fix(core): clamp compaction output budget (#36745)
Co-authored-by: Dax Raad <thdxr@users.noreply.github.com>
This commit is contained in:
parent
ecb5754f4c
commit
fbe7f26e71
2 changed files with 26 additions and 1 deletions
|
|
@ -139,6 +139,11 @@ const compactModel = Model.make({
|
|||
provider: "fake",
|
||||
route: OpenAIChat.route.with({ limits: { context: 4_000, output: 50 } }),
|
||||
})
|
||||
const fullOutputModel = Model.make({
|
||||
id: "full-output",
|
||||
provider: "fake",
|
||||
route: OpenAIChat.route.with({ limits: { context: 262_144, output: 262_144 } }),
|
||||
})
|
||||
const undersizedContextModel = Model.make({
|
||||
id: "undersized-context",
|
||||
provider: "fake",
|
||||
|
|
@ -1865,6 +1870,25 @@ describe("SessionRunnerLLM", () => {
|
|||
}),
|
||||
)
|
||||
|
||||
it.effect("does not compact immediately when the advertised output limit fills the context", () =>
|
||||
Effect.gen(function* () {
|
||||
const session = yield* setup
|
||||
currentModel = fullOutputModel
|
||||
response = reply.textWithUsage("Earlier answer", "text-full-output-first", 9_500)
|
||||
yield* admit(session, "Earlier question")
|
||||
yield* session.resume(sessionID)
|
||||
|
||||
requests.length = 0
|
||||
response = reply.text("Continued", "text-full-output-final")
|
||||
yield* admit(session, "Continue")
|
||||
yield* session.resume(sessionID)
|
||||
|
||||
expect(requests).toHaveLength(1)
|
||||
expect(userTexts(requests[0])).toContain("Continue")
|
||||
expect(yield* session.context(sessionID)).not.toContainEqual(expect.objectContaining({ type: "compaction" }))
|
||||
}),
|
||||
)
|
||||
|
||||
it.effect("stops after required automatic compaction fails", () =>
|
||||
Effect.gen(function* () {
|
||||
const session = yield* setup
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue