From 4a24be886d4d76d06f355cb5164af36fb3f36b4d Mon Sep 17 00:00:00 2001 From: tin-berri Date: Mon, 31 Aug 2026 22:12:59 -0700 Subject: [PATCH] feat(ui): one classification frequency picker for complexity auto-routers (#39042) Classification timing and session affinity are the same operator question, so Advanced: Classification Method now carries a single "How often to classify" radio: every request, every new user message, or once per session. The session choice writes session_affinity and stays disabled on custom tier sets, where the backend rejects it. Advanced: Affinity keeps the deployment switch alone. The serializer always writes classification_mode, matching session_affinity on the line below it, so an explicitly stored every_request survives an untouched save instead of being dropped back to the backend default. --- .../src/autorouter_presets.json | 4 + .../add_model/ClassificationMethodConfig.tsx | 51 ++++++++++ .../add_model/ComplexityRouterConfig.test.tsx | 96 +++++++++++++++++-- .../add_model/ComplexityRouterConfig.tsx | 41 +++++--- .../add_model/add_auto_router_tab.test.tsx | 46 ++++++++- .../add_model/add_auto_router_tab.tsx | 1 + .../build_complexity_router_config.test.ts | 19 ++++ .../build_complexity_router_config.ts | 6 ++ ...d_updated_complexity_router_config.test.ts | 28 ++++++ .../edit_auto_router_modal.test.ts | 2 + .../edit_auto_router_modal.test.tsx | 85 ++++++++++++---- .../edit_auto_router_modal.tsx | 7 ++ .../src/lib/autorouter_presets.test.ts | 32 +++++++ .../src/lib/autorouter_presets.ts | 2 + 14 files changed, 380 insertions(+), 40 deletions(-) diff --git a/ui/litellm-dashboard/src/autorouter_presets.json b/ui/litellm-dashboard/src/autorouter_presets.json index 4cbb548a855..da35d6171dc 100644 --- a/ui/litellm-dashboard/src/autorouter_presets.json +++ b/ui/litellm-dashboard/src/autorouter_presets.json @@ -14,6 +14,7 @@ }, "classifier_type": "heuristic", "escalation_keywords": ["LITELLM ESCALATE"], + "classification_mode": "every_request", "session_affinity": false, "deployment_affinity": true } @@ -30,6 +31,7 @@ }, "classifier_type": "heuristic", "escalation_keywords": ["LITELLM ESCALATE"], + "classification_mode": "every_request", "session_affinity": false, "deployment_affinity": true } @@ -56,6 +58,7 @@ }, "classifier_context_window_size": 0, "escalation_keywords": ["LITELLM ESCALATE"], + "classification_mode": "every_request", "session_affinity": false, "deployment_affinity": true } @@ -72,6 +75,7 @@ }, "classifier_type": "heuristic", "escalation_keywords": ["LITELLM ESCALATE"], + "classification_mode": "every_request", "session_affinity": false, "deployment_affinity": true } diff --git a/ui/litellm-dashboard/src/components/add_model/ClassificationMethodConfig.tsx b/ui/litellm-dashboard/src/components/add_model/ClassificationMethodConfig.tsx index 84f61c95eca..96c93306611 100644 --- a/ui/litellm-dashboard/src/components/add_model/ClassificationMethodConfig.tsx +++ b/ui/litellm-dashboard/src/components/add_model/ClassificationMethodConfig.tsx @@ -15,9 +15,12 @@ import { RestrictedSection, restrictedBy } from "./TierRestrictions"; import HeuristicScoringConfig from "./HeuristicScoringConfig"; import { useComplexityScorerDefaults } from "@/app/(dashboard)/hooks/autoRouter/useComplexityScorerDefaults"; import { + ClassificationFrequency, ClassifierFallback, ClassifierType, ComplexityRouterConfigValue, + classificationFrequency, + withClassificationFrequency, DEFAULT_CLASSIFIER_CONTEXT_BUDGET_CHARS, MIN_QUOTED_CONTEXT_TURN_CHARS, DEFAULT_CLASSIFIER_CONTEXT_WINDOW_SIZE, @@ -211,6 +214,7 @@ const ClassificationMethodConfig: React.FC = ({ const [draft, setDraft] = React.useState<{ id: string; raw: string } | null>(null); const hasDefaultModel = Boolean(defaultModel); const classifierType = effectiveClassifierType(value); + const sessionFrequencyRestriction = restrictedBy(value, "sessionAffinity"); const classifierModelMissing = showValidationErrors && usesLlmClassifier(classifierType) && !value.classifier_llm_config?.model; const usesCustomPrompt = Boolean(value.classifier_llm_config?.system_prompt?.trim()); @@ -305,6 +309,10 @@ const ClassificationMethodConfig: React.FC = ({ onChange({ ...value, classifier_fallback: fallback }); }; + const handleClassificationFrequencyChange = (frequency: ClassificationFrequency) => { + onChange(withClassificationFrequency(value, frequency)); + }; + const handleClassifierContextWindowSizeChange = (windowSize: number) => { onChange({ ...value, @@ -367,6 +375,49 @@ const ClassificationMethodConfig: React.FC = ({ )} +
+ How often to classify + + handleClassificationFrequencyChange(frequency as ClassificationFrequency) + } + > +
+ + + +
+ +

+ Holding the tier keeps an agent on one model for a whole tool loop and cuts scoring cost. A turn the router + cannot match to a held decision, such as one with no session id or an expired one, is scored again +

+
+ {usesLlmClassifier(classifierType) && (
diff --git a/ui/litellm-dashboard/src/components/add_model/ComplexityRouterConfig.test.tsx b/ui/litellm-dashboard/src/components/add_model/ComplexityRouterConfig.test.tsx index 218c99e32c5..bbee03135c7 100644 --- a/ui/litellm-dashboard/src/components/add_model/ComplexityRouterConfig.test.tsx +++ b/ui/litellm-dashboard/src/components/add_model/ComplexityRouterConfig.test.tsx @@ -598,6 +598,82 @@ describe("ComplexityRouterConfig classifier fallback", () => { }); }); +describe("ComplexityRouterConfig classification frequency", () => { + const llmValue: ComplexityRouterConfigValue = { + ...defaultValue, + classifier_type: "llm", + classifier_llm_config: { model: "gpt-3.5-turbo", timeout_ms: 3000 }, + }; + + it("defaults to every request, matching both backend field defaults", () => { + renderWithProviders(); + fireEvent.click(screen.getByText("Advanced: Classification Method")); + expect(screen.getByRole("radio", { name: /Every request/ })).toBeChecked(); + expect(screen.getByRole("radio", { name: /Every new user message/ })).not.toBeChecked(); + expect(screen.getByRole("radio", { name: /Once per session/ })).not.toBeChecked(); + }); + + it("writes both wire fields when the frequency moves to every new user message", () => { + const onChange = vi.fn(); + renderWithProviders(); + fireEvent.click(screen.getByText("Advanced: Classification Method")); + fireEvent.click(screen.getByRole("radio", { name: /Every new user message/ })); + expect(onChange).toHaveBeenCalledWith({ + ...llmValue, + classification_mode: "user_turn", + session_affinity: false, + }); + }); + + it("writes session affinity, not a classification mode, when the frequency moves to once per session", () => { + const onChange = vi.fn(); + renderWithProviders(); + fireEvent.click(screen.getByText("Advanced: Classification Method")); + fireEvent.click(screen.getByRole("radio", { name: /Once per session/ })); + expect(onChange).toHaveBeenCalledWith({ + ...llmValue, + classification_mode: "every_request", + session_affinity: true, + }); + }); + + it("shows a hand-authored config that sets both fields as once per session, matching the backend", () => { + renderWithProviders( + , + ); + fireEvent.click(screen.getByText("Advanced: Classification Method")); + expect(screen.getByRole("radio", { name: /Once per session/ })).toBeChecked(); + expect(screen.getByRole("radio", { name: /Every new user message/ })).not.toBeChecked(); + }); + + it("records a switch back to every request", () => { + const onChange = vi.fn(); + renderWithProviders( + , + ); + fireEvent.click(screen.getByText("Advanced: Classification Method")); + expect(screen.getByRole("radio", { name: /Every new user message/ })).toBeChecked(); + fireEvent.click(screen.getByRole("radio", { name: /Every request/ })); + expect(onChange).toHaveBeenCalledWith(expect.objectContaining({ classification_mode: "every_request" })); + }); + + it("offers the frequency on a heuristic router, where holding the tier still pins the model", () => { + // The backend pin is gated on the two fields alone, so a heuristic router that switches models + // mid tool loop is fixed by this control too. + renderWithProviders(); + fireEvent.click(screen.getByText("Advanced: Classification Method")); + expect(screen.getByRole("radio", { name: /Every new user message/ })).toBeInTheDocument(); + }); +}); + describe("ComplexityRouterConfig classifier rubric", () => { const llmValue: ComplexityRouterConfigValue = { ...defaultValue, @@ -761,12 +837,12 @@ describe("ComplexityRouterConfig tier labels", () => { }); describe("ComplexityRouterConfig affinity panel", () => { - it("holds both affinity switches with their backend defaults", () => { + it("holds the deployment switch at its backend default, session pinning having moved to the frequency choice", () => { renderWithProviders(); fireEvent.click(screen.getByText("Advanced: Affinity")); expect(screen.getByRole("switch", { name: "Pin a session to one deployment per model group" })).toBeChecked(); - expect(screen.getByRole("switch", { name: "Pin a session to its first model" })).not.toBeChecked(); + expect(screen.queryByRole("switch", { name: "Pin a session to its first model" })).not.toBeInTheDocument(); }); it("writes deployment_affinity through onChange without touching other keys", () => { @@ -1291,10 +1367,18 @@ describe("ComplexityRouterConfig tier editing", () => { expect(screen.getByLabelText("Fallback tier")).toBeInTheDocument(); }); - it("disables session pinning and says why, rather than letting a stripped value look saved", () => { - renderWithProviders(); - fireEvent.click(screen.getByText("Advanced: Affinity")); - expect(screen.getByLabelText("Pin a session to its first model")).toHaveAttribute("data-disabled"); + it("disables the once-per-session frequency and says why, rather than letting a stripped value look saved", () => { + renderWithProviders( + , + ); + fireEvent.click(screen.getByText("Advanced: Classification Method")); + const sessionOption = screen.getByRole("radio", { name: /Once per session/ }); + expect(sessionOption).toHaveAttribute("aria-disabled", "true"); + expect(sessionOption).not.toBeChecked(); expect( screen.getByText("Session pinning escalates along the built-in tier ladder", { exact: false }), ).toBeInTheDocument(); diff --git a/ui/litellm-dashboard/src/components/add_model/ComplexityRouterConfig.tsx b/ui/litellm-dashboard/src/components/add_model/ComplexityRouterConfig.tsx index 111a7c9f10a..5fde87fd638 100644 --- a/ui/litellm-dashboard/src/components/add_model/ComplexityRouterConfig.tsx +++ b/ui/litellm-dashboard/src/components/add_model/ComplexityRouterConfig.tsx @@ -56,6 +56,16 @@ export const MIN_QUOTED_CONTEXT_TURN_CHARS = 120; export const DEFAULT_SESSION_AFFINITY = false; export const DEFAULT_DEPLOYMENT_AFFINITY = true; +export type ClassificationMode = "every_request" | "user_turn"; + +export const DEFAULT_CLASSIFICATION_MODE: ClassificationMode = "every_request"; + +/** + * One operator-facing choice over the two wire fields that share the router's tier-pin machinery: + * session affinity pins every turn, user_turn pins every turn except a new human ask. + */ +export type ClassificationFrequency = ClassificationMode | "session"; + export type ComplexityTiers = { SIMPLE: string[]; MEDIUM: string[]; @@ -384,6 +394,7 @@ export interface ComplexityRouterConfigValue { classification_prompt?: string; /** Highest tier the scorer may decide alone under heuristic_first. Required by that type, rejected by the others. */ heuristic_first_max_tier?: string; + classification_mode?: ClassificationMode; session_affinity?: boolean; deployment_affinity?: boolean; /** Plan-mode floor as a tier ROW ID, unset meaning off. The wire carries the row's name. */ @@ -420,6 +431,21 @@ export interface ComplexityRouterConfigValue { tier_model_params?: TierModelParamsByTier; } +/** Session affinity wins where a hand-authored config sets both, matching the backend's own `or`. */ +export const classificationFrequency = (value: ComplexityRouterConfigValue): ClassificationFrequency => { + if (!value.custom_tier_set && (value.session_affinity ?? DEFAULT_SESSION_AFFINITY)) return "session"; + return value.classification_mode === "user_turn" ? "user_turn" : "every_request"; +}; + +export const withClassificationFrequency = ( + value: ComplexityRouterConfigValue, + frequency: ClassificationFrequency, +): ComplexityRouterConfigValue => ({ + ...value, + classification_mode: frequency === "user_turn" ? "user_turn" : "every_request", + session_affinity: frequency === "session", +}); + interface ComplexityRouterConfigProps { modelInfo: ModelGroup[]; value: ComplexityRouterConfigValue; @@ -498,23 +524,10 @@ const AffinityControls: React.FC<{ /> Pin a session to one deployment per model group
- + Keeps a session on the same deployment within a group, so provider prompt caches stay warm. Turn off to load-balance every turn. -
- onChange({ ...value, session_affinity: sessionAffinity })} - aria-label="Pin a session to its first model" - /> - Pin a session to its first model -
- - {restrictedBy(value, "sessionAffinity")?.reason ?? - "Keeps a session on its first turn's model instead of re-classifying each turn. Also pins the deployment."} - ); diff --git a/ui/litellm-dashboard/src/components/add_model/add_auto_router_tab.test.tsx b/ui/litellm-dashboard/src/components/add_model/add_auto_router_tab.test.tsx index 71b454dbb06..d8a955b719c 100644 --- a/ui/litellm-dashboard/src/components/add_model/add_auto_router_tab.test.tsx +++ b/ui/litellm-dashboard/src/components/add_model/add_auto_router_tab.test.tsx @@ -362,8 +362,8 @@ describe("AddAutoRouterTab", () => { await user.type(screen.getByPlaceholderText(/smart_router/i), "affinity-router"); expandDetailedConfiguration(); - await user.click(screen.getByText("Advanced: Affinity")); - expect(await screen.findByRole("switch", { name: "Pin a session to its first model" })).not.toBeChecked(); + await user.click(screen.getByText("Advanced: Classification Method")); + expect(await screen.findByRole("radio", { name: /Once per session/ })).not.toBeChecked(); await user.click(screen.getByRole("button", { name: /add auto router/i })); @@ -468,8 +468,8 @@ describe("AddAutoRouterTab", () => { await user.type(screen.getByPlaceholderText(/smart_router/i), "affinity-router"); expandDetailedConfiguration(); - await user.click(screen.getByText("Advanced: Affinity")); - await user.click(await screen.findByRole("switch", { name: "Pin a session to its first model" })); + await user.click(screen.getByText("Advanced: Classification Method")); + await user.click(await screen.findByRole("radio", { name: /Once per session/ })); await user.click(screen.getByRole("button", { name: /add auto router/i })); @@ -479,6 +479,44 @@ describe("AddAutoRouterTab", () => { }); }); + it("carries every new user message through to the create payload", async () => { + const user = userEvent.setup(); + vi.mocked(getMissingTiersError).mockReturnValue(null); + + renderWithProviders(); + + await user.type(screen.getByPlaceholderText(/smart_router/i), "user-turn-router"); + expandDetailedConfiguration(); + await user.click(screen.getByText("Advanced: Classification Method")); + await user.click(await screen.findByRole("radio", { name: /Every new user message/ })); + + await user.click(screen.getByRole("button", { name: /add auto router/i })); + + await waitFor(() => expect(handleAddAutoRouterSubmit).toHaveBeenCalled()); + expect(vi.mocked(handleAddAutoRouterSubmit).mock.calls.at(-1)?.[0].complexity_router_config).toMatchObject({ + classification_mode: "user_turn", + }); + }); + + it("writes every_request into the create payload when the default frequency stays selected", async () => { + const user = userEvent.setup(); + vi.mocked(getMissingTiersError).mockReturnValue(null); + + renderWithProviders(); + + await user.type(screen.getByPlaceholderText(/smart_router/i), "default-timing-router"); + expandDetailedConfiguration(); + await user.click(screen.getByText("Advanced: Classification Method")); + expect(await screen.findByRole("radio", { name: /Every request/ })).toBeChecked(); + + await user.click(screen.getByRole("button", { name: /add auto router/i })); + + await waitFor(() => expect(handleAddAutoRouterSubmit).toHaveBeenCalled()); + expect( + vi.mocked(handleAddAutoRouterSubmit).mock.calls.at(-1)?.[0].complexity_router_config.classification_mode, + ).toBe("every_request"); + }); + it("defaults a new router to deployment affinity on, matching the backend field default", async () => { const user = userEvent.setup(); vi.mocked(getMissingTiersError).mockReturnValue(null); diff --git a/ui/litellm-dashboard/src/components/add_model/add_auto_router_tab.tsx b/ui/litellm-dashboard/src/components/add_model/add_auto_router_tab.tsx index ed584a4882b..c60ce5e4959 100644 --- a/ui/litellm-dashboard/src/components/add_model/add_auto_router_tab.tsx +++ b/ui/litellm-dashboard/src/components/add_model/add_auto_router_tab.tsx @@ -342,6 +342,7 @@ const AddAutoRouterTab: React.FC = ({ planModeMinTier: complexityRouterConfig.plan_mode_min_tier, classificationPrompt: complexityRouterConfig.classification_prompt, heuristicFirstMaxTier: complexityRouterConfig.heuristic_first_max_tier, + classificationMode: complexityRouterConfig.classification_mode, tierLabels: complexityRouterConfig.tier_labels, classifierType: complexityRouterConfig.classifier_type, classifierLlmConfig: complexityRouterConfig.classifier_llm_config, diff --git a/ui/litellm-dashboard/src/components/add_model/build_complexity_router_config.test.ts b/ui/litellm-dashboard/src/components/add_model/build_complexity_router_config.test.ts index feddaaa0eac..40addf086b6 100644 --- a/ui/litellm-dashboard/src/components/add_model/build_complexity_router_config.test.ts +++ b/ui/litellm-dashboard/src/components/add_model/build_complexity_router_config.test.ts @@ -52,6 +52,7 @@ describe("buildComplexityRouterConfig", () => { const expected = { tiers, classifier_type: "heuristic", + classification_mode: "every_request", session_affinity: false, deployment_affinity: true, escalation_keywords: ["LITELLM ESCALATE"], @@ -790,6 +791,20 @@ describe("heuristic_first", () => { }); }); +describe("classification_mode", () => { + it("emits user_turn", () => { + const config = buildComplexityRouterConfig({ ...baseParams, classificationMode: "user_turn" }); + expect(config.classification_mode).toBe("user_turn"); + }); + + it("writes every_request explicitly, so a saved router never depends on the backend default", () => { + expect( + buildComplexityRouterConfig({ ...baseParams, classificationMode: "every_request" }).classification_mode, + ).toBe("every_request"); + expect(buildComplexityRouterConfig(baseParams).classification_mode).toBe("every_request"); + }); +}); + describe("buildComplexityRouterConfig with an edited tier set", () => { const customTierSet = { tiers: [ @@ -902,6 +917,10 @@ describe("buildComplexityRouterConfig with an edited tier set", () => { expect(payload.classifier_llm_config).toEqual({ model: "gpt-4o-mini", timeout_ms: 3000 }); }); + it("keeps classification_mode, which the backend accepts beside tier_definitions", () => { + expect(build({ classificationMode: "user_turn" }).classification_mode).toBe("user_turn"); + }); + it("carries the plan-mode floor as the row's name, not the row id the form holds", () => { expect(build({ planModeMinTier: "sec" }).plan_mode_min_tier).toBe("SECURITY_REVIEW"); }); diff --git a/ui/litellm-dashboard/src/components/add_model/build_complexity_router_config.ts b/ui/litellm-dashboard/src/components/add_model/build_complexity_router_config.ts index 9a14e956207..6a087649c00 100644 --- a/ui/litellm-dashboard/src/components/add_model/build_complexity_router_config.ts +++ b/ui/litellm-dashboard/src/components/add_model/build_complexity_router_config.ts @@ -19,10 +19,12 @@ import { import { AdaptiveEligible, AdaptiveRouterWeights, + ClassificationMode, ClassifierFallback, ClassifierLLMConfig, ClassifierType, ComplexityTierLabels, + DEFAULT_CLASSIFICATION_MODE, ComplexityRouterConfigValue, ComplexityTiers, DimensionWeights, @@ -105,6 +107,7 @@ export interface BuildComplexityRouterConfigParams { classifierFallback: ClassifierFallback | undefined; classificationPrompt: string | undefined; heuristicFirstMaxTier: string | undefined; + classificationMode: ClassificationMode | undefined; sessionAffinity: boolean; deploymentAffinity: boolean; customTechnicalKeywords: string[]; @@ -159,6 +162,7 @@ export interface ComplexityRouterConfigPayload { classifier_fallback?: ClassifierFallback; classification_prompt?: string; heuristic_first_max_tier?: string; + classification_mode: ClassificationMode; session_affinity: boolean; deployment_affinity: boolean; custom_technical_keywords?: string[]; @@ -393,6 +397,7 @@ export const buildComplexityRouterConfig = ({ classifierFallback, classificationPrompt, heuristicFirstMaxTier, + classificationMode, sessionAffinity, deploymentAffinity, customTechnicalKeywords, @@ -452,6 +457,7 @@ export const buildComplexityRouterConfig = ({ ...(cleanedTierLabels && { tier_labels: cleanedTierLabels }), classifier_type: classifierType, ...classifierWireFields(effectiveType, classifierInputs), + classification_mode: classificationMode ?? DEFAULT_CLASSIFICATION_MODE, session_affinity: sessionAffinity, deployment_affinity: deploymentAffinity, ...(customTechnicalKeywords.length > 0 && { custom_technical_keywords: customTechnicalKeywords }), diff --git a/ui/litellm-dashboard/src/components/edit_auto_router/build_updated_complexity_router_config.test.ts b/ui/litellm-dashboard/src/components/edit_auto_router/build_updated_complexity_router_config.test.ts index d8c2987ead5..3e63cbd3b32 100644 --- a/ui/litellm-dashboard/src/components/edit_auto_router/build_updated_complexity_router_config.test.ts +++ b/ui/litellm-dashboard/src/components/edit_auto_router/build_updated_complexity_router_config.test.ts @@ -257,6 +257,33 @@ describe("buildUpdatedComplexityRouterConfig session affinity", () => { }); }); +describe("buildUpdatedComplexityRouterConfig classification mode", () => { + it("round-trips a stored user_turn through hydrate then save", () => { + const stored = { ...STORED, classification_mode: "user_turn" }; + const hydrated = hydrateComplexityRouterConfig(stored, undefined); + + expect(hydrated.classification_mode).toBe("user_turn"); + expect(buildUpdatedComplexityRouterConfig(stored, hydrated).classification_mode).toBe("user_turn"); + }); + + it("round-trips an explicitly stored every_request, so an untouched save leaves it as written", () => { + const stored = { ...STORED, classification_mode: "every_request" }; + const hydrated = hydrateComplexityRouterConfig(stored, undefined); + + expect(hydrated.classification_mode).toBe("every_request"); + expect(buildUpdatedComplexityRouterConfig(stored, hydrated).classification_mode).toBe("every_request"); + }); + + it("rewrites a stored user_turn to every_request once the operator picks the default back", () => { + const stored = { ...STORED, classification_mode: "user_turn" }; + const result = buildUpdatedComplexityRouterConfig(stored, { + ...FORM_VALUE, + classification_mode: "every_request", + }); + expect(result.classification_mode).toBe("every_request"); + }); +}); + describe("buildUpdatedComplexityRouterConfig deployment affinity", () => { it("writes deployment_affinity=false when the toggle is off", () => { const result = buildUpdatedComplexityRouterConfig(STORED, { ...FORM_VALUE, deployment_affinity: false }); @@ -476,6 +503,7 @@ describe("managed keys survive an untouched open-and-save", () => { classifier_context_budget_chars: 4000, classifier_context_include_assistant_turns: true, classifier_fallback: "default_model", + classification_mode: "user_turn", session_affinity: true, deployment_affinity: false, adaptive: true, diff --git a/ui/litellm-dashboard/src/components/edit_auto_router/edit_auto_router_modal.test.ts b/ui/litellm-dashboard/src/components/edit_auto_router/edit_auto_router_modal.test.ts index 027e01a9351..d199af51b41 100644 --- a/ui/litellm-dashboard/src/components/edit_auto_router/edit_auto_router_modal.test.ts +++ b/ui/litellm-dashboard/src/components/edit_auto_router/edit_auto_router_modal.test.ts @@ -47,6 +47,7 @@ const expectedClassifiedTierConfig = { semantic_keyword_matching: true, embedding_model: "voyage-4-large", match_threshold: 0.65, + classification_mode: "every_request", session_affinity: false, deployment_affinity: true, adaptive: true, @@ -68,6 +69,7 @@ const expectedAdaptiveDisabledConfig = { semantic_keyword_matching: true, embedding_model: "voyage-4-large", match_threshold: 0.65, + classification_mode: "every_request", session_affinity: false, deployment_affinity: true, }; diff --git a/ui/litellm-dashboard/src/components/edit_auto_router/edit_auto_router_modal.test.tsx