[8.x] [inference] add support for `temperature` parameter (#206479) (#206573)

# Backport

This will backport the following commits from `main` to `8.x`:
- [[inference] add support for `temperature` parameter
(#206479)](https://github.com/elastic/kibana/pull/206479)

<!--- Backport version: 9.4.3 -->

### Questions ?
Please refer to the [Backport tool
documentation](https://github.com/sqren/backport)

<!--BACKPORT [{"author":{"name":"Pierre
Gayvallet","email":"pierre.gayvallet@elastic.co"},"sourceCommit":{"committedDate":"2025-01-14T13:05:21Z","message":"[inference]
add support for `temperature` parameter (#206479)\n\n##
Summary\r\n\r\nFix
https://github.com/elastic/kibana/issues/206542\r\n\r\nAdd a
`temperature` parameter to the `chatComplete` inference API, and\r\nwire
it accordingly on all
adapters.","sha":"81b47fc094b3a1a404fad23de7c65970a121a519","branchLabelMapping":{"^v9.0.0$":"main","^v8.18.0$":"8.x","^v(\\d+).(\\d+).\\d+$":"$1.$2"}},"sourcePullRequest":{"labels":["release_note:skip","v9.0.0","backport:version","Team:AI
Infra","v8.18.0"],"title":"[inference] add support for `temperature`
parameter","number":206479,"url":"https://github.com/elastic/kibana/pull/206479","mergeCommit":{"message":"[inference]
add support for `temperature` parameter (#206479)\n\n##
Summary\r\n\r\nFix
https://github.com/elastic/kibana/issues/206542\r\n\r\nAdd a
`temperature` parameter to the `chatComplete` inference API, and\r\nwire
it accordingly on all
adapters.","sha":"81b47fc094b3a1a404fad23de7c65970a121a519"}},"sourceBranch":"main","suggestedTargetBranches":["8.x"],"targetPullRequestStates":[{"branch":"main","label":"v9.0.0","branchLabelMappingKey":"^v9.0.0$","isSourceBranch":true,"state":"MERGED","url":"https://github.com/elastic/kibana/pull/206479","number":206479,"mergeCommit":{"message":"[inference]
add support for `temperature` parameter (#206479)\n\n##
Summary\r\n\r\nFix
https://github.com/elastic/kibana/issues/206542\r\n\r\nAdd a
`temperature` parameter to the `chatComplete` inference API, and\r\nwire
it accordingly on all
adapters.","sha":"81b47fc094b3a1a404fad23de7c65970a121a519"}},{"branch":"8.x","label":"v8.18.0","branchLabelMappingKey":"^v8.18.0$","isSourceBranch":false,"state":"NOT_CREATED"}]}]
BACKPORT-->

Co-authored-by: Pierre Gayvallet <pierre.gayvallet@elastic.co>

x-pack/platform/packages/shared/ai-infra/inference-common/src/chat_complete/api.ts

@@ -89,6 +89,11 @@ export type ChatCompleteOptions<
    * The list of messages for the current conversation
    */
   messages: Message[];
+  /**
+   * LLM temperature. All models support the 0-1 range (some supports more).
+   * Defaults to 0;
+   */
+  temperature?: number;
   /**
    * Function calling mode, defaults to "native".
    */

x-pack/platform/plugins/shared/inference/common/http_apis.ts

@@ -16,6 +16,7 @@ export type ChatCompleteRequestBody = {
   connectorId: string;
   stream?: boolean;
   system?: string;
+  temperature?: number;
   messages: Message[];
   functionCalling?: FunctionCallingMode;
 } & ToolOptions;

x-pack/platform/plugins/shared/inference/public/chat_complete.test.ts

@@ -24,6 +24,7 @@ describe('createChatCompleteApi', () => {
       connectorId: 'my-connector',
       functionCalling: 'native',
       system: 'system',
+      temperature: 0.5,
       messages: [{ role: MessageRole.User, content: 'question' }],
     };
     await chatComplete(params as ChatCompleteOptions);
@@ -44,6 +45,7 @@ describe('createChatCompleteApi', () => {
       connectorId: 'my-connector',
       functionCalling: 'native',
       stream: true,
+      temperature: 0.4,
       system: 'system',
       messages: [{ role: MessageRole.User, content: 'question' }],
     };

x-pack/platform/plugins/shared/inference/public/chat_complete.ts

@@ -24,6 +24,7 @@ export function createChatCompleteApi({ http }: { http: HttpStart }) {
     system,
     toolChoice,
     tools,
+    temperature,
     functionCalling,
     stream,
   }: ChatCompleteOptions<ToolOptions, boolean>): ChatCompleteCompositeResponse<
@@ -36,6 +37,7 @@ export function createChatCompleteApi({ http }: { http: HttpStart }) {
       messages,
       toolChoice,
       tools,
+      temperature,
       functionCalling,
     };
 

x-pack/platform/plugins/shared/inference/server/chat_complete/adapters/bedrock/bedrock_claude_adapter.test.ts

@@ -430,5 +430,22 @@ describe('bedrockClaudeAdapter', () => {
         }),
       });
     });
+
+    it('propagates the temperature parameter', () => {
+      bedrockClaudeAdapter.chatComplete({
+        logger,
+        executor: executorMock,
+        messages: [{ role: MessageRole.User, content: 'question' }],
+        temperature: 0.9,
+      });
+
+      expect(executorMock.invoke).toHaveBeenCalledTimes(1);
+      expect(executorMock.invoke).toHaveBeenCalledWith({
+        subAction: 'invokeStream',
+        subActionParams: expect.objectContaining({
+          temperature: 0.9,
+        }),
+      });
+    });
   });
 });

x-pack/platform/plugins/shared/inference/server/chat_complete/adapters/bedrock/bedrock_claude_adapter.ts

@@ -26,7 +26,15 @@ import { processCompletionChunks } from './process_completion_chunks';
 import { addNoToolUsageDirective } from './prompts';
 
 export const bedrockClaudeAdapter: InferenceConnectorAdapter = {
-  chatComplete: ({ executor, system, messages, toolChoice, tools, abortSignal }) => {
+  chatComplete: ({
+    executor,
+    system,
+    messages,
+    toolChoice,
+    tools,
+    temperature = 0,
+    abortSignal,
+  }) => {
     const noToolUsage = toolChoice === ToolChoiceType.none;
 
     const subActionParams = {
@@ -34,7 +42,7 @@ export const bedrockClaudeAdapter: InferenceConnectorAdapter = {
       messages: messagesToBedrock(messages),
       tools: noToolUsage ? [] : toolsToBedrock(tools, messages),
       toolChoice: toolChoiceToBedrock(toolChoice),
-      temperature: 0,
+      temperature,
       stopSequences: ['\n\nHuman:'],
       signal: abortSignal,
     };

x-pack/platform/plugins/shared/inference/server/chat_complete/adapters/gemini/gemini_adapter.test.ts

@@ -32,6 +32,7 @@ describe('geminiAdapter', () => {
       tools: params.tools,
       toolConfig: params.toolConfig,
       systemInstruction: params.systemInstruction,
+      temperature: params.temperature,
     };
   }
 
@@ -501,5 +502,22 @@ describe('geminiAdapter', () => {
         }),
       });
     });
+
+    it('propagates the temperature parameter', () => {
+      geminiAdapter.chatComplete({
+        logger,
+        executor: executorMock,
+        messages: [{ role: MessageRole.User, content: 'question' }],
+        temperature: 0.6,
+      });
+
+      expect(executorMock.invoke).toHaveBeenCalledTimes(1);
+      expect(executorMock.invoke).toHaveBeenCalledWith({
+        subAction: 'invokeStream',
+        subActionParams: expect.objectContaining({
+          temperature: 0.6,
+        }),
+      });
+    });
   });
 });

x-pack/platform/plugins/shared/inference/server/chat_complete/adapters/gemini/gemini_adapter.ts

@@ -22,7 +22,15 @@ import { processVertexStream } from './process_vertex_stream';
 import type { GenerateContentResponseChunk, GeminiMessage, GeminiToolConfig } from './types';
 
 export const geminiAdapter: InferenceConnectorAdapter = {
-  chatComplete: ({ executor, system, messages, toolChoice, tools, abortSignal }) => {
+  chatComplete: ({
+    executor,
+    system,
+    messages,
+    toolChoice,
+    tools,
+    temperature = 0,
+    abortSignal,
+  }) => {
     return from(
       executor.invoke({
         subAction: 'invokeStream',
@@ -31,7 +39,7 @@ export const geminiAdapter: InferenceConnectorAdapter = {
           systemInstruction: system,
           tools: toolsToGemini(tools),
           toolConfig: toolChoiceToConfig(toolChoice),
-          temperature: 0,
+          temperature,
           signal: abortSignal,
           stopSequences: ['\n\nHuman:'],
         },

x-pack/platform/plugins/shared/inference/server/chat_complete/adapters/inference/inference_adapter.test.ts

@@ -144,5 +144,24 @@ describe('inferenceAdapter', () => {
         }),
       });
     });
+
+    it('propagates the temperature parameter', () => {
+      inferenceAdapter.chatComplete({
+        logger,
+        executor: executorMock,
+        messages: [{ role: MessageRole.User, content: 'question' }],
+        temperature: 0.4,
+      });
+
+      expect(executorMock.invoke).toHaveBeenCalledTimes(1);
+      expect(executorMock.invoke).toHaveBeenCalledWith({
+        subAction: 'unified_completion_stream',
+        subActionParams: expect.objectContaining({
+          body: expect.objectContaining({
+            temperature: 0.4,
+          }),
+        }),
+      });
+    });
   });
 });

x-pack/platform/plugins/shared/inference/server/chat_complete/adapters/inference/inference_adapter.ts

@@ -30,6 +30,7 @@ export const inferenceAdapter: InferenceConnectorAdapter = {
     toolChoice,
     tools,
     functionCalling,
+    temperature = 0,
     logger,
     abortSignal,
   }) => {
@@ -44,10 +45,12 @@ export const inferenceAdapter: InferenceConnectorAdapter = {
         tools,
       });
       request = {
+        temperature,
         messages: messagesToOpenAI({ system: wrapped.system, messages: wrapped.messages }),
       };
     } else {
       request = {
+        temperature,
         messages: messagesToOpenAI({ system, messages }),
         tool_choice: toolChoiceToOpenAI(toolChoice),
         tools: toolsToOpenAI(tools),

x-pack/platform/plugins/shared/inference/server/chat_complete/adapters/openai/openai_adapter.test.ts

@@ -354,6 +354,18 @@ describe('openAIAdapter', () => {
         }),
       });
     });
+
+    it('propagates the temperature', () => {
+      openAIAdapter.chatComplete({
+        logger,
+        executor: executorMock,
+        messages: [{ role: MessageRole.User, content: 'question' }],
+        temperature: 0.7,
+      });
+
+      expect(executorMock.invoke).toHaveBeenCalledTimes(1);
+      expect(getRequest().body.temperature).toBe(0.7);
+    });
   });
 
   describe('when handling the response', () => {

x-pack/platform/plugins/shared/inference/server/chat_complete/adapters/openai/openai_adapter.ts

@@ -25,11 +25,11 @@ export const openAIAdapter: InferenceConnectorAdapter = {
     messages,
     toolChoice,
     tools,
+    temperature = 0,
     functionCalling,
     logger,
     abortSignal,
   }) => {
-    const stream = true;
     const simulatedFunctionCalling = functionCalling === 'simulated';
 
     let request: Omit<OpenAI.ChatCompletionCreateParams, 'model'> & { model?: string };
@@ -41,12 +41,14 @@ export const openAIAdapter: InferenceConnectorAdapter = {
         tools,
       });
       request = {
-        stream,
+        stream: true,
+        temperature,
         messages: messagesToOpenAI({ system: wrapped.system, messages: wrapped.messages }),
       };
     } else {
       request = {
-        stream,
+        stream: true,
+        temperature,
         messages: messagesToOpenAI({ system, messages }),
         tool_choice: toolChoiceToOpenAI(toolChoice),
         tools: toolsToOpenAI(tools),
@@ -59,7 +61,7 @@ export const openAIAdapter: InferenceConnectorAdapter = {
         subActionParams: {
           body: JSON.stringify(request),
           signal: abortSignal,
-          stream,
+          stream: true,
         },
       })
     ).pipe(

x-pack/platform/plugins/shared/inference/server/chat_complete/api.test.ts

@@ -84,6 +84,7 @@ describe('createChatCompleteApi', () => {
     await chatComplete({
       connectorId: 'connectorId',
       messages: [{ role: MessageRole.User, content: 'question' }],
+      temperature: 0.7,
     });
 
     expect(inferenceAdapter.chatComplete).toHaveBeenCalledTimes(1);
@@ -91,6 +92,7 @@ describe('createChatCompleteApi', () => {
       messages: [{ role: MessageRole.User, content: 'question' }],
       executor: inferenceExecutor,
       logger,
+      temperature: 0.7,
     });
   });
 

x-pack/platform/plugins/shared/inference/server/chat_complete/api.ts

@@ -38,6 +38,7 @@ export function createChatCompleteApi({ request, actions, logger }: CreateChatCo
     messages,
     toolChoice,
     tools,
+    temperature,
     system,
     functionCalling,
     stream,
@@ -80,6 +81,7 @@ export function createChatCompleteApi({ request, actions, logger }: CreateChatCo
           messages,
           toolChoice,
           tools,
+          temperature,
           logger,
           functionCalling,
           abortSignal,

x-pack/platform/plugins/shared/inference/server/chat_complete/types.ts

@@ -29,6 +29,7 @@ export interface InferenceConnectorAdapter {
       messages: Message[];
       system?: string;
       functionCalling?: FunctionCallingMode;
+      temperature?: number;
       abortSignal?: AbortSignal;
       logger: Logger;
     } & ToolOptions

x-pack/platform/plugins/shared/inference/server/routes/chat_complete.ts

@@ -87,6 +87,7 @@ const chatCompleteBodySchema: Type<ChatCompleteRequestBody> = schema.object({
   functionCalling: schema.maybe(
     schema.oneOf([schema.literal('native'), schema.literal('simulated')])
   ),
+  temperature: schema.maybe(schema.number()),
 });
 
 export function registerChatCompleteRoute({

x-pack/test/functional_gen_ai/inference/tests/chat_complete.ts

@@ -25,6 +25,7 @@ export const chatCompleteSuite = (
           .set('kbn-xsrf', 'kibana')
           .send({
             connectorId,
+            temperature: 0.1,
             system: 'Please answer the user question',
             messages: [{ role: 'user', content: '2+2 ?' }],
           })
@@ -154,6 +155,7 @@ export const chatCompleteSuite = (
           .set('kbn-xsrf', 'kibana')
           .send({
             connectorId,
+            temperature: 0.1,
             system: 'Please answer the user question',
             messages: [{ role: 'user', content: '2+2 ?' }],
           })