feat(router,inference,config): pin narration to a configured model_id

[router.narration] model_id selects which provider handles narration turns instead of generic capability routing — lets a small/fast model own narration while the main model keeps CHAT/STEERING. - NarrationSettings.modelId parsed from TOML, threaded via RouterConfig .narrationModelId into DefaultRouterFacade.narrate (3-arg route) - DefaultInferenceRouter: exact-id match against healthy providers, with a post-select health check; any miss logs WARN and falls back to capability routing (never fails the narration turn) - onUserInput unchanged — only narrate uses the pinned model (tested)
2026-06-10 20:53:28 +04:00
parent cc6b402a23
commit 883e23dec7
10 changed files with 192 additions and 1 deletions
@@ -117,4 +117,44 @@ class DefaultInferenceRouterTest {
        val result = router.route(stage, setOf(ModelCapability.General))
        assertSame(p2, result)
    }
+
+    // ── 3-arg route with modelId ───────────────────────────────────────────
+
+    @Test
+    fun `3-arg route selects provider by exact id match`(): Unit = runBlocking {
+        val p1 = provider("llama-cpp:fast-model", ModelCapability.General)
+        val p2 = provider("llama-cpp:big-model", ModelCapability.General)
+        val router = DefaultInferenceRouter(registryOf(p1, p2), firstStrategy())
+        val result = router.route(stage, setOf(ModelCapability.General), "llama-cpp:fast-model")
+        assertSame(p1, result)
+    }
+
+    @Test
+    fun `3-arg route falls back to capability routing when modelId matches no provider`(): Unit = runBlocking {
+        val p1 = provider("llama-cpp:only-one", ModelCapability.General)
+        val router = DefaultInferenceRouter(registryOf(p1), firstStrategy())
+        val result = router.route(stage, setOf(ModelCapability.General), "nonexistent-model")
+        assertSame(p1, result)
+    }
+
+    @Test
+    fun `3-arg route with null modelId delegates to 2-arg form`(): Unit = runBlocking {
+        val p = provider("llama-cpp:model", ModelCapability.General)
+        val router = DefaultInferenceRouter(registryOf(p), firstStrategy())
+        val result = router.route(stage, setOf(ModelCapability.General), null)
+        assertSame(p, result)
+    }
+
+    @Test
+    fun `3-arg route falls back to capability routing when matched provider is unhealthy`(): Unit = runBlocking {
+        val sick = MockInferenceProvider(
+            id = ProviderId("llama-cpp:sick"),
+            declaredCapabilities = setOf(CapabilityScore(ModelCapability.General, 1.0)),
+            health = ProviderHealth.Unavailable("down"),
+        )
+        val healthy = provider("llama-cpp:healthy", ModelCapability.General)
+        val router = DefaultInferenceRouter(registryOf(sick, healthy), firstStrategy())
+        val result = router.route(stage, setOf(ModelCapability.General), "llama-cpp:sick")
+        assertSame(healthy, result)
+    }
 }
@@ -1186,6 +1186,93 @@ class RouterFacadeTest {
        assertTrue(l3Entries.any { it.content.contains("[recalled memory]") })
    }

+    // --------------------------------------------------------------------------
+    // Narration model ID routing
+    // --------------------------------------------------------------------------
+
+    @Test
+    fun `narrate passes narrationModelId to inferenceRouter 3-arg route`(): Unit = runBlocking {
+        val capturedModelId = mutableListOf<String?>()
+        val mockInferenceRouter = object : InferenceRouter {
+            override suspend fun route(
+                stageId: StageId,
+                requiredCapabilities: Set<ModelCapability>,
+            ): InferenceProvider = mockProvider("narration text")
+
+            override suspend fun route(
+                stageId: StageId,
+                requiredCapabilities: Set<ModelCapability>,
+                modelId: String?,
+            ): InferenceProvider {
+                capturedModelId.add(modelId)
+                return mockProvider("narration text")
+            }
+        }
+        val facade = DefaultRouterFacade(
+            routerRepository = object : RouterRepository {
+                override suspend fun getRouterState(sessionId: SessionId): RouterState = RouterState()
+            },
+            routerContextBuilder = object : RouterContextBuilder {
+                override suspend fun build(state: RouterState, budget: TokenBudget): ContextPack = emptyContextPack()
+            },
+            inferenceRouter = mockInferenceRouter,
+            eventStore = mockEventStore(),
+            config = RouterConfig(
+                tokenBudget = TokenBudget(limit = 5000),
+                narrationModelId = "llama-cpp:phi-3-mini",
+            ),
+            embedder = NoopEmbedder(dimension = 8),
+            l3MemoryStore = InMemoryL3MemoryStore(),
+        )
+        facade.narrate(
+            sessionId = SessionId("test-session"),
+            trigger = com.correx.core.router.model.NarrationTrigger(
+                kind = "test",
+                instruction = "describe what happened",
+            ),
+        )
+        assertEquals(1, capturedModelId.size)
+        assertEquals("llama-cpp:phi-3-mini", capturedModelId[0])
+    }
+
+    @Test
+    fun `onUserInput does NOT pass modelId — uses 2-arg route`(): Unit = runBlocking {
+        var threeArgCallCount = 0
+        val mockInferenceRouter = object : InferenceRouter {
+            override suspend fun route(
+                stageId: StageId,
+                requiredCapabilities: Set<ModelCapability>,
+            ): InferenceProvider = mockProvider("response")
+
+            override suspend fun route(
+                stageId: StageId,
+                requiredCapabilities: Set<ModelCapability>,
+                modelId: String?,
+            ): InferenceProvider {
+                threeArgCallCount++
+                return mockProvider("response")
+            }
+        }
+        val facade = DefaultRouterFacade(
+            routerRepository = object : RouterRepository {
+                override suspend fun getRouterState(sessionId: SessionId): RouterState = RouterState()
+            },
+            routerContextBuilder = object : RouterContextBuilder {
+                override suspend fun build(state: RouterState, budget: TokenBudget): ContextPack = emptyContextPack()
+            },
+            inferenceRouter = mockInferenceRouter,
+            eventStore = mockEventStore(),
+            config = RouterConfig(
+                tokenBudget = TokenBudget(limit = 5000),
+                narrationModelId = "llama-cpp:phi-3-mini",
+            ),
+            embedder = NoopEmbedder(dimension = 8),
+            l3MemoryStore = InMemoryL3MemoryStore(),
+        )
+        facade.onUserInput(sessionId = SessionId("test-session"), input = "Hello!")
+        assertEquals(0, threeArgCallCount, "onUserInput must not call 3-arg route()")
+    }
+
    private fun emptyContextPack(): ContextPack = ContextPack(
        id = ContextPackId("empty"),
        sessionId = SessionId("unknown"),