fix(backend): distinct max_input_tokens for GPT-family models (#6191) (#6230)

This commit is contained in:
Diego Rodrigues de Sa e Souza
2026-07-04 23:02:55 -03:00
committed by GitHub
parent adde9e4bae
commit 1473261c4c
7 changed files with 102 additions and 4 deletions

View File

@@ -292,9 +292,13 @@ export async function getUnifiedModelsResponse(
registryContext ??
specContext ??
(getTokenLimit(providerId, modelId) || undefined);
const maxInputTokens = isPositiveFiniteNumber(synced?.limit_input)
const registryInputLimit = isPositiveFiniteNumber(registryModel?.maxInputTokens)
? registryModel.maxInputTokens
: undefined;
const syncedInputLimit = isPositiveFiniteNumber(synced?.limit_input)
? synced.limit_input
: contextLength;
: undefined;
const maxInputTokens = registryInputLimit ?? syncedInputLimit ?? contextLength;
const maxOutputTokens = isPositiveFiniteNumber(synced?.limit_output)
? synced.limit_output
: isPositiveFiniteNumber(spec?.maxOutputTokens)

View File

@@ -406,7 +406,11 @@ export function getResolvedModelCapabilities(input: CapabilityInput): ResolvedMo
structuredOutput: synced?.structured_output ?? null,
temperature: synced?.temperature ?? null,
contextWindow,
maxInputTokens: authoritativeContextWindow ?? synced?.limit_input ?? contextWindow,
maxInputTokens:
(typeof registryModel?.maxInputTokens === "number" ? registryModel.maxInputTokens : null) ??
authoritativeContextWindow ??
synced?.limit_input ??
contextWindow,
maxOutputTokens:
synced?.limit_output ??
(typeof registryModel?.maxOutputTokens === "number" ? registryModel.maxOutputTokens : null) ??