Merge branch 'enricoros:v2-dev' into v2-dev

Dockerfile

@@ -2,7 +2,6 @@
 FROM node:22-alpine AS base
 ENV NEXT_TELEMETRY_DISABLED 1
 
-
 # Dependencies
 FROM base AS deps
 WORKDIR /app
@@ -11,6 +10,9 @@ WORKDIR /app
 COPY package*.json ./
 COPY src/server/prisma ./src/server/prisma
 
+# link ssl3 for latest Alpine
+RUN sh -c '[ ! -e /lib/libssl.so.3 ] && ln -s /usr/lib/libssl.so.3 /lib/libssl.so.3 || echo "Link already exists"'
+
 # Install dependencies, including dev (release builds should use npm ci)
 ENV NODE_ENV development
 RUN npm ci

src/apps/chat/components/layout-bar/useLLMDropdown.tsx

@@ -14,6 +14,7 @@ import { GoodTooltip } from '~/common/components/GoodTooltip';
 import { KeyStroke } from '~/common/components/KeyStroke';
 import { OptimaBarControlMethods, OptimaBarDropdownMemo, OptimaDropdownItems } from '~/common/layout/optima/bar/OptimaBarDropdown';
 import { findModelsServiceOrNull, llmsStoreActions, useModelsStore } from '~/common/stores/llms/store-llms';
+import { isDeepEqual } from '~/common/util/hooks/useDeep';
 import { optimaActions, optimaOpenModels } from '~/common/layout/optima/useOptima';
 
 
@@ -43,6 +44,9 @@ function LLMDropdown(props: {
   }, [chatLlmId]);
 
 
+  // dropdown items - chached
+  const stabilizeLlmOptions = React.useRef<OptimaDropdownItems>();
+
   const llmDropdownItems: OptimaDropdownItems = React.useMemo(() => {
     const llmItems: OptimaDropdownItems = {};
     let prevServiceId: DModelsServiceId | null = null;
@@ -69,7 +73,8 @@ function LLMDropdown(props: {
         llmItems[`sep-${llm.sId}`] = {
           type: 'separator',
           title: serviceLabel,
-          icon: vendor?.Icon ? <vendor.Icon /> : undefined,
+          // NOTE: commenting because not useful, and creates a recursive issue in isDeepEqual - not needed, so kthxbye
+          // icon: vendor?.Icon ? <vendor.Icon /> : undefined,
         };
         prevServiceId = llm.sId;
         sepCount++;
@@ -81,6 +86,7 @@ function LLMDropdown(props: {
         // icon: llm.id.startsWith('some vendor') ? <VendorIcon /> : undefined,
       };
     }
+
     // if there's a single separator (i.e. only one source), remove it
     if (sepCount === 1) {
       for (const key in llmItems) {
@@ -90,7 +96,13 @@ function LLMDropdown(props: {
         }
       }
     }
-    return llmItems;
+
+    // stabilize the items: reuse the full array if nothing changed
+    const prev = stabilizeLlmOptions.current;
+    if (prev && isDeepEqual(prev, llmItems)) return prev;
+
+    // otherwise update the cache and return the new items
+    return stabilizeLlmOptions.current = llmItems;
   }, [chatLlmId, llms, filterString]);
 
 

src/common/stores/llms/llms.parameters.ts

@@ -63,7 +63,7 @@ export const DModelParameterRegistry = {
     incompatibleWith: ['temperature'] as const,
   } as const,
 
-  'vnd.oai.reasoning_effort': {
+  llmVndOaiReasoningEffort: {
     label: 'Reasoning Effort',
     type: 'enum' as const,
     description: 'Constrains effort on reasoning for OpenAI reasoning models',
@@ -87,7 +87,7 @@ export type DModelParameterValues = {
   [K in DModelParameterId]?: DModelParameterValue<K>;
 }
 
-export type DModelParameterId = keyof typeof DModelParameterRegistry; // max_tokens, temperature, top_p, vnd.oai.reasoning_effort, ...
+export type DModelParameterId = keyof typeof DModelParameterRegistry;
 // type _ExtendedParameterId = keyof typeof _ExtendedParameterRegistry;
 
 type _EnumValues<T> = T extends { type: 'enum', values: readonly (infer U)[] } ? U : never;

src/common/stores/llms/store-llms.ts

@@ -8,8 +8,9 @@ import { persist } from 'zustand/middleware';
 import type { DOpenRouterServiceSettings } from '~/modules/llms/vendors/openrouter/openrouter.vendor';
 import type { ModelVendorId } from '~/modules/llms/vendors/vendors.registry';
 
-import { DLLM, DLLMId, LLM_IF_OAI_Fn, LLM_IF_OAI_Vision } from './llms.types';
+import type { DModelParameterId } from './llms.parameters';
 import type { DModelsService, DModelsServiceId } from './modelsservice.types';
+import { DLLM, DLLMId, LLM_IF_OAI_Fn, LLM_IF_OAI_Vision } from './llms.types';
 import { getLlmCostForTokens, portModelPricingV2toV3 } from './llms.pricing';
 
 
@@ -33,6 +34,7 @@ interface LlmsActions {
   rerankLLMsByServices: (serviceIdOrder: DModelsServiceId[]) => void;
   updateLLM: (id: DLLMId, partial: Partial<DLLM>) => void;
   updateLLMUserParameters: (id: DLLMId, partial: Partial<DLLM['userParameters']>) => void;
+  deleteLLMUserParameter: (id: DLLMId, parameterId: DModelParameterId) => void;
 
   addService: (service: DModelsService) => void;
   removeService: (id: DModelsServiceId) => void;
@@ -133,14 +135,22 @@ export const useModelsStore = create<LlmsState & LlmsActions>()(persist(
       })),
 
     updateLLMUserParameters: (id: DLLMId, partialUserParameters: Partial<DLLM['userParameters']>) =>
-      set(state => ({
-        llms: state.llms.map((llm: DLLM): DLLM =>
+      set(({ llms }) => ({
+        llms: llms.map((llm: DLLM): DLLM =>
           llm.id === id
             ? { ...llm, userParameters: { ...llm.userParameters, ...partialUserParameters } }
             : llm,
         ),
       })),
 
+    deleteLLMUserParameter: (id: DLLMId, parameterId: DModelParameterId) =>
+      set(({ llms }) => ({
+        llms: llms.map((llm: DLLM): DLLM =>
+          llm.id === id && llm.userParameters
+            ? { ...llm, userParameters: Object.fromEntries(Object.entries(llm.userParameters).filter(([key]) => key !== parameterId)) }
+            : llm,
+        ),
+      })),
 
     addService: (service: DModelsService) =>
       set(state => {

src/modules/aix/client/aix.client.ts

@@ -29,23 +29,31 @@ export function aixCreateChatGenerateContext(name: AixAPI_Context_ChatGenerate['
 }
 
 export function aixCreateModelFromLLMOptions(
-  llmOptions: DModelParameterValues | undefined,
-  llmOptionsOverride: DModelParameterValues | undefined,
+  llmOptions: DModelParameterValues,
+  llmOptionsOverride: Omit<DModelParameterValues, 'llmRef'> | undefined,
   debugLlmId: string,
 ): AixAPI_Model {
-  // model params (llm)
-  let { llmRef, llmTemperature, llmResponseTokens } = llmOptions || {};
-  if (!llmRef || llmTemperature === undefined)
+
+  // destructure input with the overrides
+  const { llmRef, llmTemperature, llmResponseTokens, llmTopP, llmVndOaiReasoningEffort } = {
+    ...llmOptions,
+    ...llmOptionsOverride,
+  };
+
+  // llmRef is absolutely required
+  if (!llmRef)
     throw new Error(`AIX: Error in configuration for model ${debugLlmId} (missing ref, temperature): ${JSON.stringify(llmOptions)}`);
 
