@llumiverse/drivers 0.22.0 → 0.23.0-dev-20251118

package/src/vertexai/models/llama.ts

@@ -1,5 +1,5 @@
 import {
-    AIModel, Completion, CompletionChunk, CompletionChunkObject, ExecutionOptions, ModelType,
+    AIModel, Completion, CompletionChunkObject, ExecutionOptions, ModelType,
     PromptOptions, PromptRole, PromptSegment,
     TextFallbackOptions
 } from "@llumiverse/core";
@@ -195,7 +195,7 @@ export class LLamaModelDefinition implements ModelDefinition<LLamaPrompt> {
         });
 
         return {
-            result: text,
+            result: [{ type: "text", value: text }],
             token_usage: {
                 prompt: result.usage.prompt_tokens,
                 result: result.usage.completion_tokens,
@@ -206,11 +206,11 @@ export class LLamaModelDefinition implements ModelDefinition<LLamaPrompt> {
         };
     }
 
-    async requestTextCompletionStream(driver: VertexAIDriver, prompt: LLamaPrompt, options: ExecutionOptions): Promise<AsyncIterable<CompletionChunk>> {
+    async requestTextCompletionStream(driver: VertexAIDriver, prompt: LLamaPrompt, options: ExecutionOptions): Promise<AsyncIterable<CompletionChunkObject>> {
         const splits = options.model.split("/");
         const modelName = splits[splits.length - 1];
 
-        let conversation = updateConversation(options.conversation as LLamaPrompt, prompt);
+        const conversation = updateConversation(options.conversation as LLamaPrompt, prompt);
 
         const modelOptions = options.model_options as TextFallbackOptions;
 
@@ -247,8 +247,9 @@ export class LLamaModelDefinition implements ModelDefinition<LLamaPrompt> {
         return transformSSEStream(stream, (data: string): CompletionChunkObject => {
             const json = JSON.parse(data) as LLamaStreamResponse;
             const choice = json.choices?.[0];
+            const content = choice?.delta?.content ?? '';
             return {
-                result: choice?.delta?.content ?? '',
+                result: content ? [{ type: "text", value: content }] : [],
                 finish_reason: choice?.finish_reason,
                 token_usage: json.usage ? {
                     prompt: json.usage.prompt_tokens,

package/src/vertexai/models.ts

@@ -1,5 +1,5 @@
-import { AIModel, Completion, PromptSegment, ExecutionOptions, CompletionChunk } from "@llumiverse/core";
-import { VertexAIDriver , trimModelName} from "./index.js";
+import { AIModel, Completion, PromptSegment, ExecutionOptions, CompletionChunkObject } from "@llumiverse/core";
+import { VertexAIDriver, trimModelName } from "./index.js";
 import { GeminiModelDefinition } from "./models/gemini.js";
 import { ClaudeModelDefinition } from "./models/claude.js";
 import { LLamaModelDefinition } from "./models/llama.js";
@@ -9,15 +9,27 @@ export interface ModelDefinition<PromptT = any> {
     versions?: string[]; // the versions of the model that are available. ex: ['001', '002']
     createPrompt: (driver: VertexAIDriver, segments: PromptSegment[], options: ExecutionOptions) => Promise<PromptT>;
     requestTextCompletion: (driver: VertexAIDriver, prompt: PromptT, options: ExecutionOptions) => Promise<Completion>;
-    requestTextCompletionStream: (driver: VertexAIDriver, prompt: PromptT, options: ExecutionOptions) => Promise<AsyncIterable<CompletionChunk>>;
+    requestTextCompletionStream: (driver: VertexAIDriver, prompt: PromptT, options: ExecutionOptions) => Promise<AsyncIterable<CompletionChunkObject>>;
     preValidationProcessing?(result: Completion, options: ExecutionOptions): { result: Completion, options: ExecutionOptions };
 }
 