b/ui/litellm-dashboard/src/components/edit_auto_router/edit_auto_router_modal.test.tsx index a4921fcfcb5..96e8549eac8 100644 --- a/ui/litellm-dashboard/src/components/edit_auto_router/edit_auto_router_modal.test.tsx +++ b/ui/litellm-dashboard/src/components/edit_auto_router/edit_auto_router_modal.test.tsx @@ -364,7 +364,7 @@ describe("EditAutoRouterModal assistant turns", () => { }); }); -describe("EditAutoRouterModal session affinity", () => { +describe("EditAutoRouterModal classification frequency", () => { beforeEach(() => { modelPatchUpdateCall.mockClear(); }); @@ -381,15 +381,15 @@ describe("EditAutoRouterModal session affinity", () => { />, ); - // A stored config with no session_affinity key now runs with affinity OFF, because the backend - // field defaults to False. The toggle has to render what the router actually does, and an - // untouched save must not flip it. - it("shows a stored config with no session_affinity key as off", async () => { + // A stored config with neither key now runs with affinity OFF, because both backend fields + // default that way. The picker has to render what the router actually does, and an untouched + // save must not flip it. + it("shows a stored config with neither key as every request", async () => { const user = userEvent.setup(); renderWithStoredConfig(STORED_CONFIG); - await user.click(await screen.findByText("Advanced: Affinity")); - expect(await screen.findByRole("switch", { name: "Pin a session to its first model" })).not.toBeChecked(); + await user.click(await screen.findByText("Advanced: Classification Method")); + expect(await screen.findByRole("radio", { name: /Every request/ })).toBeChecked(); await user.click(screen.getByRole("button", { name: /save changes/i })); @@ -397,12 +397,12 @@ describe("EditAutoRouterModal session affinity", () => { expect(savedConfig().session_affinity).toBe(false); }); - it("shows a stored session_affinity=true as on and preserves it through an untouched save", async () => { + it("shows a stored session_affinity=true as once per session and preserves it through an untouched save", async () => { const user = userEvent.setup(); renderWithStoredConfig({ ...STORED_CONFIG, session_affinity: true }); - await user.click(await screen.findByText("Advanced: Affinity")); - expect(await screen.findByRole("switch", { name: "Pin a session to its first model" })).toBeChecked(); + await user.click(await screen.findByText("Advanced: Classification Method")); + expect(await screen.findByRole("radio", { name: /Once per session/ })).toBeChecked(); await user.click(screen.getByRole("button", { name: /save changes/i })); @@ -410,12 +410,12 @@ describe("EditAutoRouterModal session affinity", () => { expect(savedConfig().session_affinity).toBe(true); }); - it("persists turning session affinity on", async () => { + it("persists picking once per session", async () => { const user = userEvent.setup(); renderWithStoredConfig(STORED_CONFIG); - await user.click(await screen.findByText("Advanced: Affinity")); - await user.click(await screen.findByRole("switch", { name: "Pin a session to its first model" })); + await user.click(await screen.findByText("Advanced: Classification Method")); + await user.click(await screen.findByRole("radio", { name: /Once per session/ })); await user.click(screen.getByRole("button", { name: /save changes/i })); @@ -423,18 +423,71 @@ describe("EditAutoRouterModal session affinity", () => { expect(savedConfig().session_affinity).toBe(true); }); - it("persists turning session affinity back off", async () => { + it("persists picking every request back over a stored session pin", async () => { const user = userEvent.setup(); renderWithStoredConfig({ ...STORED_CONFIG, session_affinity: true }); - await user.click(await screen.findByText("Advanced: Affinity")); - await user.click(await screen.findByRole("switch", { name: "Pin a session to its first model" })); + await user.click(await screen.findByText("Advanced: Classification Method")); + await user.click(await screen.findByRole("radio", { name: /Every request/ })); await user.click(screen.getByRole("button", { name: /save changes/i })); await waitFor(() => expect(modelPatchUpdateCall).toHaveBeenCalled()); expect(savedConfig().session_affinity).toBe(false); }); + + it("clears a stored session pin when the operator moves to every new user message", async () => { + const user = userEvent.setup(); + renderWithStoredConfig({ ...STORED_CONFIG, session_affinity: true }); + + await user.click(await screen.findByText("Advanced: Classification Method")); + await user.click(await screen.findByRole("radio", { name: /Every new user message/ })); + + await user.click(screen.getByRole("button", { name: /save changes/i })); + + await waitFor(() => expect(modelPatchUpdateCall).toHaveBeenCalled()); + expect(savedConfig().session_affinity).toBe(false); + expect(savedConfig().classification_mode).toBe("user_turn"); + }); + + it("shows a stored user_turn as selected and