248d7da01c
The merged Qoder work also rewrote shared translator/handler code so that /v1/responses clients got token usage on response.completed. That changed behaviour for every provider, not just Qoder: proxies saw input tokens rise by the 2000-token context buffer, and the plain token mapping was replaced by one that always adds input_tokens_details. A probe confirms the Qoder benefit does not depend on those edits: the executor's coalescer already emits one include_usage-style finish chunk, so a Claude client receives input_tokens and cache_read_input_tokens with every shared file at its original state. Only the Responses path relies on the shared translator, and that path has no Qoder-owned seam to put it in. Reverts the shared files to their pre-PR state and drops the Responses usage test. The Cline envelope unwrap in nonStreamingHandler.js, which landed after the PR in the same file, is kept.
70 lines
3.5 KiB
JavaScript
70 lines
3.5 KiB
JavaScript
// Build OpenAI usage object. Caller computes prompt/completion/total (provider math).
|
|
// Optional details added only when > 0 (matches existing claude/gemini/codex behavior).
|
|
export function buildUsage({ promptTokens, completionTokens, totalTokens, cachedTokens = 0, cacheCreationTokens = 0, reasoningTokens = 0 }) {
|
|
const usage = { prompt_tokens: promptTokens, completion_tokens: completionTokens, total_tokens: totalTokens };
|
|
if (cachedTokens > 0 || cacheCreationTokens > 0) {
|
|
usage.prompt_tokens_details = {};
|
|
if (cachedTokens > 0) usage.prompt_tokens_details.cached_tokens = cachedTokens;
|
|
if (cacheCreationTokens > 0) usage.prompt_tokens_details.cache_creation_tokens = cacheCreationTokens;
|
|
}
|
|
if (reasoningTokens > 0) {
|
|
usage.completion_tokens_details = { reasoning_tokens: reasoningTokens };
|
|
}
|
|
return usage;
|
|
}
|
|
|
|
const n = (v) => (typeof v === "number" ? v : 0);
|
|
|
|
// Per-provider raw token field-map + math. Returns buildUsage() args (NOT the usage object).
|
|
// Keeps each provider's exact semantics: claude/gemini fold cache+reasoning, others don't.
|
|
const USAGE_EXTRACTORS = {
|
|
claude(raw) {
|
|
const input = n(raw.input_tokens), output = n(raw.output_tokens);
|
|
const cacheRead = n(raw.cache_read_input_tokens), cacheCreate = n(raw.cache_creation_input_tokens);
|
|
const prompt = input + cacheRead + cacheCreate;
|
|
return { promptTokens: prompt, completionTokens: output, totalTokens: prompt + output, cachedTokens: cacheRead, cacheCreationTokens: cacheCreate };
|
|
},
|
|
gemini(raw) {
|
|
const cached = n(raw.cachedContentTokenCount);
|
|
const prompt = n(raw.promptTokenCount);
|
|
const thoughts = n(raw.thoughtsTokenCount);
|
|
const total = n(raw.totalTokenCount);
|
|
let candidates = n(raw.candidatesTokenCount);
|
|
// Fallback: derive candidates from total when upstream omits it
|
|
if (candidates === 0 && total > 0) {
|
|
candidates = total - prompt - thoughts;
|
|
if (candidates < 0) candidates = 0;
|
|
}
|
|
return { promptTokens: prompt, completionTokens: candidates + thoughts, totalTokens: total, cachedTokens: cached, reasoningTokens: thoughts };
|
|
},
|
|
kiro(raw) {
|
|
const input = n(raw.inputTokens), output = n(raw.outputTokens);
|
|
// ponytail: Amazon Q (Kiro upstream) does not expose cache fields today,
|
|
// but pass through any cache_read/cache_creation/cached_tokens if the
|
|
// event shape grows them later so cost tracking keeps working without
|
|
// a second pass.
|
|
const cached = n(raw.cache_read_input_tokens) || n(raw.cachedTokens) || n(raw.cached_tokens);
|
|
const cacheCreation = n(raw.cache_creation_input_tokens);
|
|
const out = { promptTokens: input, completionTokens: output, totalTokens: input + output };
|
|
if (cached > 0) out.cachedTokens = cached;
|
|
if (cacheCreation > 0) out.cacheCreationTokens = cacheCreation;
|
|
return out;
|
|
},
|
|
ollama(raw) {
|
|
const input = n(raw.prompt_eval_count), output = n(raw.eval_count);
|
|
return { promptTokens: input, completionTokens: output, totalTokens: input + output };
|
|
},
|
|
commandcode(raw) {
|
|
const input = n(raw.inputTokens), output = n(raw.outputTokens);
|
|
const total = typeof raw.totalTokens === "number" ? raw.totalTokens : input + output;
|
|
return { promptTokens: input, completionTokens: output, totalTokens: total };
|
|
},
|
|
};
|
|
|
|
// Convert provider-native usage object → OpenAI usage. Returns null if no extractor/raw.
|
|
export function toOpenAIUsage(raw, kind) {
|
|
const extract = USAGE_EXTRACTORS[kind];
|
|
if (!extract || !raw || typeof raw !== "object") return null;
|
|
return buildUsage(extract(raw));
|
|
}
|