modelfusion 0.13.0 → 0.15.0

package/model-provider/llamacpp/LlamaCppTextGenerationModel.d.ts

@@ -3,8 +3,8 @@ import { AbstractModel } from "../../model-function/AbstractModel.js";
 import { FunctionOptions } from "../../model-function/FunctionOptions.js";
 import { DeltaEvent } from "../../model-function/generate-text/DeltaEvent.js";
 import { TextGenerationModel, TextGenerationModelSettings } from "../../model-function/generate-text/TextGenerationModel.js";
-import { PromptMapping } from "../../prompt/PromptMapping.js";
-import { PromptMappingTextGenerationModel } from "../../prompt/PromptMappingTextGenerationModel.js";
+import { PromptFormat } from "../../prompt/PromptFormat.js";
+import { PromptFormatTextGenerationModel } from "../../prompt/PromptFormatTextGenerationModel.js";
 import { RetryFunction } from "../../util/api/RetryFunction.js";
 import { ThrottleFunction } from "../../util/api/ThrottleFunction.js";
 import { ResponseHandler } from "../../util/api/postToApi.js";
@@ -25,9 +25,7 @@ export interface LlamaCppTextGenerationModelSettings<CONTEXT_WINDOW_SIZE extends
     temperature?: number;
     topK?: number;
     topP?: number;
-    nPredict?: number;
     nKeep?: number;
-    stop?: string[];
     tfsZ?: number;
     typicalP?: number;
     repeatPenalty?: number;
@@ -59,8 +57,8 @@ export declare class LlamaCppTextGenerationModel<CONTEXT_WINDOW_SIZE extends num
             model: string;
             stream: boolean;
             seed: number;
-            stop: string[];
             mirostat: number;
+            stop: string[];
             frequency_penalty: number;
             ignore_eos: boolean;
             logit_bias: number[];
@@ -102,11 +100,8 @@ export declare class LlamaCppTextGenerationModel<CONTEXT_WINDOW_SIZE extends num
     extractText(response: LlamaCppTextGenerationResponse): string;
     generateDeltaStreamResponse(prompt: string, options?: FunctionOptions<LlamaCppTextGenerationModelSettings<CONTEXT_WINDOW_SIZE>>): Promise<AsyncIterable<DeltaEvent<LlamaCppTextGenerationDelta>>>;
     extractTextDelta(fullDelta: LlamaCppTextGenerationDelta): string | undefined;
-    mapPrompt<INPUT_PROMPT>(promptMapping: PromptMapping<INPUT_PROMPT, string>): PromptMappingTextGenerationModel<INPUT_PROMPT, string, LlamaCppTextGenerationResponse, LlamaCppTextGenerationDelta, LlamaCppTextGenerationModelSettings<CONTEXT_WINDOW_SIZE>, this>;
+    withPromptFormat<INPUT_PROMPT>(promptFormat: PromptFormat<INPUT_PROMPT, string>): PromptFormatTextGenerationModel<INPUT_PROMPT, string, LlamaCppTextGenerationResponse, LlamaCppTextGenerationDelta, LlamaCppTextGenerationModelSettings<CONTEXT_WINDOW_SIZE>, this>;
     withSettings(additionalSettings: Partial<LlamaCppTextGenerationModelSettings<CONTEXT_WINDOW_SIZE>>): this;
-    get maxCompletionTokens(): number | undefined;
-    withMaxCompletionTokens(maxCompletionTokens: number): this;
-    withStopTokens(stopTokens: string[]): this;
 }
 declare const llamaCppTextGenerationResponseSchema: z.ZodObject<{
     content: z.ZodString;
@@ -139,8 +134,8 @@ declare const llamaCppTextGenerationResponseSchema: z.ZodObject<{
         model: string;
         stream: boolean;
         seed: number;
-        stop: string[];
         mirostat: number;
+        stop: string[];
         frequency_penalty: number;
         ignore_eos: boolean;
         logit_bias: number[];
@@ -163,8 +158,8 @@ declare const llamaCppTextGenerationResponseSchema: z.ZodObject<{
         model: string;
         stream: boolean;
         seed: number;
-        stop: string[];
         mirostat: number;
+        stop: string[];
         frequency_penalty: number;
         ignore_eos: boolean;
         logit_bias: number[];
@@ -231,8 +226,8 @@ declare const llamaCppTextGenerationResponseSchema: z.ZodObject<{
         model: string;
         stream: boolean;
         seed: number;
-        stop: string[];
         mirostat: number;
+        stop: string[];
         frequency_penalty: number;
         ignore_eos: boolean;
         logit_bias: number[];
@@ -279,8 +274,8 @@ declare const llamaCppTextGenerationResponseSchema: z.ZodObject<{
         model: string;
         stream: boolean;
         seed: number;
-        stop: string[];
         mirostat: number;
+        stop: string[];
         frequency_penalty: number;
         ignore_eos: boolean;
         logit_bias: number[];
@@ -344,8 +339,8 @@ export declare const LlamaCppTextGenerationResponseFormat: {
                 model: string;
                 stream: boolean;
                 seed: number;
-                stop: string[];
                 mirostat: number;
+                stop: string[];
                 frequency_penalty: number;
                 ignore_eos: boolean;
                 logit_bias: number[];

