mirror of
https://github.com/diegosouzapw/OmniRoute.git
synced 2026-08-23 15:42:12 +03:00
28924fe05b
Validado no worktree combinado: mesmos gates + testes focados verdes. Fix bem medido (context window real vs anunciado divergindo por até 24h para modelos sincronizados fora do ciclo). CI vermelho é o base-red já rastreado em #9985.
59 lines
2.8 KiB
TypeScript
59 lines
2.8 KiB
TypeScript
// Per-model default reasoning effort (#6879, "Ask 1"). Many models think by
|
|
// default with no client-visible way to turn it off (measured:
|
|
// gemini-flash-lite-latest burns ~277 reasoning tokens on a plain request with
|
|
// no reasoning params). ModelSpec.defaultReasoningEffort lets an operator
|
|
// configure a strip-by-default (or steer-by-default) value fleet-wide without
|
|
// patching every client.
|
|
//
|
|
// Semantics: applied ONLY when the request carries no reasoning field of any
|
|
// shape (`reasoning_effort`, `reasoning`, `thinking`) — an explicit client
|
|
// value, including one forwarded verbatim through a combo leg, always wins
|
|
// and this is a no-op. Models without a configured default are untouched
|
|
// (regression-safe). Wired at the OpenAI-format dispatch chokepoint in
|
|
// chatCore.ts, after model resolution, so the *upstream* model's default is
|
|
// used even when a combo/route substituted it.
|
|
import { getModelSpec } from "@/shared/constants/modelSpecs.ts";
|
|
|
|
/** True when `body` already expresses a reasoning-effort choice, in any known shape. */
|
|
function hasExplicitReasoningField(body: Record<string, unknown>): boolean {
|
|
return (
|
|
body.reasoning_effort !== undefined ||
|
|
body.reasoning !== undefined ||
|
|
body.thinking !== undefined
|
|
);
|
|
}
|
|
|
|
/**
|
|
* Inject a resolved reasoning effort as `reasoning_effort` when the request has no
|
|
* reasoning field. Returns `body` unchanged (same reference) when there is nothing to
|
|
* inject, so callers can chain it without extra guards.
|
|
*
|
|
* `suffixEffort` (#7694) is the tier a `<prefix>/<model>-{effort}` synced-model alias
|
|
* resolved to (`src/sse/services/model.ts`'s `resolveSyncedModelIdAndEffort`) — an
|
|
* explicit, request-time model selection, so it takes priority over both defaults
|
|
* below when present.
|
|
*
|
|
* `syncedDefaultEffort` is the vendor-declared default captured at sync time
|
|
* (`reasoning.default_effort`, e.g. OpenRouter `stealth/ox-alpha` declares `max`) —
|
|
* see `detectDefaultThinkingEffort`. A model that only produces usable output with
|
|
* an explicit effort gets the vendor default instead of an empty upstream response.
|
|
* It is the LOWEST-priority default: an explicit client value wins, the suffix alias
|
|
* wins, and a static `ModelSpec.defaultReasoningEffort` (operator-configured
|
|
* strip-by-default, #6879) also wins over the vendor default.
|
|
*/
|
|
export function applyDefaultReasoningEffort<T extends Record<string, unknown>>(
|
|
body: T,
|
|
modelId: string,
|
|
suffixEffort?: string | null,
|
|
syncedDefaultEffort?: string | null
|
|
): T {
|
|
if (!body || typeof body !== "object") return body;
|
|
if (hasExplicitReasoningField(body)) return body;
|
|
|
|
const defaultEffort =
|
|
suffixEffort || getModelSpec(modelId)?.defaultReasoningEffort || syncedDefaultEffort;
|
|
if (!defaultEffort) return body;
|
|
|
|
return { ...body, reasoning_effort: defaultEffort };
|
|
}
|