-  // model params overrides
-  if (llmOptionsOverride?.llmTemperature !== undefined) llmTemperature = llmOptionsOverride.llmTemperature;
-  if (llmOptionsOverride?.llmResponseTokens !== undefined) llmResponseTokens = llmOptionsOverride.llmResponseTokens;
+  // llmTemperature is highly recommended, so we display a note if it's missing
+  if (llmTemperature === undefined)
+    console.warn(`[DEV] AIX: Missing temperature for model ${debugLlmId}, using default.`);
 
   return {
     id: llmRef,
-    temperature: llmTemperature,
-    ...(llmResponseTokens ? { maxTokens: llmResponseTokens } : {}),
+    ...(llmTemperature !== undefined ? { temperature: llmTemperature } : {}),
+    ...(llmResponseTokens /* null: similar to undefined, will omit the value */ ? { maxTokens: llmResponseTokens } : {}),
+    ...(llmTopP !== undefined ? { topP: llmTopP } : {}),
+    ...(llmVndOaiReasoningEffort ? { vndOaiReasoningEffort: llmVndOaiReasoningEffort } : {}),
   };
 }
 
@@ -70,7 +78,7 @@ type StreamMessageStatus = {
 interface AixClientOptions {
   abortSignal: AbortSignal | 'NON_ABORTABLE'; // 'NON_ABORTABLE' is a special case for non-abortable operations
   throttleParallelThreads?: number; // 0: disable, 1: default throttle (12Hz), 2+ reduce frequency with the square root
-  llmOptionsOverride?: DModelParameterValues;
+  llmOptionsOverride?: Omit<DModelParameterValues, 'llmRef'>; // overrides for the LLM options
 }
 
 

src/modules/aix/server/api/aix.wiretypes.ts

@@ -376,6 +376,8 @@ export namespace AixWire_API {
     id: z.string(),
     temperature: z.number().min(0).max(2).optional(),
     maxTokens: z.number().min(1).optional(),
+    topP: z.number().min(0).max(1).optional(),
+    vndOaiReasoningEffort: z.enum(['low', 'medium', 'high']).optional(),
   });
 
   /// Context

src/modules/aix/server/dispatch/chatGenerate/adapters/anthropic.messageCreate.ts

@@ -88,6 +88,12 @@ export function aixToAnthropicMessageCreate(model: AixAPI_Model, chatGenerate: A
     // top_p: undefined,
   };
 
+  // Top-P instead of temperature
+  if (model.topP !== undefined) {
+    payload.top_p = model.topP;
+    delete payload.temperature
+  }
+
   // Preemptive error detection with server-side payload validation before sending it upstream
   const validated = AnthropicWire_API_Message_Create.Request_schema.safeParse(payload);
   if (!validated.success) {

src/modules/aix/server/dispatch/chatGenerate/adapters/gemini.generateContent.ts

@@ -51,6 +51,12 @@ export function aixToGeminiGenerateContent(model: AixAPI_Model, chatGenerate: Ai
     },
   };
 
+  // Top-P instead of temperature
+  if (model.topP !== undefined) {
+    delete payload.generationConfig!.temperature;
+    payload.generationConfig!.topP = model.topP;
+  }
+
   // Preemptive error detection with server-side payload validation before sending it upstream
   const validated = GeminiWire_API_Generate_Content.Request_schema.safeParse(payload);
   if (!validated.success) {

src/modules/aix/server/dispatch/chatGenerate/adapters/openai.chatCompletions.ts

@@ -79,6 +79,17 @@ export function aixToOpenAIChatCompletions(openAIDialect: OpenAIDialects, model:
     user: undefined,
   };
 
+  // Top-P instead of temperature
+  if (model.topP !== undefined) {
+    delete payload.temperature;
+    payload.top_p = model.topP;
+  }
+
+  // [OpenAI] Vendor-specific reasoning effort, for o1 models only as of 2024-12-24
+  if (model.vndOaiReasoningEffort) {
+    payload.reasoning_effort = model.vndOaiReasoningEffort;
+  }
+
   if (hotFixOpenAIo1Family)
     payload = _fixRequestForOpenAIO1_maxCompletionTokens(payload);