modelfusion 0.98.0 → 0.99.0

package/model-provider/ollama/OllamaTextGenerationModel.cjs

@@ -66,7 +66,7 @@ class OllamaTextGenerationModel extends AbstractModel_js_1.AbstractModel {
                         num_ctx: this.settings.contextWindowSize,
                         num_gpu: this.settings.numGpu,
                         num_gqa: this.settings.numGqa,
-                        num_predict: this.settings.maxCompletionTokens,
+                        num_predict: this.settings.maxGenerationTokens,
                         num_threads: this.settings.numThreads,
                         repeat_last_n: this.settings.repeatLastN,
                         repeat_penalty: this.settings.repeatPenalty,
@@ -90,7 +90,7 @@ class OllamaTextGenerationModel extends AbstractModel_js_1.AbstractModel {
     }
     get settingsForEvent() {
         const eventSettingProperties = [
-            "maxCompletionTokens",
+            "maxGenerationTokens",
             "stopSequences",
             "contextWindowSize",
             "temperature",
@@ -114,14 +114,14 @@ class OllamaTextGenerationModel extends AbstractModel_js_1.AbstractModel {
         ];
         return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
     }
-    async doGenerateText(prompt, options) {
+    async doGenerateTexts(prompt, options) {
         const response = await this.callAPI(prompt, {
             ...options,
             responseFormat: exports.OllamaTextGenerationResponseFormat.json,
         });
         return {
             response,
-            text: response.response,
+            texts: [response.response],
         };
     }
     doStreamText(prompt, options) {

package/model-provider/ollama/OllamaTextGenerationModel.d.ts

@@ -132,7 +132,7 @@ export declare class OllamaTextGenerationModel<CONTEXT_WINDOW_SIZE extends numbe
         responseFormat: OllamaTextGenerationResponseFormatType<RESPONSE>;
     } & FunctionOptions): Promise<RESPONSE>;
     get settingsForEvent(): Partial<OllamaTextGenerationModelSettings<CONTEXT_WINDOW_SIZE>>;
-    doGenerateText(prompt: OllamaTextGenerationPrompt, options?: FunctionOptions): Promise<{
+    doGenerateTexts(prompt: OllamaTextGenerationPrompt, options?: FunctionOptions): Promise<{
         response: {
             response: string;
             model: string;
@@ -144,7 +144,7 @@ export declare class OllamaTextGenerationModel<CONTEXT_WINDOW_SIZE extends numbe
             load_duration?: number | undefined;
             context?: number[] | undefined;
         };
-        text: string;
+        texts: string[];
     }>;
     doStreamText(prompt: OllamaTextGenerationPrompt, options?: FunctionOptions): Promise<AsyncIterable<Delta<string>>>;
     asToolCallGenerationModel<INPUT_PROMPT>(promptTemplate: ToolCallPromptTemplate<INPUT_PROMPT, OllamaTextGenerationPrompt>): TextGenerationToolCallModel<INPUT_PROMPT, OllamaTextGenerationPrompt, this>;

package/model-provider/ollama/OllamaTextGenerationModel.js

@@ -63,7 +63,7 @@ export class OllamaTextGenerationModel extends AbstractModel {
                         num_ctx: this.settings.contextWindowSize,
                         num_gpu: this.settings.numGpu,
                         num_gqa: this.settings.numGqa,
-                        num_predict: this.settings.maxCompletionTokens,
+                        num_predict: this.settings.maxGenerationTokens,
                         num_threads: this.settings.numThreads,
                         repeat_last_n: this.settings.repeatLastN,
                         repeat_penalty: this.settings.repeatPenalty,
@@ -87,7 +87,7 @@ export class OllamaTextGenerationModel extends AbstractModel {
     }
     get settingsForEvent() {
         const eventSettingProperties = [
-            "maxCompletionTokens",
+            "maxGenerationTokens",
             "stopSequences",
             "contextWindowSize",
             "temperature",
@@ -111,14 +111,14 @@ export class OllamaTextGenerationModel extends AbstractModel {
         ];
         return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
     }
-    async doGenerateText(prompt, options) {
+    async doGenerateTexts(prompt, options) {
         const response = await this.callAPI(prompt, {
             ...options,
             responseFormat: OllamaTextGenerationResponseFormat.json,
         });
         return {
             response,
-            text: response.response,
+            texts: [response.response],
         };
     }
     doStreamText(prompt, options) {

package/model-provider/openai/OpenAICompletionModel.cjs

@@ -137,7 +137,7 @@ exports.calculateOpenAICompletionCostInMillicents = calculateOpenAICompletionCos
  * const model = new OpenAICompletionModel({
  *   model: "gpt-3.5-turbo-instruct",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -180,33 +180,60 @@ class OpenAICompletionModel extends AbstractModel_js_1.AbstractModel {
         return (0, countTokens_js_1.countTokens)(this.tokenizer, input);
     }
     async callAPI(prompt, options) {
-        const { run, responseFormat } = options;
-        const callSettings = {
-            user: this.settings.isUserIdForwardingEnabled ? run?.userId : undefined,
-            // Copied settings:
-            ...this.settings,
-            // map to OpenAI API names:
-            stop: this.settings.stopSequences,
-            maxTokens: this.settings.maxCompletionTokens,
-            // other settings:
-            abortSignal: run?.abortSignal,
-            prompt,
-            responseFormat,
-        };
+        const api = this.settings.api ?? new OpenAIApiConfiguration_js_1.OpenAIApiConfiguration();
+        const user = this.settings.isUserIdForwardingEnabled
+            ? options.run?.userId
+            : undefined;
+        const abortSignal = options.run?.abortSignal;
+        let { stopSequences } = this.settings;
+        const openaiResponseFormat = options.responseFormat;
         return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
-            retry: callSettings.api?.retry,
-            throttle: callSettings.api?.throttle,
-            call: async () => callOpenAICompletionAPI(callSettings),
+            retry: api.retry,
+            throttle: api.throttle,
+            call: async () => {
+                // empty arrays are not allowed for stop:
+                if (stopSequences != null &&
+                    Array.isArray(stopSequences) &&
+                    stopSequences.length === 0) {
+                    stopSequences = undefined;
+                }
+                return (0, postToApi_js_1.postJsonToApi)({
+                    url: api.assembleUrl("/completions"),
+                    headers: api.headers,
+                    body: {
+                        stream: openaiResponseFormat.stream,
+                        model: this.settings.model,
+                        prompt,
+                        suffix: this.settings.suffix,
+                        max_tokens: this.settings.maxGenerationTokens,
+                        temperature: this.settings.temperature,
+                        top_p: this.settings.topP,
+                        n: this.settings.numberOfGenerations,
+                        logprobs: this.settings.logprobs,
+                        echo: this.settings.echo,
+                        stop: this.settings.stopSequences,
+                        seed: this.settings.seed,
+                        presence_penalty: this.settings.presencePenalty,
+                        frequency_penalty: this.settings.frequencyPenalty,
+                        best_of: this.settings.bestOf,
+                        logit_bias: this.settings.logitBias,
+                        user,
+                    },
+                    failedResponseHandler: OpenAIError_js_1.failedOpenAICallResponseHandler,
+                    successfulResponseHandler: openaiResponseFormat.handler,
+                    abortSignal,
+                });
+            },
         });
     }
     get settingsForEvent() {
         const eventSettingProperties = [
-            "maxCompletionTokens",
+            "maxGenerationTokens",
             "stopSequences",
+            "numberOfGenerations",
             "suffix",
             "temperature",
             "topP",
-            "n",
             "logprobs",
             "echo",
             "presencePenalty",
@@ -217,14 +244,14 @@ class OpenAICompletionModel extends AbstractModel_js_1.AbstractModel {
         ];
         return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
     }
-    async doGenerateText(prompt, options) {
+    async doGenerateTexts(prompt, options) {
         const response = await this.callAPI(prompt, {
             ...options,
             responseFormat: exports.OpenAITextResponseFormat.json,
         });
         return {
             response,
-            text: response.choices[0].text,
+            texts: response.choices.map((choice) => choice.text),
             usage: {
                 promptTokens: response.usage.prompt_tokens,
                 completionTokens: response.usage.completion_tokens,
@@ -287,38 +314,6 @@ const OpenAICompletionResponseSchema = zod_1.z.object({
         total_tokens: zod_1.z.number(),
     }),
 });
-async function callOpenAICompletionAPI({ api = new OpenAIApiConfiguration_js_1.OpenAIApiConfiguration(), abortSignal, responseFormat, model, prompt, suffix, maxTokens, temperature, topP, n, logprobs, echo, stop, presencePenalty, frequencyPenalty, bestOf, logitBias, seed, user, }) {
-    // empty arrays are not allowed for stop:
-    if (stop != null && Array.isArray(stop) && stop.length === 0) {
-        stop = undefined;
-    }
-    return (0, postToApi_js_1.postJsonToApi)({
-        url: api.assembleUrl("/completions"),
-        headers: api.headers,
-        body: {
-            stream: responseFormat.stream,
-            model,
-            prompt,
-            suffix,
-            max_tokens: maxTokens,
-            temperature,
-            top_p: topP,
-            n,
-            logprobs,
-            echo,
-            stop,
-            seed,
-            presence_penalty: presencePenalty,
-            frequency_penalty: frequencyPenalty,
-            best_of: bestOf,
-            logit_bias: logitBias,
-            user,
-        },
-        failedResponseHandler: OpenAIError_js_1.failedOpenAICallResponseHandler,
-        successfulResponseHandler: responseFormat.handler,
-        abortSignal,
-    });
-}
 exports.OpenAITextResponseFormat = {
     /**
      * Returns the response as a JSON object.

package/model-provider/openai/OpenAICompletionModel.d.ts

@@ -101,13 +101,10 @@ export interface OpenAICompletionCallSettings {
     api?: ApiConfiguration;
     model: OpenAICompletionModelType;
     suffix?: string;
-    maxTokens?: number;
     temperature?: number;
     topP?: number;
-    n?: number;
     logprobs?: number;
     echo?: boolean;
-    stop?: string | string[];
     presencePenalty?: number;
     frequencyPenalty?: number;
     bestOf?: number;
@@ -126,7 +123,7 @@ export interface OpenAICompletionModelSettings extends TextGenerationModelSettin
  * const model = new OpenAICompletionModel({
  *   model: "gpt-3.5-turbo-instruct",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -146,7 +143,7 @@ export declare class OpenAICompletionModel extends AbstractModel<OpenAICompletio
         responseFormat: OpenAITextResponseFormatType<RESULT>;
     } & FunctionOptions): Promise<RESULT>;
     get settingsForEvent(): Partial<OpenAICompletionModelSettings>;
-    doGenerateText(prompt: string, options?: FunctionOptions): Promise<{
+    doGenerateTexts(prompt: string, options?: FunctionOptions): Promise<{
         response: {
             object: "text_completion";
             usage: {
@@ -165,7 +162,7 @@ export declare class OpenAICompletionModel extends AbstractModel<OpenAICompletio
             }[];
             system_fingerprint?: string | undefined;
         };
-        text: string;
+        texts: string[];
         usage: {
             promptTokens: number;
             completionTokens: number;

package/model-provider/openai/OpenAICompletionModel.js

@@ -131,7 +131,7 @@ export const calculateOpenAICompletionCostInMillicents = ({ model, response, })
  * const model = new OpenAICompletionModel({
  *   model: "gpt-3.5-turbo-instruct",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -174,33 +174,60 @@ export class OpenAICompletionModel extends AbstractModel {
         return countTokens(this.tokenizer, input);
     }
     async callAPI(prompt, options) {
-        const { run, responseFormat } = options;
-        const callSettings = {
-            user: this.settings.isUserIdForwardingEnabled ? run?.userId : undefined,
-            // Copied settings:
-            ...this.settings,
-            // map to OpenAI API names:
-            stop: this.settings.stopSequences,
-            maxTokens: this.settings.maxCompletionTokens,
-            // other settings:
-            abortSignal: run?.abortSignal,
-            prompt,
-            responseFormat,
-        };
+        const api = this.settings.api ?? new OpenAIApiConfiguration();
+        const user = this.settings.isUserIdForwardingEnabled
+            ? options.run?.userId
+            : undefined;
+        const abortSignal = options.run?.abortSignal;
+        let { stopSequences } = this.settings;
+        const openaiResponseFormat = options.responseFormat;
         return callWithRetryAndThrottle({
-            retry: callSettings.api?.retry,
-            throttle: callSettings.api?.throttle,
-            call: async () => callOpenAICompletionAPI(callSettings),
+            retry: api.retry,
+            throttle: api.throttle,
+            call: async () => {
+                // empty arrays are not allowed for stop:
+                if (stopSequences != null &&
+                    Array.isArray(stopSequences) &&
+                    stopSequences.length === 0) {
+                    stopSequences = undefined;
+                }
+                return postJsonToApi({
+                    url: api.assembleUrl("/completions"),
+                    headers: api.headers,
+                    body: {
+                        stream: openaiResponseFormat.stream,
+                        model: this.settings.model,
+                        prompt,
+                        suffix: this.settings.suffix,
+                        max_tokens: this.settings.maxGenerationTokens,
+                        temperature: this.settings.temperature,
+                        top_p: this.settings.topP,
+                        n: this.settings.numberOfGenerations,
+                        logprobs: this.settings.logprobs,
+                        echo: this.settings.echo,
+                        stop: this.settings.stopSequences,
+                        seed: this.settings.seed,
+                        presence_penalty: this.settings.presencePenalty,
+                        frequency_penalty: this.settings.frequencyPenalty,
+                        best_of: this.settings.bestOf,
+                        logit_bias: this.settings.logitBias,
+                        user,
+                    },
+                    failedResponseHandler: failedOpenAICallResponseHandler,
+                    successfulResponseHandler: openaiResponseFormat.handler,
+                    abortSignal,
+                });
+            },
         });
     }
     get settingsForEvent() {
         const eventSettingProperties = [
-            "maxCompletionTokens",
+            "maxGenerationTokens",
             "stopSequences",
+            "numberOfGenerations",
             "suffix",
             "temperature",
             "topP",
-            "n",
             "logprobs",
             "echo",
             "presencePenalty",
@@ -211,14 +238,14 @@ export class OpenAICompletionModel extends AbstractModel {
         ];
         return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
     }
-    async doGenerateText(prompt, options) {
+    async doGenerateTexts(prompt, options) {
         const response = await this.callAPI(prompt, {
             ...options,
             responseFormat: OpenAITextResponseFormat.json,
         });
         return {
             response,
-            text: response.choices[0].text,
+            texts: response.choices.map((choice) => choice.text),
             usage: {
                 promptTokens: response.usage.prompt_tokens,
                 completionTokens: response.usage.completion_tokens,
@@ -280,38 +307,6 @@ const OpenAICompletionResponseSchema = z.object({
         total_tokens: z.number(),
     }),
 });
-async function callOpenAICompletionAPI({ api = new OpenAIApiConfiguration(), abortSignal, responseFormat, model, prompt, suffix, maxTokens, temperature, topP, n, logprobs, echo, stop, presencePenalty, frequencyPenalty, bestOf, logitBias, seed, user, }) {
-    // empty arrays are not allowed for stop:
-    if (stop != null && Array.isArray(stop) && stop.length === 0) {
-        stop = undefined;
-    }
-    return postJsonToApi({
-        url: api.assembleUrl("/completions"),
-        headers: api.headers,
-        body: {
-            stream: responseFormat.stream,
-            model,
-            prompt,
-            suffix,
-            max_tokens: maxTokens,
-            temperature,
-            top_p: topP,
-            n,
-            logprobs,
-            echo,
-            stop,
-            seed,
-            presence_penalty: presencePenalty,
-            frequency_penalty: frequencyPenalty,
-            best_of: bestOf,
-            logit_bias: logitBias,
-            user,
-        },
-        failedResponseHandler: failedOpenAICallResponseHandler,
-        successfulResponseHandler: responseFormat.handler,
-        abortSignal,
-    });
-}
 export const OpenAITextResponseFormat = {
     /**
      * Returns the response as a JSON object.

package/model-provider/openai/OpenAIFacade.cjs

@@ -17,7 +17,7 @@ const OpenAIChatModel_js_1 = require("./chat/OpenAIChatModel.cjs");
  * const model = openai.CompletionTextGenerator({
  *   model: "gpt-3.5-turbo-instruct",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -41,7 +41,7 @@ exports.CompletionTextGenerator = CompletionTextGenerator;
  * const model = openai.ChatTextGenerator({
  *   model: "gpt-3.5-turbo",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  * });
  *
  * const text = await generateText(

package/model-provider/openai/OpenAIFacade.d.ts

@@ -14,7 +14,7 @@ import { OpenAIChatModel, OpenAIChatSettings } from "./chat/OpenAIChatModel.js";
  * const model = openai.CompletionTextGenerator({
  *   model: "gpt-3.5-turbo-instruct",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -35,7 +35,7 @@ export declare function CompletionTextGenerator(settings: OpenAICompletionModelS
  * const model = openai.ChatTextGenerator({
  *   model: "gpt-3.5-turbo",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  * });
  *
  * const text = await generateText(

package/model-provider/openai/OpenAIFacade.js

@@ -14,7 +14,7 @@ import { OpenAIChatModel } from "./chat/OpenAIChatModel.js";
  * const model = openai.CompletionTextGenerator({
  *   model: "gpt-3.5-turbo-instruct",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -37,7 +37,7 @@ export function CompletionTextGenerator(settings) {
  * const model = openai.ChatTextGenerator({
  *   model: "gpt-3.5-turbo",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  * });
  *
  * const text = await generateText(

package/model-provider/openai/chat/AbstractOpenAIChatModel.cjs

@@ -20,38 +20,67 @@ class AbstractOpenAIChatModel extends AbstractModel_js_1.AbstractModel {
         super({ settings });
     }
     async callAPI(messages, options) {
+        const api = this.settings.api ?? new OpenAIApiConfiguration_js_1.OpenAIApiConfiguration();
+        const responseFormat = options.responseFormat;
+        const abortSignal = options.run?.abortSignal;
+        const user = this.settings.isUserIdForwardingEnabled
+            ? options.run?.userId
+            : undefined;
+        const openAIResponseFormat = this.settings.responseFormat;
+        // function & tool calling:
+        const functions = options.functions ?? this.settings.functions;
+        const functionCall = options.functionCall ?? this.settings.functionCall;
+        const tools = options.tools ?? this.settings.tools;
+        const toolChoice = options.toolChoice ?? this.settings.toolChoice;
+        let { stopSequences } = this.settings;
         return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
             retry: this.settings.api?.retry,
             throttle: this.settings.api?.throttle,
-            call: async () => callOpenAIChatCompletionAPI({
-                ...this.settings,
-                // function & tool calling:
-                functions: options.functions ?? this.settings.functions,
-                functionCall: options.functionCall ?? this.settings.functionCall,
-                tools: options.tools ?? this.settings.tools,
-                toolChoice: options.toolChoice ?? this.settings.toolChoice,
-                // map to OpenAI API names:
-                stop: this.settings.stopSequences,
-                maxTokens: this.settings.maxCompletionTokens,
-                openAIResponseFormat: this.settings.responseFormat,
-                // other settings:
-                user: this.settings.isUserIdForwardingEnabled
-                    ? options.run?.userId
-                    : undefined,
-                abortSignal: options.run?.abortSignal,
-                responseFormat: options.responseFormat,
-                messages,
-            }),
+            call: async () => {
+                // empty arrays are not allowed for stopSequences:
+                if (stopSequences != null &&
+                    Array.isArray(stopSequences) &&
+                    stopSequences.length === 0) {
+                    stopSequences = undefined;
+                }
+                return (0, postToApi_js_1.postJsonToApi)({
+                    url: api.assembleUrl("/chat/completions"),
+                    headers: api.headers,
+                    body: {
+                        stream: responseFormat.stream,
+                        model: this.settings.model,
+                        messages,
+                        functions,
+                        function_call: functionCall,
+                        tools,
+                        tool_choice: toolChoice,
+                        temperature: this.settings.temperature,
+                        top_p: this.settings.topP,
+                        n: this.settings.numberOfGenerations,
+                        stop: this.settings.stopSequences,
+                        max_tokens: this.settings.maxGenerationTokens,
+                        presence_penalty: this.settings.presencePenalty,
+                        frequency_penalty: this.settings.frequencyPenalty,
+                        logit_bias: this.settings.logitBias,
+                        seed: this.settings.seed,
+                        response_format: openAIResponseFormat,
+                        user,
+                    },
+                    failedResponseHandler: OpenAIError_js_1.failedOpenAICallResponseHandler,
+                    successfulResponseHandler: responseFormat.handler,
+                    abortSignal,
+                });
+            },
         });
     }
-    async doGenerateText(prompt, options) {
+    async doGenerateTexts(prompt, options) {
         const response = await this.callAPI(prompt, {
             ...options,
             responseFormat: exports.OpenAIChatResponseFormat.json,
         });
         return {
             response,
-            text: response.choices[0].message.content,
+            texts: response.choices.map((choice) => choice.message.content ?? ""),
             usage: this.extractUsage(response),
         };
     }
@@ -173,39 +202,6 @@ const openAIChatResponseSchema = zod_1.z.object({
         total_tokens: zod_1.z.number(),
     }),
 });
-async function callOpenAIChatCompletionAPI({ api = new OpenAIApiConfiguration_js_1.OpenAIApiConfiguration(), abortSignal, responseFormat, model, messages, functions, functionCall, tools, toolChoice, temperature, topP, n, stop, maxTokens, presencePenalty, frequencyPenalty, logitBias, user, openAIResponseFormat, seed, }) {
-    // empty arrays are not allowed for stop:
-    if (stop != null && Array.isArray(stop) && stop.length === 0) {
-        stop = undefined;
-    }
-    return (0, postToApi_js_1.postJsonToApi)({
-        url: api.assembleUrl("/chat/completions"),
-        headers: api.headers,
-        body: {
-            stream: responseFormat.stream,
-            model,
-            messages,
-            functions,
-            function_call: functionCall,
-            tools,
-            tool_choice: toolChoice,
-            temperature,
-            top_p: topP,
-            n,
-            stop,
-            max_tokens: maxTokens,
-            presence_penalty: presencePenalty,
-            frequency_penalty: frequencyPenalty,
-            logit_bias: logitBias,
-            seed,
-            response_format: openAIResponseFormat,
-            user,
-        },
-        failedResponseHandler: OpenAIError_js_1.failedOpenAICallResponseHandler,
-        successfulResponseHandler: responseFormat.handler,
-        abortSignal,
-    });
-}
 exports.OpenAIChatResponseFormat = {
     /**
      * Returns the response as a JSON object.