@promptbook/cli 0.92.0-31 → 0.92.0-32

package/esm/index.es.js

@@ -26,8 +26,8 @@ import http from 'http';
 import { Server } from 'socket.io';
 import swaggerUi from 'swagger-ui-express';
 import Anthropic from '@anthropic-ai/sdk';
-import { OpenAIClient, AzureKeyCredential } from '@azure/openai';
 import Bottleneck from 'bottleneck';
+import { OpenAIClient, AzureKeyCredential } from '@azure/openai';
 import OpenAI from 'openai';
 import { Readability } from '@mozilla/readability';
 import { JSDOM } from 'jsdom';
@@ -47,7 +47,7 @@ const BOOK_LANGUAGE_VERSION = '1.0.0';
  * @generated
  * @see https://github.com/webgptorg/promptbook
  */
-const PROMPTBOOK_ENGINE_VERSION = '0.92.0-31';
+const PROMPTBOOK_ENGINE_VERSION = '0.92.0-32';
 /**
  * TODO: string_promptbook_version should be constrained to the all versions of Promptbook engine
  * Note: [💞] Ignore a discrepancy between file name and entity name
@@ -3119,23 +3119,17 @@ class MultipleLlmExecutionTools {
      * Check the configuration of all execution tools
      */
     async checkConfiguration() {
-        // TODO: Maybe do it in parallel
-        for (const llmExecutionTools of this.llmExecutionTools) {
-            await llmExecutionTools.checkConfiguration();
-        }
+        // Note: Run checks in parallel
+        await Promise.all(this.llmExecutionTools.map((tools) => tools.checkConfiguration()));
     }
     /**
      * List all available models that can be used
      * This lists is a combination of all available models from all execution tools
      */
     async listModels() {
-        const availableModels = [];
-        for (const llmExecutionTools of this.llmExecutionTools) {
-            // TODO: [🪂] Obtain models in parallel
-            const models = await llmExecutionTools.listModels();
-            availableModels.push(...models);
-        }
-        return availableModels;
+        // Obtain all models in parallel and flatten
+        const modelArrays = await Promise.all(this.llmExecutionTools.map((tools) => tools.listModels()));
+        return modelArrays.flat();
     }
     /**
      * Calls the best available chat model
@@ -15249,6 +15243,8 @@ class AnthropicClaudeExecutionTools {
          * Anthropic Claude API client.
          */
         this.client = null;
+        const rate = this.options.maxRequestsPerMinute || DEFAULT_MAX_REQUESTS_PER_MINUTE;
+        this.limiter = new Bottleneck({ minTime: 60000 / rate });
     }
     get title() {
         return 'Anthropic Claude';
@@ -15300,8 +15296,6 @@ class AnthropicClaudeExecutionTools {
             //                                            <- TODO: [🌾] Make some global max cap for maxTokens
             temperature: modelRequirements.temperature,
             system: modelRequirements.systemMessage,
-            // <- TODO: [🈁] Use `seed` here AND/OR use is `isDeterministic` for entire execution tools
-            // <- Note: [🧆]
             messages: [
                 {
                     role: 'user',
@@ -15310,14 +15304,14 @@ class AnthropicClaudeExecutionTools {
                     //    @see https://docs.anthropic.com/en/docs/test-and-evaluate/strengthen-guardrails/increase-consistency#specify-the-desired-output-format
                 },
             ],
-            // TODO: Is here some equivalent of user identification?> user: this.options.user,
         };
         const start = $getCurrentDate();
-        let complete;
         if (this.options.isVerbose) {
             console.info(colors.bgWhite('rawRequest'), JSON.stringify(rawRequest, null, 4));
         }
-        const rawResponse = await client.messages.create(rawRequest).catch((error) => {
+        const rawResponse = await this.limiter
+            .schedule(() => client.messages.create(rawRequest))
+            .catch((error) => {
             if (this.options.isVerbose) {
                 console.info(colors.bgRed('error'), error);
             }
@@ -15337,12 +15331,11 @@ class AnthropicClaudeExecutionTools {
             throw new PipelineExecutionError(`Returned content is not "text" type but "${contentBlock.type}"`);
         }
         const resultContent = contentBlock.text;
-        // eslint-disable-next-line prefer-const
-        complete = $getCurrentDate();
+        const complete = $getCurrentDate();
         const usage = computeAnthropicClaudeUsage(rawPromptContent || '', resultContent || '', rawResponse);
         return exportJson({
             name: 'promptResult',
-            message: `Result of \`AzureOpenAiExecutionTools.callChatModel\``,
+            message: `Result of \`AnthropicClaudeExecutionTools.callChatModel\``,
             order: [],
             value: {
                 content: resultContent,
@@ -15359,83 +15352,59 @@ class AnthropicClaudeExecutionTools {
             },
         });
     }
-    /*
-    TODO: [👏]
-    public async callCompletionModel(
-        prompt: Pick<Prompt, 'content' | 'parameters' | 'modelRequirements'>,
-    ): Promise<CompletionPromptResult> {
-
+    /**
+     * Calls Anthropic Claude API to use a completion model.
+     */
+    async callCompletionModel(prompt) {
         if (this.options.isVerbose) {
             console.info('🖋 Anthropic Claude callCompletionModel call');
         }
-
         const { content, parameters, modelRequirements } = prompt;
-
-        // TODO: [☂] Use here more modelRequirements
         if (modelRequirements.modelVariant !== 'COMPLETION') {
             throw new PipelineExecutionError('Use callCompletionModel only for COMPLETION variant');
         }
-
+        const client = await this.getClient();
         const modelName = modelRequirements.modelName || this.getDefaultChatModel().modelName;
-        const modelSettings = {
+        const rawPromptContent = templateParameters(content, { ...parameters, modelName });
+        const rawRequest = {
             model: modelName,
-            max_tokens: modelRequirements.maxTokens || 2000, // <- Note: 2000 is for lagacy reasons
-            //                                                  <- TODO: [🌾] Make some global max cap for maxTokens
-            // <- TODO: Use here `systemMessage`, `temperature` and `seed`
-        };
-
-        const rawRequest: xxxx.Completions.CompletionCreateParamsNonStreaming = {
-            ...modelSettings,
+            max_tokens_to_sample: modelRequirements.maxTokens || 2000,
+            temperature: modelRequirements.temperature,
             prompt: rawPromptContent,
-            user: this.options.user,
         };
-        const start: string_date_iso8601 = $getCurrentDate();
-        let complete: string_date_iso8601;
-
-        if (this.options.isVerbose) {
-            console.info(colors.bgWhite('rawRequest'), JSON.stringify(rawRequest, null, 4));
-        }
-        const rawResponse = await this.client.completions.create(rawRequest).catch((error) => {
-                if (this.options.isVerbose) {
-                    console.info(colors.bgRed('error'), error);
-                }
-                throw error;
-            });
-
-
+        const start = $getCurrentDate();
+        const rawResponse = await this.limiter
+            .schedule(() => client.completions.create(rawRequest))
+            .catch((error) => {
+            if (this.options.isVerbose) {
+                console.info(colors.bgRed('error'), error);
+            }
+            throw error;
+        });
         if (this.options.isVerbose) {
             console.info(colors.bgWhite('rawResponse'), JSON.stringify(rawResponse, null, 4));
         }
-
-        if (!rawResponse.choices[0]) {
-            throw new PipelineExecutionError('No choises from Anthropic Claude');
+        if (!rawResponse.completion) {
+            throw new PipelineExecutionError('No completion from Anthropic Claude');
         }
-
-        if (rawResponse.choices.length > 1) {
-            // TODO: This should be maybe only warning
-            throw new PipelineExecutionError('More than one choise from Anthropic Claude');
-        }
-
-        const resultContent = rawResponse.choices[0].text;
-        // eslint-disable-next-line prefer-const
-        complete = $getCurrentDate();
-        const usage = { price: 'UNKNOWN', inputTokens: 0, outputTokens: 0 /* <- TODO: [🐞] Compute usage * / } satisfies Usage;
-
-
-
-        return $exportJson({ name: 'promptResult',message: Result of \`AzureOpenAiExecutionTools callChatModel\`, order: [],value:{
-            content: resultContent,
-            modelName: rawResponse.model || model,
-            timing: {
-                start,
-                complete,
+        const resultContent = rawResponse.completion;
+        const complete = $getCurrentDate();
+        const usage = computeAnthropicClaudeUsage(rawPromptContent, resultContent, rawResponse);
+        return exportJson({
+            name: 'promptResult',
+            message: `Result of \`AnthropicClaudeExecutionTools.callCompletionModel\``,
+            order: [],
+            value: {
+                content: resultContent,
+                modelName: rawResponse.model || modelName,
+                timing: { start, complete },
+                usage,
+                rawPromptContent,
+                rawRequest,
+                rawResponse,
             },
-            usage,
-            rawResponse,
-            // <- [🗯]
         });
     }
-    */
     // <- Note: [🤖] callXxxModel
     /**
      * Get the model that should be used as default
@@ -15444,7 +15413,7 @@ class AnthropicClaudeExecutionTools {
         const model = ANTHROPIC_CLAUDE_MODELS.find(({ modelName }) => modelName.startsWith(defaultModelName));
         if (model === undefined) {
             throw new UnexpectedError(spaceTrim((block) => `
-                          Cannot find model in OpenAI models with name "${defaultModelName}" which should be used as default.
+                          Cannot find model in Anthropic Claude models with name "${defaultModelName}" which should be used as default.
 
                           Available models:
                           ${block(ANTHROPIC_CLAUDE_MODELS.map(({ modelName }) => `- "${modelName}"`).join('\n'))}
@@ -17078,11 +17047,9 @@ function computeOpenAiUsage(promptContent, // <- Note: Intentionally using [] to
 resultContent, rawResponse) {
     var _a, _b;
     if (rawResponse.usage === undefined) {
-        console.log('!!! computeOpenAiUsage', 'The usage is not defined in the response from OpenAI');
         throw new PipelineExecutionError('The usage is not defined in the response from OpenAI');
     }
     if (((_a = rawResponse.usage) === null || _a === void 0 ? void 0 : _a.prompt_tokens) === undefined) {
-        console.log('!!! computeOpenAiUsage', 'In OpenAI response `usage.prompt_tokens` not defined');
         throw new PipelineExecutionError('In OpenAI response `usage.prompt_tokens` not defined');
     }
     const inputTokens = rawResponse.usage.prompt_tokens;
@@ -17096,15 +17063,6 @@ resultContent, rawResponse) {
             isUncertain = true;
         }
     }
-    console.log('!!! computeOpenAiUsage', {
-        inputTokens,
-        outputTokens,
-        rawResponse,
-        'rawResponse.model': rawResponse.model,
-        OPENAI_MODELS,
-        resultContent,
-        modelInfo,
-    });
     let price;
     if (modelInfo === undefined || modelInfo.pricing === undefined) {
         price = uncertainNumber();
@@ -17248,7 +17206,6 @@ class OpenAiExecutionTools {
             user: (_a = this.options.userId) === null || _a === void 0 ? void 0 : _a.toString(),
         };
         const start = $getCurrentDate();
-        let complete;
         if (this.options.isVerbose) {
             console.info(colors.bgWhite('rawRequest'), JSON.stringify(rawRequest, null, 4));
         }
@@ -17264,6 +17221,7 @@ class OpenAiExecutionTools {
         if (this.options.isVerbose) {
             console.info(colors.bgWhite('rawResponse'), JSON.stringify(rawResponse, null, 4));
         }
+        const complete = $getCurrentDate();
         if (!rawResponse.choices[0]) {
             throw new PipelineExecutionError('No choises from OpenAI');
         }
@@ -17272,8 +17230,6 @@ class OpenAiExecutionTools {
             throw new PipelineExecutionError('More than one choise from OpenAI');
         }
         const resultContent = rawResponse.choices[0].message.content;
-        // eslint-disable-next-line prefer-const
-        complete = $getCurrentDate();
         const usage = computeOpenAiUsage(content || '', resultContent || '', rawResponse);
         if (resultContent === null) {
             throw new PipelineExecutionError('No response message from OpenAI');
@@ -17327,7 +17283,6 @@ class OpenAiExecutionTools {
             user: (_a = this.options.userId) === null || _a === void 0 ? void 0 : _a.toString(),
         };
         const start = $getCurrentDate();
-        let complete;
         if (this.options.isVerbose) {
             console.info(colors.bgWhite('rawRequest'), JSON.stringify(rawRequest, null, 4));
         }
@@ -17343,6 +17298,7 @@ class OpenAiExecutionTools {
         if (this.options.isVerbose) {
             console.info(colors.bgWhite('rawResponse'), JSON.stringify(rawResponse, null, 4));
         }
+        const complete = $getCurrentDate();
         if (!rawResponse.choices[0]) {
             throw new PipelineExecutionError('No choises from OpenAI');
         }
@@ -17351,8 +17307,6 @@ class OpenAiExecutionTools {
             throw new PipelineExecutionError('More than one choise from OpenAI');
         }
         const resultContent = rawResponse.choices[0].text;
-        // eslint-disable-next-line prefer-const
-        complete = $getCurrentDate();
         const usage = computeOpenAiUsage(content || '', resultContent || '', rawResponse);
         return exportJson({
             name: 'promptResult',
@@ -17393,7 +17347,6 @@ class OpenAiExecutionTools {
             model: modelName,
         };
         const start = $getCurrentDate();
-        let complete;
         if (this.options.isVerbose) {
             console.info(colors.bgWhite('rawRequest'), JSON.stringify(rawRequest, null, 4));
         }
@@ -17409,12 +17362,11 @@ class OpenAiExecutionTools {
         if (this.options.isVerbose) {
             console.info(colors.bgWhite('rawResponse'), JSON.stringify(rawResponse, null, 4));
         }
+        const complete = $getCurrentDate();
         if (rawResponse.data.length !== 1) {
             throw new PipelineExecutionError(`Expected exactly 1 data item in response, got ${rawResponse.data.length}`);
         }
         const resultContent = rawResponse.data[0].embedding;
-        // eslint-disable-next-line prefer-const
-        complete = $getCurrentDate();
         const usage = computeOpenAiUsage(content || '', '', 
         // <- Note: Embedding does not have result content
         rawResponse);
@@ -17442,7 +17394,8 @@ class OpenAiExecutionTools {
      * Get the model that should be used as default
      */
     getDefaultModel(defaultModelName) {
-        const model = OPENAI_MODELS.find(({ modelName }) => modelName === defaultModelName);
+        // Note: Match exact or prefix for model families
+        const model = OPENAI_MODELS.find(({ modelName }) => modelName === defaultModelName || modelName.startsWith(defaultModelName));
         if (model === undefined) {
             throw new UnexpectedError(spaceTrim((block) => `
                             Cannot find model in OpenAI models with name "${defaultModelName}" which should be used as default.