Skip to content

Open source CapCut alternative.

TypeScript

refactor(site): share the usage cost function

David Leecc63f846879
diff --git a/site/src/lib/credits/inference.ts b/site/src/lib/credits/inference.ts
index c4fef525..3115c123 100644
--- a/site/src/lib/credits/inference.ts
+++ b/site/src/lib/credits/inference.ts
@@ -8,6 +8,8 @@ import {
} from "@/lib/credits/amounts";
import {
providerCreditPricing,
+ providerPricingCostMicros,
+ type NormalizedUsage,
type ProviderCreditPricing,
} from "@/lib/credits/provider-pricing";
import { isJsonObject, toJsonValue } from "@/lib/inference/json";
@@ -35,19 +37,6 @@ type CreditRateSnapshot = {
providerPricing?: ProviderCreditPricing;
};
-type NormalizedUsage = {
- billableOutputTokens: bigint;
- cachedInputTokens: bigint;
- durationMillis: bigint;
- generationCount: bigint;
- inputTokens: bigint;
- inputAudioTokens: bigint;
- outputTokens: bigint;
- outputAudioTokens: bigint;
- totalTokens: bigint;
- characterCost: bigint;
-};
-
type InferenceUsageStatus = "failed" | "succeeded";
type InferenceUsageInput = {
@@ -1239,7 +1228,17 @@ function zeroCreditRate(): CreditRateSnapshot {
};
}
-const unitScalePerMillion = BigInt("1000000");
+/** What one unit of usage on a route costs, through the same rate lookup and
+ * cost function as the charge: a database override wins over the code price. */
+export async function usageCostMicros(
+ route: string,
+ provider: string,
+ model: string,
+ usage: NormalizedUsage,
+): Promise<bigint> {
+ const rate = await resolveCreditRate(prisma, route, provider, model);
+ return costForUsage(rate, usage);
+}
function costForUsage(rate: CreditRateSnapshot, usage: NormalizedUsage) {
if (rate.providerPricing) {
@@ -1270,104 +1269,10 @@ function costForProviderPricing(
rate: CreditRateSnapshot,
usage: NormalizedUsage,
) {
- const pricing = providerPricingForUsage(rate.providerPricing, usage);
- if (!pricing) {
- return rate.fallbackCostMicros;
- }
-
- const inputAudioTokens = minBigint(usage.inputAudioTokens, usage.inputTokens);
- const outputAudioTokens = minBigint(usage.outputAudioTokens, usage.outputTokens);
- const cachedInputTokens = minBigint(
- usage.cachedInputTokens,
- usage.inputTokens - inputAudioTokens,
- );
- const uncachedInputTokens = maxBigint(
- zeroCreditMicros,
- usage.inputTokens - inputAudioTokens - cachedInputTokens,
- );
- const textOutputTokens = maxBigint(
- zeroCreditMicros,
- usage.billableOutputTokens - outputAudioTokens,
- );
- const inputTokenRate = pricing.inputTokenCostMicrosPerMillion ?? zeroCreditMicros;
- const cachedInputTokenRate =
- pricing.cachedInputTokenCostMicrosPerMillion ?? inputTokenRate;
- const outputTokenRate = pricing.outputTokenCostMicrosPerMillion ?? zeroCreditMicros;
- const inputAudioTokenRate =
- pricing.inputAudioTokenCostMicrosPerMillion ?? inputTokenRate;
- const outputAudioTokenRate =
- pricing.outputAudioTokenCostMicrosPerMillion ?? outputTokenRate;
-
- const tokenAndUnitCostMicros =
- perMillionCost(uncachedInputTokens, inputTokenRate) +
- perMillionCost(cachedInputTokens, cachedInputTokenRate) +
- perMillionCost(inputAudioTokens, inputAudioTokenRate) +
- perMillionCost(textOutputTokens, outputTokenRate) +
- perMillionCost(outputAudioTokens, outputAudioTokenRate) +
- usage.characterCost * (pricing.characterCostMicros ?? zeroCreditMicros) +
- usage.generationCount * (pricing.generationCostMicros ?? zeroCreditMicros);
- const durationCostMicros = ceilDivide(
- usage.durationMillis * (pricing.durationSecondCostMicros ?? zeroCreditMicros),
- BigInt(1000),
- );
- const costMicros =
- tokenAndUnitCostMicros +
- (pricing.durationCostOnlyWhenNoTokenUsage && tokenAndUnitCostMicros > zeroCreditMicros
- ? zeroCreditMicros
- : durationCostMicros);
-
+ const costMicros = providerPricingCostMicros(rate.providerPricing, usage);
return costMicros > zeroCreditMicros ? costMicros : rate.fallbackCostMicros;
}
-function providerPricingForUsage(
- pricing: ProviderCreditPricing | undefined,
- usage: NormalizedUsage,