preserves it through an untouched save", async () => { + const user = userEvent.setup(); + renderWithStoredConfig({ ...STORED_CONFIG, classification_mode: "user_turn" }); + + await user.click(await screen.findByText("Advanced: Classification Method")); + expect(await screen.findByRole("radio", { name: /Every new user message/ })).toBeChecked(); + + await user.click(screen.getByRole("button", { name: /save changes/i })); + + await waitFor(() => expect(modelPatchUpdateCall).toHaveBeenCalled()); + expect(savedConfig().classification_mode).toBe("user_turn"); + }); + + it("persists switching a stored config to every new user message", async () => { + const user = userEvent.setup(); + renderWithStoredConfig(STORED_CONFIG); + + await user.click(await screen.findByText("Advanced: Classification Method")); + await user.click(await screen.findByRole("radio", { name: /Every new user message/ })); + + await user.click(screen.getByRole("button", { name: /save changes/i })); + + await waitFor(() => expect(modelPatchUpdateCall).toHaveBeenCalled()); + expect(savedConfig().classification_mode).toBe("user_turn"); + }); + + it("rewrites the stored mode to every_request when the operator picks it back", async () => { + const user = userEvent.setup(); + renderWithStoredConfig({ ...STORED_CONFIG, classification_mode: "user_turn" }); + + await user.click(await screen.findByText("Advanced: Classification Method")); + await user.click(await screen.findByRole("radio", { name: /Every request/ })); + + await user.click(screen.getByRole("button", { name: /save changes/i })); + + await waitFor(() => expect(modelPatchUpdateCall).toHaveBeenCalled()); + expect(savedConfig().classification_mode).toBe("every_request"); + }); }); describe("EditAutoRouterModal deployment affinity", () => { diff --git a/ui/litellm-dashboard/src/components/edit_auto_router/edit_auto_router_modal.tsx b/ui/litellm-dashboard/src/components/edit_auto_router/edit_auto_router_modal.tsx index 4751c2e64b7..2aecf0b3483 100644 --- a/ui/litellm-dashboard/src/components/edit_auto_router/edit_auto_router_modal.tsx +++ b/ui/litellm-dashboard/src/components/edit_auto_router/edit_auto_router_modal.tsx @@ -96,6 +96,7 @@ export interface StoredComplexityRouterConfig { classifier_context_budget_chars?: unknown; classifier_context_include_assistant_turns?: unknown; classifier_fallback?: unknown; + classification_mode?: unknown; tier_boundaries?: unknown; token_thresholds?: unknown; dimension_weights?: unknown; @@ -165,6 +166,10 @@ export const hydrateComplexityRouterConfig = ( typeof parsedConfig.heuristic_first_max_tier === "string" && parsedConfig.heuristic_first_max_tier.trim() !== "" ? parsedConfig.heuristic_first_max_tier : undefined, + classification_mode: + parsedConfig.classification_mode === "user_turn" || parsedConfig.classification_mode === "every_request" + ? parsedConfig.classification_mode + : undefined, tier_boundaries: hydrateTierBoundaries(parsedConfig.tier_boundaries), token_thresholds: hydrateTokenThresholds(parsedConfig.token_thresholds), dimension_weights: hydrateDimensionWeights(parsedConfig.dimension_weights), @@ -207,6 +212,7 @@ export const MANAGED_COMPLEXITY_ROUTER_KEYS = new Set([ "classifier_fallback", "classification_prompt", "heuristic_first_max_tier", + "classification_mode", "session_affinity", "deployment_affinity", "adaptive", @@ -294,6 +300,7 @@ export const buildUpdatedComplexityRouterConfig = ( planModeMinTier: value.plan_mode_min_tier, classificationPrompt: value.classification_prompt, heuristicFirstMaxTier: value.heuristic_first_max_tier, + classificationMode: value.classification_mode, tierLabels: value.tier_labels, classifierType: value.classifier_type, classifierLlmConfig: value.classifier_llm_config, diff --git a/ui/litellm-dashboard/src/lib/autorouter_presets.test.ts b/ui/litellm-dashboard/src/lib/autorouter_presets.test.ts index f14d6279e32..5632d6d947f 100644 --- a/ui/litellm-dashboard/src/lib/autorouter_presets.test.ts +++ b/ui/litellm-dashboard/src/lib/autorouter_presets.test.ts @@ -230,6 +230,7 @@ describe("autorouter_presets", () => { const config = { tiers: { SIMPLE: [presetModel], MEDIUM: [], COMPLEX: [], REASONING: [] }, classifier_type: "heuristic" as const, + classification_mode: "every_request" as const, session_affinity: false, deployment_affinity: true, }; @@ -273,6 +274,7 @@ describe("autorouter_presets", () => { const config = { tiers: { SIMPLE: ["claude-opus-5"], MEDIUM: [], COMPLEX: [], REASONING: [] }, classifier_type: "heuristic" as const, + classification_mode: "every_request" as const, session_affinity: false, deployment_affinity: true, }; @@ -287,6 +289,7 @@ describe("autorouter_presets", () => { const config = { tiers: { SIMPLE: ["claude-opus-5"], MEDIUM: [], COMPLEX: [], REASONING: [] }, classifier_type: "heuristic" as const, + classification_mode: "every_request" as const, session_affinity: false, deployment_affinity: true, }; @@ -320,6 +323,7 @@ describe("autorouter_presets", () => { const