@plurnk/plurnk-providers 1.3.12 → 1.4.0

package/src/usage.test.ts

@@ -23,7 +23,7 @@ test("normalizeUsage: OpenAI-style — reasoning split out of completion_tokens"
     assert.equal(u.completion + u.reasoning, 100); // billable output unchanged
 });
 
-test("normalizeUsage: xAI/Grok-style — reasoning is ADDITIVE, not subtracted from completion (#28)", () => {
+test("normalizeUsage: xAI/Grok-style — reasoning is ADDITIVE, not subtracted from completion", () => {
     // Real grok-4.3 shape: completion_tokens is visible-only; reasoning_tokens is
     // detailed but ADDITIVE — total = prompt + completion + reasoning.
     const u = normalizeUsage({
@@ -48,6 +48,17 @@ test("normalizeUsage: top-level cached_tokens still honored", () => {
     assert.equal(u.cached, 12);
 });
 
+test("#157: normalizeUsage maps DeepSeek's prompt cache hit count", () => {
+    const u = normalizeUsage({
+        prompt_tokens: 50,
+        prompt_cache_hit_tokens: 30,
+        prompt_cache_miss_tokens: 20,
+        completion_tokens: 10,
+        total_tokens: 60,
+    });
+    assert.deepEqual(u, { prompt: 50, completion: 10, reasoning: 0, cached: 30, total: 60 });
+});
+
 test("normalizeUsage: no reasoning — plain prompt+completion", () => {
     const u = normalizeUsage({ prompt_tokens: 10, completion_tokens: 20, total_tokens: 30 });
     assert.deepEqual(u, { prompt: 10, completion: 20, reasoning: 0, cached: 0, total: 30 });
@@ -63,9 +74,9 @@ test("normalizeUsage: absent usage → all zeros", () => {
     assert.deepEqual(normalizeUsage(undefined), { prompt: 0, completion: 0, reasoning: 0, cached: 0, total: 0 });
 });
 
-// -- Fireworks-style: reasoning shipped as TEXT, folded into completion, not itemized (#425) --
+// -- Fireworks-style: reasoning shipped as TEXT, folded into completion, not itemized --
 
-test("normalizeUsage: fireworks folds reasoning into completion -- re-split by text proportion, sum preserved (#425)", () => {
+test("normalizeUsage: fireworks folds reasoning into completion -- re-split by text proportion, sum preserved", () => {
     // total = prompt + completion (no gap), reasoning_tokens absent, but 750 vs 250
     // chars of reasoning vs content came back. Split completion 75/25; cost base held.
     const u = normalizeUsage(
@@ -78,7 +89,7 @@ test("normalizeUsage: fireworks folds reasoning into completion -- re-split by t
     assert.equal(u.prompt + u.completion + u.reasoning, u.total); // invariant
 });
 
-test("normalizeUsage: pure-reasoning turn (empty content) attributes all completion to reasoning (#425)", () => {
+test("normalizeUsage: pure-reasoning turn (empty content) attributes all completion to reasoning", () => {
     // The run52 runaway shape: 0 visible content, the whole budget spent reasoning.
     const u = normalizeUsage(
         { prompt_tokens: 100, completion_tokens: 500, total_tokens: 600 },

package/src/usage.ts

@@ -20,16 +20,39 @@ export type RawUsage = {
     completion_tokens?: number;
     total_tokens?: number;
     cached_tokens?: number;
+    prompt_cache_hit_tokens?: number;
+    prompt_cache_miss_tokens?: number;
     prompt_tokens_details?: { cached_tokens?: number };
     completion_tokens_details?: { reasoning_tokens?: number };
 };
 
+// Some providers return distinct reasoning text while reporting one combined
+// output count. Attribute that count without disturbing an upstream split.
+export const attributeUnitemizedReasoning = (
+    usage: ProviderUsage,
+    reasoningText: string,
+    contentText: string,
+): ProviderUsage => {
+    if (usage.reasoning !== 0 || usage.completion === 0 || reasoningText.length === 0) return usage;
+    const reasoning = contentText.length === 0
+        ? usage.completion
+        : Math.round(usage.completion * reasoningText.length / (reasoningText.length + contentText.length));
+    return {
+        ...usage,
+        completion: usage.completion - reasoning,
+        reasoning,
+    };
+};
+
 export const normalizeUsage = (raw: RawUsage | null | undefined, reasoningText = "", contentText = ""): ProviderUsage => {
     const prompt = raw?.prompt_tokens ?? 0;
     const completionRaw = raw?.completion_tokens ?? 0;
     const reportedTotal = raw?.total_tokens ?? 0;
     // OpenAI nests cached under prompt_tokens_details; others put it top-level.
-    const cached = raw?.prompt_tokens_details?.cached_tokens ?? raw?.cached_tokens ?? 0;
+    const cached = raw?.prompt_tokens_details?.cached_tokens
+        ?? raw?.prompt_cache_hit_tokens
+        ?? raw?.cached_tokens
+        ?? 0;
     const reasoningDetail = raw?.completion_tokens_details?.reasoning_tokens;
 
     let completion: number;
@@ -40,7 +63,7 @@ export const normalizeUsage = (raw: RawUsage | null | undefined, reasoningText =
         // o-series folds it INTO completion_tokens (subset: total = prompt + completion,
         // so completion must have reasoning subtracted out). xAI/Grok reports it
         // ADDITIVE to a visible-only completion_tokens (total = prompt + completion +
-        // reasoning), where subtracting wrongly zeroes the visible output (#28). Tell
+        // reasoning), where subtracting wrongly zeroes the visible output. Tell
         // them apart by the total identity; with no total reported, fall back on the
         // impossible-subset signal — reasoning can't exceed the completion it's a
         // subset of.
@@ -53,20 +76,15 @@ export const normalizeUsage = (raw: RawUsage | null | undefined, reasoningText =
         // reasoning. Only trust the gap when a total was actually reported.
         reasoning = reportedTotal > 0 ? Math.max(0, reportedTotal - prompt - completionRaw) : 0;
         completion = completionRaw;
-        // Fireworks folds reasoning INTO completion_tokens and itemizes no
-        // reasoning_tokens, so a turn that shipped only reasoning reads reasoning=0
-        // though 300k chars of it arrived (#425). When reasoning TEXT came back but
-        // the reported totals leave no gap, re-split the reported completion by the
-        // emitted text proportions. Sum-preserving: billable output
-        // (completion+reasoning) and cost are byte-identical; only the
-        // visible/reasoning gauge is corrected (pure-reasoning turn -> completion 0).
-        if (reasoning === 0 && reasoningText.length > 0 && reportedTotal > 0 && completionRaw > 0) {
-            reasoning = Math.round(completionRaw * reasoningText.length / (reasoningText.length + contentText.length));
-            completion = completionRaw - reasoning;
-        }
     }
     const total = reportedTotal > 0 ? reportedTotal : prompt + completion + reasoning;
-    return { prompt, completion, reasoning, cached, total };
+    const usage = { prompt, completion, reasoning, cached, total };
+    // Fireworks folds reasoning INTO completion_tokens and itemizes no
+    // reasoning_tokens. A reported total establishes that completion is an
+    // upstream quantity rather than a locally synthesized fallback.
+    return reasoningDetail === undefined && reportedTotal > 0
+        ? attributeUnitemizedReasoning(usage, reasoningText, contentText)
+        : usage;
 };
 
 // Conventional provider pricing: USD per million tokens, matching Models.dev.

package/src/warnings.test.ts

@@ -4,24 +4,24 @@ import { emitWarningOnce, resetEmittedWarnings } from "./warnings.ts";
 
 test.afterEach(() => { mock.restoreAll(); resetEmittedWarnings(); });
 
-test("#40: same (code, message) fires once per process", () => {
+test("same (code, message) fires once per process", () => {
     const seen: string[] = [];
     mock.method(process, "emitWarning", (msg: string | Error) => { seen.push(String(msg)); });
-    emitWarningOnce("openai provider: heuristic", "PLURNK_TOKENIZER_HEURISTIC");
-    emitWarningOnce("openai provider: heuristic", "PLURNK_TOKENIZER_HEURISTIC");
-    emitWarningOnce("openai provider: heuristic", "PLURNK_TOKENIZER_HEURISTIC");
-    assert.deepEqual(seen, ["openai provider: heuristic"]);
+    emitWarningOnce("openai provider: estimate", "PLURNK_PROMPT_COUNT_ESTIMATE");
+    emitWarningOnce("openai provider: estimate", "PLURNK_PROMPT_COUNT_ESTIMATE");
+    emitWarningOnce("openai provider: estimate", "PLURNK_PROMPT_COUNT_ESTIMATE");
+    assert.deepEqual(seen, ["openai provider: estimate"]);
 });
 
-test("#40: dedup is by (code, MESSAGE), not code — a second provider's surfacing is never suppressed", () => {
+test("dedup is by (code, MESSAGE), not code — a second provider's surfacing is never suppressed", () => {
     const seen: string[] = [];
     mock.method(process, "emitWarning", (msg: string | Error) => { seen.push(String(msg)); });
-    emitWarningOnce("openai provider: heuristic", "PLURNK_TOKENIZER_HEURISTIC");
-    emitWarningOnce("groq provider: heuristic", "PLURNK_TOKENIZER_HEURISTIC"); // same code, different provider
-    assert.deepEqual(seen, ["openai provider: heuristic", "groq provider: heuristic"]);
+    emitWarningOnce("openai provider: estimate", "PLURNK_PROMPT_COUNT_ESTIMATE");
+    emitWarningOnce("groq provider: estimate", "PLURNK_PROMPT_COUNT_ESTIMATE"); // same code, different provider
+    assert.deepEqual(seen, ["openai provider: estimate", "groq provider: estimate"]);
 });
 
-test("#40: resetEmittedWarnings clears the set (test-order independence)", () => {
+test("resetEmittedWarnings clears the set (test-order independence)", () => {
     const seen: string[] = [];
     mock.method(process, "emitWarning", (msg: string | Error) => { seen.push(String(msg)); });
     emitWarningOnce("m", "C");

package/dist/OpenAICompat.d.ts

@@ -1,76 +0,0 @@
-import type { ChatMessage, Provider, ProviderResponse, ProviderUsage } from "./types.ts";
-import type { Reasoning, ReserveSpec } from "./env.ts";
-export type ProviderFetch = typeof globalThis.fetch;
-export type ReasoningStyle = "none" | "think" | "include_reasoning" | "effort" | "effort_explicit" | "template" | "anthropic";
-export type GrammarStyle = "none" | "llamacpp";
-export type OpenAICompatConfig = {
-    model: string;
-    url: string;
-    fetchTimeoutMs: number;
-    streamIdleTimeoutMs?: number;
-    headers?: Record<string, string>;
-    fetch?: ProviderFetch;
-    contextWindow?: number | null;
-    reasoningStyle?: ReasoningStyle;
-    countTokens?: (text: string) => number;
-    calculateCost?: (usage: ProviderUsage) => number;
-    source?: string;
-    grammarStyle?: GrammarStyle;
-    promptCacheKey?: boolean;
-    serviceTier?: string;
-    gbnfDebug?: boolean;
-    streaming?: boolean;
-    firstPartyMetadata?: boolean;
-    apiKeyRejectedMessage?: string;
-    eosText?: string;
-    supportsSlotPinning?: boolean;
-    slotCount?: number | null;
-    tokenizeUrl?: string;
-    servedModel?: string;
-    requiresMaxTokens?: boolean;
-    reasoning: Reasoning;
-    temperature: number;
-    repeatPenalty: number;
-    frequencyPenalty?: number;
-    dryMultiplier?: number;
-    dryBase?: number;
-    dryAllowedLength?: number;
-    repeatLastN?: number;
-    retryAttempts: number;
-    topLogprobs?: number | null;
-    rawBody?: boolean;
-    reasoningReserve?: ReserveSpec;
-    completionReserve?: ReserveSpec;
-    tuningFloors?: boolean;
-};
-export declare const effortFromBudget: (budget: number) => "low" | "medium" | "high";
-export default class OpenAICompatProvider implements Provider {
-    #private;
-    tokenize?: (text: string) => Promise<number[]>;
-    constructor(config: OpenAICompatConfig);
-    get contextWindow(): number | null;
-    get reasoningReserve(): number | null;
-    get completionReserve(): number | null;
-    get model(): string;
-    get servedModel(): string | undefined;
-    get requiresMaxTokens(): boolean | undefined;
-    get constrainsOutput(): boolean;
-    countTokens(text: string): number;
-    calculateCost(usage: ProviderUsage): number;
-    generate({ messages, workerId, primaryWorkerId, signal, grammar, maxTokens, attributions, client, strikes, workspaceId, loop, turn, sampling }: {
-        messages: ChatMessage[];
-        workerId: string;
-        primaryWorkerId?: string;
-        signal?: AbortSignal;
-        grammar?: string;
-        maxTokens?: number;
-        attributions?: string[];
-        client?: string;
-        strikes?: number;
-        workspaceId?: string;
-        loop?: number;
-        turn?: number;
-        sampling?: Record<string, unknown>;
-    }): Promise<ProviderResponse>;
-}
-//# sourceMappingURL=OpenAICompat.d.ts.map