-) {
- if (!pricing) {
- return undefined;
- }
-
- const contextTokens = usage.totalTokens > zeroCreditMicros
- ? usage.totalTokens
- : usage.inputTokens + usage.outputTokens;
- if (
- pricing.longContext &&
- pricing.longContextThresholdTokens &&
- contextTokens > pricing.longContextThresholdTokens
- ) {
- return {
- ...pricing,
- ...pricing.longContext,
- longContext: undefined,
- };
- }
-
- return pricing;
-}
-
-function perMillionCost(units: bigint, microsPerMillion: bigint) {
- if (units <= zeroCreditMicros || microsPerMillion <= zeroCreditMicros) {
- return zeroCreditMicros;
- }
-
- return ceilDivide(units * microsPerMillion, unitScalePerMillion);
-}
-
-function ceilDivide(numerator: bigint, denominator: bigint) {
- if (denominator <= zeroCreditMicros) {
- throw new Error("Cannot divide by zero.");
- }
- if (numerator <= zeroCreditMicros) {
- return zeroCreditMicros;
- }
-
- return (numerator + denominator - BigInt(1)) / denominator;
-}
-
-function minBigint(left: bigint, right: bigint) {
- return left < right ? left : right;
-}
-
function maxBigint(left: bigint, right: bigint) {
return left > right ? left : right;
}
diff --git a/site/src/lib/credits/provider-pricing.ts b/site/src/lib/credits/provider-pricing.ts
index 83459d80..0e36d716 100644
--- a/site/src/lib/credits/provider-pricing.ts
+++ b/site/src/lib/credits/provider-pricing.ts
@@ -576,3 +576,117 @@ function ceilDivide(numerator: bigint, denominator: bigint) {
return (numerator + denominator - BigInt(1)) / denominator;
}
+
+/** Usage in the units provider pricing bills: tokens, characters, rendered
+ * milliseconds, and flat generations. */
+export type NormalizedUsage = {
+ billableOutputTokens: bigint;
+ cachedInputTokens: bigint;
+ durationMillis: bigint;
+ generationCount: bigint;
+ inputTokens: bigint;
+ inputAudioTokens: bigint;
+ outputTokens: bigint;
+ outputAudioTokens: bigint;
+ totalTokens: bigint;
+ characterCost: bigint;
+};
+
+const unitScalePerMillion = BigInt("1000000");
+const millisPerSecond = BigInt(1000);
+
+/** What a usage costs at a provider's code price, in micros; zero with no
+ * price. The charge and the price sheet both run through here. */
+export function providerPricingCostMicros(
+ providerPricing: ProviderCreditPricing | undefined,
+ usage: NormalizedUsage,
+): bigint {
+ const pricing = providerPricingForUsage(providerPricing, usage);
+ if (!pricing) {
+ return zeroCreditMicros;
+ }
+
+ const inputAudioTokens = minBigint(usage.inputAudioTokens, usage.inputTokens);
+ const outputAudioTokens = minBigint(usage.outputAudioTokens, usage.outputTokens);
+ const cachedInputTokens = minBigint(
+ usage.cachedInputTokens,
+ usage.inputTokens - inputAudioTokens,
+ );
+ const uncachedInputTokens = maxBigint(
+ zeroCreditMicros,
+ usage.inputTokens - inputAudioTokens - cachedInputTokens,
+ );
+ const textOutputTokens = maxBigint(
+ zeroCreditMicros,
+ usage.billableOutputTokens - outputAudioTokens,
+ );
+ const inputTokenRate = pricing.inputTokenCostMicrosPerMillion ?? zeroCreditMicros;
+ const cachedInputTokenRate =
+ pricing.cachedInputTokenCostMicrosPerMillion ?? inputTokenRate;
+ const outputTokenRate = pricing.outputTokenCostMicrosPerMillion ?? zeroCreditMicros;
+ const inputAudioTokenRate =
+ pricing.inputAudioTokenCostMicrosPerMillion ?? inputTokenRate;
+ const outputAudioTokenRate =
+ pricing.outputAudioTokenCostMicrosPerMillion ?? outputTokenRate;
+
+ const tokenAndUnitCostMicros =
+ perMillionCost(uncachedInputTokens, inputTokenRate) +
+ perMillionCost(cachedInputTokens, cachedInputTokenRate) +
+ perMillionCost(inputAudioTokens, inputAudioTokenRate) +
+ perMillionCost(textOutputTokens, outputTokenRate) +
+ perMillionCost(outputAudioTokens, outputAudioTokenRate) +
+ usage.characterCost * (pricing.characterCostMicros ?? zeroCreditMicros) +
+ usage.generationCount * (pricing.generationCostMicros ?? zeroCreditMicros);
+ const durationCostMicros = ceilDivide(
+ usage.durationMillis * (pricing.durationSecondCostMicros ?? zeroCreditMicros),