simpleTierConfig = (presetModel: string) => ({ tiers: { SIMPLE: [presetModel], MEDIUM: [], COMPLEX: [], REASONING: [] }, classifier_type: "heuristic" as const, + classification_mode: "every_request" as const, session_affinity: false, deployment_affinity: true, }); @@ -563,6 +567,7 @@ describe("autorouter_presets", () => { const config = { tiers: { SIMPLE: ["gpt-5-nano"], MEDIUM: [], COMPLEX: [], REASONING: [] }, classifier_type: "heuristic" as const, + classification_mode: "every_request" as const, session_affinity: false, deployment_affinity: true, match_threshold: 0, @@ -578,6 +583,7 @@ describe("autorouter_presets", () => { { tiers: { SIMPLE: ["gpt-5-nano"], MEDIUM: [], COMPLEX: [], REASONING: [] }, classifier_type: "heuristic", + classification_mode: "every_request", session_affinity: false, deployment_affinity: true, enable_context_window_escalation: false, @@ -589,11 +595,29 @@ describe("autorouter_presets", () => { expect(prefill.complexityRouterConfig.context_window_escalation_buffer).toBe(0.9); }); + it("carries a preset's classification_mode and defaults it when the preset omits one", () => { + const tiers = { SIMPLE: ["gpt-5-nano"], MEDIUM: [], COMPLEX: [], REASONING: [] }; + const base = { + tiers, + classifier_type: "heuristic" as const, + classification_mode: "every_request" as const, + session_affinity: false, + deployment_affinity: true, + }; + const availability = groupsOnly(["gpt-5-nano"]); + expect( + buildPresetPrefill({ ...base, classification_mode: "user_turn" }, availability).complexityRouterConfig + .classification_mode, + ).toBe("user_turn"); + expect(buildPresetPrefill(base, availability).complexityRouterConfig.classification_mode).toBe("every_request"); + }); + it("falls back to the defaults when a preset omits match_threshold and escalation_keywords", () => { const prefill = buildPresetPrefill( { tiers: { SIMPLE: ["gpt-5-nano"], MEDIUM: [], COMPLEX: [], REASONING: [] }, classifier_type: "heuristic", + classification_mode: "every_request" as const, session_affinity: false, deployment_affinity: true, }, @@ -610,6 +634,7 @@ describe("autorouter_presets", () => { const base = { tiers: { SIMPLE: ["gpt-5-nano"], MEDIUM: [], COMPLEX: [], REASONING: [] }, classifier_type: "heuristic" as const, + classification_mode: "every_request" as const, session_affinity: false, deployment_affinity: true, }; @@ -625,6 +650,7 @@ describe("autorouter_presets", () => { const config = { tiers: { SIMPLE: ["claude-sonnet-4-5"], MEDIUM: [], COMPLEX: [], REASONING: [] }, classifier_type: "heuristic" as const, + classification_mode: "every_request" as const, session_affinity: false, deployment_affinity: true, }; @@ -639,6 +665,7 @@ describe("autorouter_presets", () => { REASONING: [{ model_name: "o3", litellm_params: { reasoning_effort: "high" } }], }, classifier_type: "heuristic" as const, + classification_mode: "every_request" as const, session_affinity: false, deployment_affinity: true, }; @@ -658,6 +685,7 @@ describe("autorouter_presets", () => { REASONING: [{ model_name: "claude-sonnet-4-5", litellm_params: { reasoning_effort: "high" } }], }, classifier_type: "heuristic" as const, + classification_mode: "every_request" as const, session_affinity: false, deployment_affinity: true, }; @@ -680,6 +708,9 @@ describe("autorouter_presets", () => { ], }, classifier_type: "heuristic" as const, + classification_mode: "every_request" as const, + session_affinity: false, + deployment_affinity: true, }; const prefill = buildPresetPrefill(config, groupsOnly(["claude-sonnet-4.5"])); // temperature survives from the spelling that would otherwise have been overwritten; @@ -693,6 +724,7 @@ describe("autorouter_presets", () => { const config = { tiers: { SIMPLE: ["gpt-5-nano"], MEDIUM: [], COMPLEX: [], REASONING: [] }, classifier_type: "heuristic" as const, + classification_mode: "every_request" as const, session_affinity: false, deployment_affinity: true, }; diff --git a/ui/litellm-dashboard/src/lib/autorouter_presets.ts b/ui/litellm-dashboard/src/lib/autorouter_presets.ts index f96dd5ddb4c..4c1f08b62e9 100644 --- a/ui/litellm-dashboard/src/lib/autorouter_presets.ts +++ b/ui/litellm-dashboard/src/lib/autorouter_presets.ts @@ -6,6 +6,7 @@ import { ComplexityRouterConfigValue, ClassifierType, ClassifierLLMConfig, + DEFAULT_CLASSIFICATION_MODE, DEFAULT_SESSION_AFFINITY, DEFAULT_DEPLOYMENT_AFFINITY, usesLlmClassifier, @@ -288,6 +289,7 @@ export const buildPresetPrefill = ( classifier_context_budget_chars: config.classifier_context_budget_chars, classifier_context_per_turn_chars: config.classifier_context_per_turn_chars, classifier_context_include_assistant_turns: config.classifier_context_include_assistant_turns, + classification_mode: config.classification_mode ?? DEFAULT_CLASSIFICATION_MODE, session_affinity: config.session_affinity ?? DEFAULT_SESSION_AFFINITY, deployment_affinity: config.deployment_affinity ?? DEFAULT_DEPLOYMENT_AFFINITY, adaptive: config.adaptive,