 export function getModelDefinition(model: string): ModelDefinition {
     const splits = model.split("/");
-    const publisher = splits[1];
-    const modelName = trimModelName(splits[splits.length - 1]);
-    
+
+    // Handle both formats: "publishers/anthropic/models/..." and "locations/.../publishers/anthropic/models/..."
+    let publisher: string | undefined;
+    let modelName: string;
+
+    const publisherIndex = splits.indexOf("publishers");
+    if (publisherIndex !== -1 && publisherIndex + 1 < splits.length) {
+        publisher = splits[publisherIndex + 1];
+        modelName = trimModelName(splits[splits.length - 1]);
+    } else {
+        // Fallback to old logic for backward compatibility
+        publisher = splits[1];
+        modelName = trimModelName(splits[splits.length - 1]);
+    }
+
     if (publisher?.includes("anthropic")) {
         return new ClaudeModelDefinition(modelName);
     } else if (publisher?.includes("google")) {

package/src/watsonx/index.ts

@@ -1,4 +1,4 @@
-import { AIModel, AbstractDriver, Completion, CompletionChunk, DriverOptions, EmbeddingsOptions, EmbeddingsResult, ExecutionOptions, TextFallbackOptions } from "@llumiverse/core";
+import { AIModel, AbstractDriver, Completion, CompletionChunkObject, DriverOptions, EmbeddingsOptions, EmbeddingsResult, ExecutionOptions, TextFallbackOptions } from "@llumiverse/core";
 import { transformSSEStream } from "@llumiverse/core/async";
 import { FetchClient } from "@vertesia/api-fetch-client";
 import { GenerateEmbeddingPayload, GenerateEmbeddingResponse, WatsonAuthToken, WatsonxListModelResponse, WatsonxModelSpec, WatsonxTextGenerationPayload, WatsonxTextGenerationResponse } from "./interfaces.js";
@@ -29,9 +29,9 @@ export class WatsonxDriver extends AbstractDriver<WatsonxDriverOptions, string>
         this.fetchClient = new FetchClient(this.endpoint_url).withAuthCallback(async () => this.getAuthToken().then(token => `Bearer ${token}`));
     }
 
-    async requestTextCompletion(prompt: string, options: ExecutionOptions): Promise<Completion<any>> {
+    async requestTextCompletion(prompt: string, options: ExecutionOptions): Promise<Completion> {
         if (options.model_options?._option_id !== "text-fallback") {
-            this.logger.warn("Invalid model options", { options: options.model_options });
+            this.logger.warn({ options: options.model_options }, "Invalid model options");
         }
         options.model_options = options.model_options as TextFallbackOptions | undefined;
 
@@ -53,7 +53,7 @@ export class WatsonxDriver extends AbstractDriver<WatsonxDriverOptions, string>
         const result = res.results[0];
 
         return {
-            result: result.generated_text,
+            result: [{ type: "text", value: result.generated_text }],
             token_usage: {
                 prompt: result.input_token_count,
                 result: result.generated_token_count,
@@ -64,9 +64,9 @@ export class WatsonxDriver extends AbstractDriver<WatsonxDriverOptions, string>
         }
     }
 
-    async requestTextCompletionStream(prompt: string, options: ExecutionOptions): Promise<AsyncIterable<CompletionChunk>> {
+    async requestTextCompletionStream(prompt: string, options: ExecutionOptions): Promise<AsyncIterable<CompletionChunkObject>> {
         if (options.model_options?._option_id !== "text-fallback") {
-            this.logger.warn("Invalid model options", { options: options.model_options });
+            this.logger.warn({ options: options.model_options }, "Invalid model options");
         }
         options.model_options = options.model_options as TextFallbackOptions | undefined;
         const payload: WatsonxTextGenerationPayload = {
@@ -90,7 +90,7 @@ export class WatsonxDriver extends AbstractDriver<WatsonxDriverOptions, string>
         return transformSSEStream(stream, (data: string) => {
             const json = JSON.parse(data) as WatsonxTextGenerationResponse;
             return {
-                result: json.results[0]?.generated_text ?? '',
+                result: json.results[0]?.generated_text ? [{ type: "text", value: json.results[0].generated_text }] : [],
                 finish_reason: watsonFinishReason(json.results[0]?.stop_reason),
                 token_usage: {
                     prompt: json.results[0].input_token_count,
@@ -152,7 +152,7 @@ export class WatsonxDriver extends AbstractDriver<WatsonxDriverOptions, string>
         return this.listModels()
             .then(() => true)
             .catch((err) => {
-                this.logger.warn("Failed to connect to WatsonX", { error: err });
+                this.logger.warn({ error: err }, "Failed to connect to WatsonX");
                 return false
             });
     }

package/src/xai/index.ts

@@ -5,7 +5,7 @@ import OpenAI from "openai";
 import { BaseOpenAIDriver } from "../openai/index.js";
 
 export interface xAiDriverOptions extends DriverOptions {
-    
+
     apiKey: string;
 
     endpoint?: string;
@@ -17,7 +17,7 @@ export class xAIDriver extends BaseOpenAIDriver {
     service: OpenAI;
     provider: "xai";
     xai_service: FetchClient;
-    DEFAULT_ENDPOINT = "https://api.x.ai/v1"; 
+    DEFAULT_ENDPOINT = "https://api.x.ai/v1";
 
     constructor(opts: xAiDriverOptions) {
         super(opts);
@@ -29,8 +29,8 @@ export class xAIDriver extends BaseOpenAIDriver {
         this.service = new OpenAI({
             apiKey: opts.apiKey,
             baseURL: opts.endpoint ?? this.DEFAULT_ENDPOINT,
-          });
-        this.xai_service = new FetchClient(opts.endpoint ?? this.DEFAULT_ENDPOINT ).withAuthCallback(async () => `Bearer ${opts.apiKey}`);
+        });
+        this.xai_service = new FetchClient(opts.endpoint ?? this.DEFAULT_ENDPOINT).withAuthCallback(async () => `Bearer ${opts.apiKey}`);
         this.provider = "xai";
         //this.formatPrompt = this._formatPrompt; //TODO: fix xai prompt formatting
     }
@@ -43,7 +43,7 @@ export class xAIDriver extends BaseOpenAIDriver {
             useToolForFormatting: false,
         }
 
-        const p = await formatOpenAILikeMultimodalPrompt(segments, {...options, ...opts}) as OpenAI.Chat.Completions.ChatCompletionMessageParam[];
+        const p = await formatOpenAILikeMultimodalPrompt(segments, { ...options, ...opts }) as OpenAI.Chat.Completions.ChatCompletionMessageParam[];
 
         return p;
 
@@ -51,7 +51,7 @@ export class xAIDriver extends BaseOpenAIDriver {
 
     extractDataFromResponse(_options: ExecutionOptions, result: OpenAI.Chat.Completions.ChatCompletion): Completion {
         return {
-            result: result.choices[0].message.content,
+            result: result.choices[0].message.content ? [{ type: "text", value: result.choices[0].message.content }] : [],
             finish_reason: result.choices[0].finish_reason,
             token_usage: {
                 prompt: result.usage?.prompt_tokens,
@@ -63,7 +63,7 @@ export class xAIDriver extends BaseOpenAIDriver {
 
     async listModels(): Promise<AIModel[]> {
         const [lm, em] = await Promise.all([
-            this.xai_service.get("/language-models") ,
+            this.xai_service.get("/language-models"),
             this.xai_service.get("/embedding-models")
         ]) as [xAIModelResponse, xAIModelResponse];
 
@@ -93,9 +93,9 @@ export class xAIDriver extends BaseOpenAIDriver {
 
 interface xAIModelResponse {
     models: xAIModel[];
-  }
-  
-  interface xAIModel {
+}
+
+interface xAIModel {
     completion_text_token_price: number;
     created: number;
     id: string;
@@ -105,5 +105,4 @@ interface xAIModelResponse {
     owned_by: string;
     prompt_image_token_price: number;
     prompt_text_token_price: number;
-  }
-  
+}