toor
1e9ea89d86
Import the client, server, skills, tests, and build scripts as the starting point for the rebuild, then trim the surfaces this project does not ship: Electron desktop packaging, the marketing website, and the ESP32 firmware tree. CI keeps the web build, e2e, lockfile, release, and Docker workflows. Rename the package and CLI entry points to hermes-ui, drop the prepare hook so npm ci no longer triggers a build, and drop the removed tsconfig and nodemon references. Rewrite AGENTS.md, ARCHITECTURE.md, DEVELOPMENT.md, and the READMEs so the documented package map and hard rules match this repository. reference_code/ is ignored: it is a read-only snapshot, not an input to the build.
92 lines
2.7 KiB
TypeScript
92 lines
2.7 KiB
TypeScript
import { describe, expect, it, vi } from 'vitest'
|
|
import { countTokens } from '../../packages/server/src/lib/context-compressor'
|
|
import {
|
|
contextTokensWithCachedOverhead,
|
|
estimateUsageTokensFromMessages,
|
|
updateMessageContextTokenUsage,
|
|
} from '../../packages/server/src/services/hermes/run-chat/usage'
|
|
|
|
describe('run-chat usage token estimates', () => {
|
|
it('counts message content instead of serialized message payloads', () => {
|
|
const messages = [
|
|
{ role: 'user', content: 'hello from user' },
|
|
{ role: 'assistant', content: 'hello from assistant' },
|
|
]
|
|
|
|
const usage = estimateUsageTokensFromMessages(messages)
|
|
|
|
expect(usage.inputTokens).toBe(countTokens('hello from user'))
|
|
expect(usage.outputTokens).toBe(countTokens('hello from assistant'))
|
|
expect(usage.inputTokens + usage.outputTokens).toBeLessThan(countTokens(JSON.stringify(messages)))
|
|
})
|
|
|
|
it('keeps assistant tool call tokens on the output side', () => {
|
|
const messages = [
|
|
{
|
|
role: 'assistant',
|
|
content: 'calling tool',
|
|
tool_calls: [{ id: 'call_1', type: 'function', function: { name: 'lookup', arguments: '{"q":"x"}' } }],
|
|
},
|
|
]
|
|
|
|
const usage = estimateUsageTokensFromMessages(messages)
|
|
|
|
expect(usage.inputTokens).toBe(0)
|
|
expect(usage.outputTokens).toBe(countTokens('calling tool') + countTokens(String(messages[0].tool_calls || '')))
|
|
})
|
|
|
|
it('adds cached bridge fixed context when updating full context usage', () => {
|
|
const emit = vi.fn()
|
|
const state = {
|
|
messages: [],
|
|
isWorking: false,
|
|
events: [],
|
|
queue: [],
|
|
bridgeContext: { fixedContextTokens: 20_000 },
|
|
} as any
|
|
|
|
const contextTokens = updateMessageContextTokenUsage(
|
|
'session-1',
|
|
state,
|
|
emit,
|
|
1_569,
|
|
{ inputTokens: 1_200, outputTokens: 369 },
|
|
)
|
|
|
|
expect(contextTokens).toBe(21_569)
|
|
expect(state.contextTokens).toBe(21_569)
|
|
expect(emit).toHaveBeenCalledWith('usage.updated', expect.objectContaining({
|
|
session_id: 'session-1',
|
|
inputTokens: 1_200,
|
|
outputTokens: 369,
|
|
contextTokens: 21_569,
|
|
}))
|
|
})
|
|
|
|
it('falls back to message tokens when bridge fixed context is missing', () => {
|
|
const emit = vi.fn()
|
|
const state = {
|
|
messages: [],
|
|
isWorking: false,
|
|
events: [],
|
|
queue: [],
|
|
} as any
|
|
|
|
expect(contextTokensWithCachedOverhead(state, 1_569)).toBe(1_569)
|
|
|
|
const contextTokens = updateMessageContextTokenUsage(
|
|
'session-1',
|
|
state,
|
|
emit,
|
|
1_569,
|
|
{ inputTokens: 1_200, outputTokens: 369 },
|
|
)
|
|
|
|
expect(contextTokens).toBe(1_569)
|
|
expect(state.contextTokens).toBe(1_569)
|
|
expect(emit).toHaveBeenCalledWith('usage.updated', expect.objectContaining({
|
|
contextTokens: 1_569,
|
|
}))
|
|
})
|
|
})
|