package/model-provider/llamacpp/LlamaCppTextGenerationModel.js

@@ -3,7 +3,7 @@ import z from "zod";
 import { AbstractModel } from "../../model-function/AbstractModel.js";
 import { AsyncQueue } from "../../model-function/generate-text/AsyncQueue.js";
 import { parseEventSourceReadableStream } from "../../model-function/generate-text/parseEventSourceReadableStream.js";
-import { PromptMappingTextGenerationModel } from "../../prompt/PromptMappingTextGenerationModel.js";
+import { PromptFormatTextGenerationModel } from "../../prompt/PromptFormatTextGenerationModel.js";
 import { callWithRetryAndThrottle } from "../../util/api/callWithRetryAndThrottle.js";
 import { createJsonResponseHandler, postJsonToApi, } from "../../util/api/postToApi.js";
 import { failedLlamaCppCallResponseHandler } from "./LlamaCppError.js";
@@ -37,11 +37,18 @@ export class LlamaCppTextGenerationModel extends AbstractModel {
     }
     async callAPI(prompt, options) {
         const { run, settings, responseFormat } = options;
-        const callSettings = Object.assign(this.settings, settings, {
+        const combinedSettings = {
+            ...this.settings,
+            ...settings,
+        };
+        const callSettings = {
+            ...combinedSettings,
+            nPredict: combinedSettings.maxCompletionTokens,
+            stop: combinedSettings.stopSequences,
             abortSignal: run?.abortSignal,
             prompt,
             responseFormat,
-        });
+        };
         return callWithRetryAndThrottle({
             retry: this.settings.retry,
             throttle: this.settings.throttle,
@@ -70,24 +77,17 @@ export class LlamaCppTextGenerationModel extends AbstractModel {
     extractTextDelta(fullDelta) {
         return fullDelta.delta;
     }
-    mapPrompt(promptMapping) {
-        return new PromptMappingTextGenerationModel({
-            model: this.withStopTokens(promptMapping.stopTokens),
-            promptMapping,
+    withPromptFormat(promptFormat) {
+        return new PromptFormatTextGenerationModel({
+            model: this.withSettings({
+                stopSequences: promptFormat.stopSequences,
+            }),
+            promptFormat,
         });
     }
     withSettings(additionalSettings) {
         return new LlamaCppTextGenerationModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.nPredict;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ nPredict: maxCompletionTokens });
-    }
-    withStopTokens(stopTokens) {
-        return this.withSettings({ stop: stopTokens });
-    }
 }
 const llamaCppTextGenerationResponseSchema = z.object({
     content: z.string(),

package/model-provider/openai/OpenAITextGenerationModel.cjs

@@ -10,7 +10,7 @@ const AbstractModel_js_1 = require("../../model-function/AbstractModel.cjs");
 const AsyncQueue_js_1 = require("../../model-function/generate-text/AsyncQueue.cjs");
 const parseEventSourceReadableStream_js_1 = require("../../model-function/generate-text/parseEventSourceReadableStream.cjs");
 const countTokens_js_1 = require("../../model-function/tokenize-text/countTokens.cjs");
-const PromptMappingTextGenerationModel_js_1 = require("../../prompt/PromptMappingTextGenerationModel.cjs");
+const PromptFormatTextGenerationModel_js_1 = require("../../prompt/PromptFormatTextGenerationModel.cjs");
 const callWithRetryAndThrottle_js_1 = require("../../util/api/callWithRetryAndThrottle.cjs");
 const postToApi_js_1 = require("../../util/api/postToApi.cjs");
 const OpenAIError_js_1 = require("./OpenAIError.cjs");
@@ -75,7 +75,7 @@ exports.calculateOpenAITextGenerationCostInMillicents = calculateOpenAITextGener
  * const model = new OpenAITextGenerationModel({
  *   model: "text-davinci-003",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -124,14 +124,23 @@ class OpenAITextGenerationModel extends AbstractModel_js_1.AbstractModel {
     }
     async callAPI(prompt, options) {
         const { run, settings, responseFormat } = options;
-        const callSettings = Object.assign({
+        const combinedSettings = {
+            ...this.settings,
+            ...settings,
+        };
+        const callSettings = {
             apiKey: this.apiKey,
             user: this.settings.isUserIdForwardingEnabled ? run?.userId : undefined,
-        }, this.settings, settings, {
+            // Copied settings:
+            ...combinedSettings,
+            // map to OpenAI API names:
+            stop: combinedSettings.stopSequences,
+            maxTokens: combinedSettings.maxCompletionTokens,
+            // other settings:
             abortSignal: run?.abortSignal,
             prompt,
             responseFormat,
-        });
+        };
         return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
             retry: callSettings.retry,
             throttle: callSettings.throttle,
@@ -156,24 +165,17 @@ class OpenAITextGenerationModel extends AbstractModel_js_1.AbstractModel {
     extractTextDelta(fullDelta) {
         return fullDelta[0].delta;
     }
-    mapPrompt(promptMapping) {
-        return new PromptMappingTextGenerationModel_js_1.PromptMappingTextGenerationModel({
-            model: this.withStopTokens(promptMapping.stopTokens),
-            promptMapping,
+    withPromptFormat(promptFormat) {
+        return new PromptFormatTextGenerationModel_js_1.PromptFormatTextGenerationModel({
+            model: this.withSettings({
+                stopSequences: promptFormat.stopSequences,
+            }),
+            promptFormat,
         });
     }
     withSettings(additionalSettings) {
         return new OpenAITextGenerationModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.maxTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ maxTokens: maxCompletionTokens });
-    }
-    withStopTokens(stopTokens) {
-        return this.withSettings({ stop: stopTokens });
-    }
 }
 exports.OpenAITextGenerationModel = OpenAITextGenerationModel;
 const openAITextGenerationResponseSchema = zod_1.default.object({

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

@@ -3,8 +3,8 @@ import { AbstractModel } from "../../model-function/AbstractModel.js";
 import { FunctionOptions } from "../../model-function/FunctionOptions.js";
 import { DeltaEvent } from "../../model-function/generate-text/DeltaEvent.js";
 import { TextGenerationModel, TextGenerationModelSettings } from "../../model-function/generate-text/TextGenerationModel.js";
-import { PromptMapping } from "../../prompt/PromptMapping.js";
-import { PromptMappingTextGenerationModel } from "../../prompt/PromptMappingTextGenerationModel.js";
+import { PromptFormat } from "../../prompt/PromptFormat.js";
+import { PromptFormatTextGenerationModel } from "../../prompt/PromptFormatTextGenerationModel.js";
 import { RetryFunction } from "../../util/api/RetryFunction.js";
 import { ThrottleFunction } from "../../util/api/ThrottleFunction.js";
 import { ResponseHandler } from "../../util/api/postToApi.js";
@@ -72,13 +72,11 @@ export interface OpenAITextGenerationModelSettings extends TextGenerationModelSe
     throttle?: ThrottleFunction;
     isUserIdForwardingEnabled?: boolean;
     suffix?: string;
-    maxTokens?: number;
     temperature?: number;
     topP?: number;
     n?: number;
     logprobs?: number;
     echo?: boolean;
-    stop?: string | string[];
     presencePenalty?: number;
     frequencyPenalty?: number;
     bestOf?: number;
@@ -92,7 +90,7 @@ export interface OpenAITextGenerationModelSettings extends TextGenerationModelSe
  * const model = new OpenAITextGenerationModel({
  *   model: "text-davinci-003",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -134,11 +132,8 @@ export declare class OpenAITextGenerationModel extends AbstractModel<OpenAITextG
     extractText(response: OpenAITextGenerationResponse): string;
     generateDeltaStreamResponse(prompt: string, options?: FunctionOptions<OpenAITextGenerationModelSettings>): Promise<AsyncIterable<DeltaEvent<OpenAITextGenerationDelta>>>;
     extractTextDelta(fullDelta: OpenAITextGenerationDelta): string | undefined;
-    mapPrompt<INPUT_PROMPT>(promptMapping: PromptMapping<INPUT_PROMPT, string>): PromptMappingTextGenerationModel<INPUT_PROMPT, string, OpenAITextGenerationResponse, OpenAITextGenerationDelta, OpenAITextGenerationModelSettings, this>;
+    withPromptFormat<INPUT_PROMPT>(promptFormat: PromptFormat<INPUT_PROMPT, string>): PromptFormatTextGenerationModel<INPUT_PROMPT, string, OpenAITextGenerationResponse, OpenAITextGenerationDelta, OpenAITextGenerationModelSettings, this>;
     withSettings(additionalSettings: Partial<OpenAITextGenerationModelSettings>): this;
-    get maxCompletionTokens(): number | undefined;
-    withMaxCompletionTokens(maxCompletionTokens: number): this;
-    withStopTokens(stopTokens: string[]): this;
 }
 declare const openAITextGenerationResponseSchema: z.ZodObject<{
     id: z.ZodString;

