zen: stat worker
This commit is contained in:
parent
784796e27a
commit
77e51b0a32
6 changed files with 26 additions and 16 deletions
|
|
@ -320,6 +320,7 @@ export async function handler(
|
|||
await modelTpsLimiter?.track(
|
||||
providerInfo.id,
|
||||
providerInfo.model,
|
||||
providerInfo.tpsGoal,
|
||||
timestampFirstByte,
|
||||
timestampLastByte,
|
||||
usageInfo,
|
||||
|
|
@ -525,7 +526,7 @@ export async function handler(
|
|||
})
|
||||
.filter((provider) => {
|
||||
if (!provider.tpsGoal) return true
|
||||
const isLowTps = modelTpsLimits?.[`${provider.id}/${provider.model}`] ?? false
|
||||
const isLowTps = modelTpsLimits?.[`${provider.id}/${provider.model}/${provider.tpsGoal}`] ?? false
|
||||
return !isLowTps
|
||||
})
|
||||
.map((provider) => {
|
||||
|
|
|
|||
|
|
@ -5,7 +5,7 @@ import { UsageInfo } from "./provider/provider"
|
|||
export function createModelTpsLimiter(providers: { id: string; model: string; tpsGoal?: number }[]) {
|
||||
const tpsGoals = Object.fromEntries(
|
||||
providers.flatMap((p) => {
|
||||
return p.tpsGoal ? [[`${p.id}/${p.model}`, p.tpsGoal]] : []
|
||||
return p.tpsGoal ? [[`${p.id}/${p.model}/${p.tpsGoal}`, p.tpsGoal]] : []
|
||||
}),
|
||||
)
|
||||
const ids = Object.keys(tpsGoals)
|
||||
|
|
@ -56,11 +56,17 @@ export function createModelTpsLimiter(providers: { id: string; model: string; tp
|
|||
}),
|
||||
)
|
||||
},
|
||||
track: async (provider: string, model: string, tsFirstByte: number, tsLastByte: number, usageInfo: UsageInfo) => {
|
||||
const id = `${provider}/${model}`
|
||||
if (!ids.includes(id)) return
|
||||
const tpsGoal = tpsGoals[id]
|
||||
track: async (
|
||||
provider: string,
|
||||
model: string,
|
||||
tpsGoal: number | undefined,
|
||||
tsFirstByte: number,
|
||||
tsLastByte: number,
|
||||
usageInfo: UsageInfo,
|
||||
) => {
|
||||
if (!tpsGoal) return
|
||||
const id = `${provider}/${model}/${tpsGoal}`
|
||||
if (!ids.includes(id)) return
|
||||
if (tsFirstByte <= 0 || tsLastByte <= 0) return
|
||||
const tokens = usageInfo.outputTokens
|
||||
if (tokens <= 10) return
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue