@llumiverse/drivers 0.10.0 → 0.11.0

package/src/vertexai/models/gemini.ts

@@ -1,10 +1,10 @@
-import { Content, GenerateContentRequest, GenerativeModel, HarmBlockThreshold, HarmCategory, TextPart } from "@google-cloud/vertexai";
+import { Content, GenerateContentRequest, HarmBlockThreshold, HarmCategory, TextPart } from "@google-cloud/vertexai";
 import { AIModel, Completion, ExecutionOptions, ExecutionTokenUsage, ModelType, PromptOptions, PromptRole, PromptSegment } from "@llumiverse/core";
 import { asyncMap } from "@llumiverse/core/async";
 import { VertexAIDriver } from "../index.js";
 import { ModelDefinition } from "../models.js";
 
-function getGenerativeModel(driver: VertexAIDriver, options: ExecutionOptions): GenerativeModel {
+function getGenerativeModel(driver: VertexAIDriver, options: ExecutionOptions) {
     return driver.vertexai.preview.getGenerativeModel({
         model: options.model,
         //TODO pass in the options
@@ -106,8 +106,8 @@ export class GeminiModelDefinition implements ModelDefinition<GenerateContentReq
         const response = await r.response;
         const usage = response.usageMetadata;
         const token_usage: ExecutionTokenUsage = {
-            prompt: usage?.prompt_token_count,
-            result: usage?.candidates_token_count,
+            prompt: usage?.promptTokenCount,
+            result: usage?.candidatesTokenCount,
             total: usage?.totalTokenCount,
         }
 

package/src/vertexai/models/palm-model-base.ts

@@ -2,9 +2,25 @@ import { AIModel, Completion, ExecutionOptions, PromptOptions, PromptSegment } f
 import { transformSSEStream } from "@llumiverse/core/async";
 import { VertexAIDriver } from "../index.js";
 import { ModelDefinition } from "../models.js";
-import { PromptParamatersBase } from "../utils/prompts.js";
 import { generateStreamingPrompt } from "../utils/tensor.js";
 
+
+export interface PromptParamatersBase {
+    temperature?: number,
+    maxOutputTokens?: number,
+    topK?: number,
+    topP?: number,
+    groundingConfig?: string,
+    stopSequences?: string[],
+    candidateCount?: number,
+    logprobs?: number,
+    presencePenalty?: number,
+    frequencyPenalty?: number,
+    logitBias?: Record<string, number>,
+    seed?: number,
+    echo?: boolean
+}
+
 export interface NonStreamingPromptBase<InstanceType = any> {
     instances: InstanceType[];
     parameters: PromptParamatersBase;

package/src/vertexai/models/palm2-chat.ts

@@ -1,6 +1,6 @@
 import { AIModel, ModelType, PromptOptions, PromptRole, PromptSegment } from "@llumiverse/core";
+import { getJSONSafetyNotice } from "@llumiverse/core/formatters";
 import { VertexAIDriver } from "../index.js";
-import { getJSONSafetyNotice } from "../utils/prompts.js";
 import { AbstractPalmModelDefinition, NonStreamingPromptBase, PalmResponseMetadata, StreamingPromptBase } from "./palm-model-base.js";
 
 export interface Palm2ChatMessage {

package/src/vertexai/models/palm2-text.ts

@@ -1,6 +1,6 @@
 import { AIModel, ModelType, PromptOptions, PromptSegment } from "@llumiverse/core";
+import { formatTextPrompt } from "@llumiverse/core/formatters";
 import { VertexAIDriver } from "../index.js";
-import { getPromptAsText } from "../utils/prompts.js";
 import { AbstractPalmModelDefinition, NonStreamingPromptBase, PalmResponseMetadata, StreamingPromptBase } from "./palm-model-base.js";
 
 export type Palm2TextPrompt = NonStreamingPromptBase<{
@@ -50,7 +50,7 @@ export class Palm2TextDefinition extends AbstractPalmModelDefinition<Palm2TextPr
     createNonStreamingPrompt(_driver: VertexAIDriver, segments: PromptSegment[], opts: PromptOptions): Palm2TextPrompt {
         return {
             instances: [{
-                prompt: getPromptAsText(segments, opts)
+                prompt: formatTextPrompt(segments, opts.resultSchema)
             }],
             parameters: {
                 // put defauilts here