package/dist/OpenAICompat.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"OpenAICompat.d.ts","sourceRoot":"","sources":["../src/OpenAICompat.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EAAE,WAAW,EAAgB,QAAQ,EAAE,gBAAgB,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AACvG,OAAO,KAAK,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,UAAU,CAAC;AAMvD,MAAM,MAAM,aAAa,GAAG,OAAO,UAAU,CAAC,KAAK,CAAC;AAapD,MAAM,MAAM,cAAc,GAAG,MAAM,GAAG,OAAO,GAAG,mBAAmB,GAAG,QAAQ,GAAG,iBAAiB,GAAG,UAAU,GAAG,WAAW,CAAC;AAI9H,MAAM,MAAM,YAAY,GAAG,MAAM,GAAG,UAAU,CAAC;AAE/C,MAAM,MAAM,kBAAkB,GAAG;IAC7B,KAAK,EAAE,MAAM,CAAC;IACd,GAAG,EAAE,MAAM,CAAC;IACZ,cAAc,EAAE,MAAM,CAAC;IACvB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,KAAK,CAAC,EAAE,aAAa,CAAC;IACtB,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,cAAc,CAAC,EAAE,cAAc,CAAC;IAChC,WAAW,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,MAAM,CAAC;IACvC,aAAa,CAAC,EAAE,CAAC,KAAK,EAAE,aAAa,KAAK,MAAM,CAAC;IACjD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,YAAY,CAAC,EAAE,YAAY,CAAC;IAM5B,cAAc,CAAC,EAAE,OAAO,CAAC;IAGzB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAC7B,qBAAqB,CAAC,EAAE,MAAM,CAAC;IAC/B,OAAO,CAAC,EAAE,MAAM,CAAC;IAEjB,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAI1B,WAAW,CAAC,EAAE,MAAM,CAAC;IAKrB,WAAW,CAAC,EAAE,MAAM,CAAC;IAIrB,iBAAiB,CAAC,EAAE,OAAO,CAAC;IAM5B,SAAS,EAAE,SAAS,CAAC;IAOrB,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,MAAM,CAAC;IAKtB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAO1B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,WAAW,CAAC,EAAE,MAAM,CAAC;IAIrB,aAAa,EAAE,MAAM,CAAC;IAQtB,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,OAAO,CAAC,EAAE,OAAO,CAAC;IAOlB,gBAAgB,CAAC,EAAE,WAAW,CAAC;IAC/B,iBAAiB,CAAC,EAAE,WAAW,CAAC;IAIhC,YAAY,CAAC,EAAE,OAAO,CAAC;CAC1B,CAAC;AAeF,eAAO,MAAM,gBAAgB,WAAY,MAAM,KAAG,KAAK,GAAG,QAAQ,GAAG,MAIpE,CAAC;AAwCF,MAAM,CAAC,OAAO,OAAO,oBAAqB,YAAW,QAAQ;;IA4CzD,QAAQ,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;IAE/C,YAAY,MAAM,EAAE,kBAAkB,EA8DrC;IAED,IAAI,aAAa,IAAI,MAAM,GAAG,IAAI,CAAgC;IAQlE,IAAI,gBAAgB,IAAI,MAAM,GAAG,IAAI,CAAyD;IAC9F,IAAI,iBAAiB,IAAI,MAAM,GAAG,IAAI,CAA0D;IAChG,IAAI,KAAK,IAAI,MAAM,CAAwB;IAE3C,IAAI,WAAW,IAAI,MAAM,GAAG,SAAS,CAA8B;IAEnE,IAAI,iBAAiB,IAAI,OAAO,GAAG,SAAS,CAAoC;IAIhF,IAAI,gBAAgB,IAAI,OAAO,CAA0C;IAEzE,WAAW,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAAoC;IACrE,aAAa,CAAC,KAAK,EAAE,aAAa,GAAG,MAAM,CAAuC;IAuN5E,QAAQ,CAAC,EAAE,QAAQ,EAAE,QAAQ,EAAE,eAAe,EAAE,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,YAAY,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE;QAAE,QAAQ,EAAE,WAAW,EAAE,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAC;QAAC,eAAe,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,WAAW,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,SAAS,CAAC,EAAE,MAAM,CAAC;QAAC,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,WAAW,CAAC,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;KAAE,GAAG,OAAO,CAAC,gBAAgB,CAAC,CA0Ixc;CACJ"}