+ millisPerSecond,
+ );
+ return (
+ tokenAndUnitCostMicros +
+ (pricing.durationCostOnlyWhenNoTokenUsage && tokenAndUnitCostMicros > zeroCreditMicros
+ ? zeroCreditMicros
+ : durationCostMicros)
+ );
+}
+
+function providerPricingForUsage(
+ pricing: ProviderCreditPricing | undefined,
+ usage: NormalizedUsage,
+) {
+ if (!pricing) {
+ return undefined;
+ }
+
+ const contextTokens = usage.totalTokens > zeroCreditMicros
+ ? usage.totalTokens
+ : usage.inputTokens + usage.outputTokens;
+ if (
+ pricing.longContext &&
+ pricing.longContextThresholdTokens &&
+ contextTokens > pricing.longContextThresholdTokens
+ ) {
+ return {
+ ...pricing,
+ ...pricing.longContext,
+ longContext: undefined,
+ };
+ }
+
+ return pricing;
+}
+
+function perMillionCost(units: bigint, microsPerMillion: bigint) {
+ if (units <= zeroCreditMicros || microsPerMillion <= zeroCreditMicros) {
+ return zeroCreditMicros;
+ }
+
+ return ceilDivide(units * microsPerMillion, unitScalePerMillion);
+}
+
+function minBigint(left: bigint, right: bigint) {
+ return left < right ? left : right;
+}
+
+function maxBigint(left: bigint, right: bigint) {
+ return left > right ? left : right;
+}
diff --git a/site/src/lib/inference/adapters/gemini-omni-video.ts b/site/src/lib/inference/adapters/gemini-omni-video.ts
index fcbf7491..2d8dd761 100644
--- a/site/src/lib/inference/adapters/gemini-omni-video.ts
+++ b/site/src/lib/inference/adapters/gemini-omni-video.ts
@@ -16,9 +16,8 @@ import {
geminiOmniDurationSeconds,
geminiOmniMaxReferenceImages,
geminiOmniModels,
+ geminiOmniRenderUnits,
geminiOmniResolutions,
- geminiOmniTokensPerSecond,
- geminiOmniUnitTokens,
type GeminiOmniResolution,
} from "@/lib/inference/gemini-models";
import { ensureConfigured } from "@/lib/inference/http";
@@ -140,17 +139,12 @@ export function createGeminiOmniVideoAssetProvider(
return { resolution: resolutionRaw as GeminiOmniResolution, durationSeconds };
}
- // Billing units for a request: seconds of 360p-equivalent output, from the
- // length (the model's ceiling when unset) times the resolution's tokens per
- // second. The route holds the balance to this count before the submit, and
- // the submit bills the same count.
+ // Billing units for a request (geminiOmniRenderUnits). The route holds the
+ // balance to this count before the submit, and the submit bills the same
+ // count.
function generationUnits(parameters: JsonObject): number {
const { resolution, durationSeconds } = renderShape(parameters);
- const seconds = durationSeconds ?? geminiOmniDurationSeconds.max;
- return Math.max(
- 1,
- Math.ceil((seconds * geminiOmniTokensPerSecond[resolution]) / geminiOmniUnitTokens),
- );
+ return geminiOmniRenderUnits(resolution, durationSeconds);
}
async function generateAsset({
diff --git a/site/src/lib/inference/gemini-models.ts b/site/src/lib/inference/gemini-models.ts
index 6bdc4b77..b5b80376 100644
--- a/site/src/lib/inference/gemini-models.ts
+++ b/site/src/lib/inference/gemini-models.ts
@@ -66,6 +66,18 @@ export const geminiOmniTokensPerSecond: Record<GeminiOmniResolution, number> = {
* count the same thing. */
export const geminiOmniUnitTokens = geminiOmniTokensPerSecond["360p"];
+/** Billing units for a render: seconds of 360p-equivalent output, from the
+ * length (the model's ceiling when unset) times the resolution's tokens per
+ * second. A 10s 720p clip is 30 units. The submit bills this count and the
+ * price on the Generate button reads it. */
+export function geminiOmniRenderUnits(
+ resolution: GeminiOmniResolution,
+ durationSeconds: number | undefined,
+): number {
+ const seconds = durationSeconds ?? geminiOmniDurationSeconds.max;
+ return Math.max(1, Math.ceil((seconds * geminiOmniTokensPerSecond[resolution]) / geminiOmniUnitTokens));
+}
+
// Generative speech (Gemini TTS): text in, spoken audio out, with prompt-driven
// style direction and inline audio tags. Bump here when adopting a newer TTS model.
export const geminiTtsModels = {