/index.js CHANGED
@@ -50953,11 +50953,13 @@ function usageTotals(protocol, usage) {
50953
50953
  }
50954
50954
  var CONTEXT_OVERFLOW_PATTERNS = [
50955
50955
  /context_length_exceeded/i,
50956
+ /context_window_exceeded/i,
50956
50957
  /context length exceeded/i,
50957
50958
  /maximum context length/i,
50958
50959
  /max context length/i,
50959
50960
  /maximum context size/i,
50960
50961
  /exceeds the context window/i,
50962
+ /out of room in the model/i,
50961
50963
  /exceeded model token limit/i,
50962
50964
  /prompt is too long/i,
50963
50965
  /prompt_too_long/i,
@@ -55865,7 +55867,16 @@ ${hdrText}
55865
55867
  s3.metadata.learnedContextLimits = learnedMap;
55866
55868
  log2("warn", `[${s3.id}] upstream context overflow \u2014 learned real window ${info.window} for ${reqModel ?? "(unknown model)"} (was ${prev ?? "unset"}); arming emergency shrink`);
55867
55869
  } else {
55868
- log2("warn", `[${s3.id}] upstream context overflow (window not parseable): ${info.message}`);
55870
+ const payloadEstimate = estimateCoreMessages(prepared.processedMessages);
55871
+ const prev = (reqModel ? learnedMap[reqModel] : void 0) ?? s3.metadata.learnedContextLimit;
55872
+ if (payloadEstimate >= 1e3 && (prev === void 0 || payloadEstimate < prev)) {
55873
+ if (reqModel) learnedMap[reqModel] = payloadEstimate;
55874
+ else s3.metadata.learnedContextLimit = payloadEstimate;
55875
+ s3.metadata.learnedContextLimits = learnedMap;
55876
+ log2("warn", `[${s3.id}] upstream context overflow (window not parseable) \u2014 learned conservative window ${payloadEstimate} for ${reqModel ?? "(unknown model)"} from rejected payload size (was ${prev ?? "unset"}); arming emergency shrink`);
55877
+ } else {
55878
+ log2("warn", `[${s3.id}] upstream context overflow (window not parseable): ${info.message}`);
55879
+ }
55869
55880
  }
55870
55881
  const floor = info.window ?? (reqModel ? learnedMap[reqModel] : void 0) ?? s3.metadata.learnedContextLimit ?? s3.metadata.effectiveContextLimit ?? 0;
55871
55882
  if (floor > 0) s3.stats.lastInputTokens = Math.max(s3.stats.lastInputTokens, floor);