package/model-provider/openai/OpenAITextGenerationModel.js

@@ -4,7 +4,7 @@ import { AbstractModel } from "../../model-function/AbstractModel.js";
 import { AsyncQueue } from "../../model-function/generate-text/AsyncQueue.js";
 import { parseEventSourceReadableStream } from "../../model-function/generate-text/parseEventSourceReadableStream.js";
 import { countTokens } from "../../model-function/tokenize-text/countTokens.js";
-import { PromptMappingTextGenerationModel } from "../../prompt/PromptMappingTextGenerationModel.js";
+import { PromptFormatTextGenerationModel } from "../../prompt/PromptFormatTextGenerationModel.js";
 import { callWithRetryAndThrottle } from "../../util/api/callWithRetryAndThrottle.js";
 import { createJsonResponseHandler, postJsonToApi, } from "../../util/api/postToApi.js";
 import { failedOpenAICallResponseHandler } from "./OpenAIError.js";
@@ -67,7 +67,7 @@ export const calculateOpenAITextGenerationCostInMillicents = ({ model, response,
  * const model = new OpenAITextGenerationModel({
  *   model: "text-davinci-003",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -116,14 +116,23 @@ export class OpenAITextGenerationModel extends AbstractModel {
     }
     async callAPI(prompt, options) {
         const { run, settings, responseFormat } = options;
-        const callSettings = Object.assign({
+        const combinedSettings = {
+            ...this.settings,
+            ...settings,
+        };
+        const callSettings = {
             apiKey: this.apiKey,
             user: this.settings.isUserIdForwardingEnabled ? run?.userId : undefined,
-        }, this.settings, settings, {
+            // Copied settings:
+            ...combinedSettings,
+            // map to OpenAI API names:
+            stop: combinedSettings.stopSequences,
+            maxTokens: combinedSettings.maxCompletionTokens,
+            // other settings:
             abortSignal: run?.abortSignal,
             prompt,
             responseFormat,
-        });
+        };
         return callWithRetryAndThrottle({
             retry: callSettings.retry,
             throttle: callSettings.throttle,
@@ -148,24 +157,17 @@ export class OpenAITextGenerationModel extends AbstractModel {
     extractTextDelta(fullDelta) {
         return fullDelta[0].delta;
     }
-    mapPrompt(promptMapping) {
-        return new PromptMappingTextGenerationModel({
-            model: this.withStopTokens(promptMapping.stopTokens),
-            promptMapping,
+    withPromptFormat(promptFormat) {
+        return new PromptFormatTextGenerationModel({
+            model: this.withSettings({
+                stopSequences: promptFormat.stopSequences,
+            }),
+            promptFormat,
         });
     }
     withSettings(additionalSettings) {
         return new OpenAITextGenerationModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.maxTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ maxTokens: maxCompletionTokens });
-    }
-    withStopTokens(stopTokens) {
-        return this.withSettings({ stop: stopTokens });
-    }
 }
 const openAITextGenerationResponseSchema = z.object({
     id: z.string(),

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

@@ -7,7 +7,7 @@ exports.OpenAIChatResponseFormat = exports.OpenAIChatModel = exports.calculateOp
 const secure_json_parse_1 = __importDefault(require("secure-json-parse"));
 const zod_1 = __importDefault(require("zod"));
 const AbstractModel_js_1 = require("../../../model-function/AbstractModel.cjs");
-const PromptMappingTextGenerationModel_js_1 = require("../../../prompt/PromptMappingTextGenerationModel.cjs");
+const PromptFormatTextGenerationModel_js_1 = require("../../../prompt/PromptFormatTextGenerationModel.cjs");
 const callWithRetryAndThrottle_js_1 = require("../../../util/api/callWithRetryAndThrottle.cjs");
 const postToApi_js_1 = require("../../../util/api/postToApi.cjs");
 const OpenAIError_js_1 = require("../OpenAIError.cjs");
@@ -93,7 +93,7 @@ exports.calculateOpenAIChatCostInMillicents = calculateOpenAIChatCostInMillicent
  * const model = new OpenAIChatModel({
  *   model: "gpt-3.5-turbo",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  * });
  *
  * const text = await generateText([
@@ -150,14 +150,20 @@ class OpenAIChatModel extends AbstractModel_js_1.AbstractModel {
     }
     async callAPI(messages, options) {
         const { run, settings, responseFormat } = options;
-        const callSettings = Object.assign({
+        const combinedSettings = {
+            ...this.settings,
+            ...settings,
+        };
+        const callSettings = {
             apiKey: this.apiKey,
             user: this.settings.isUserIdForwardingEnabled ? run?.userId : undefined,
-        }, this.settings, settings, {
+            ...combinedSettings,
+            stop: combinedSettings.stopSequences,
+            maxTokens: combinedSettings.maxCompletionTokens,
             abortSignal: run?.abortSignal,
             messages,
             responseFormat,
-        });
+        };
         return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
             retry: callSettings.retry,
             throttle: callSettings.throttle,
@@ -205,24 +211,15 @@ class OpenAIChatModel extends AbstractModel_js_1.AbstractModel {
         const jsonText = response.choices[0].message.function_call.arguments;
         return secure_json_parse_1.default.parse(jsonText);
     }
-    mapPrompt(promptMapping) {
-        return new PromptMappingTextGenerationModel_js_1.PromptMappingTextGenerationModel({
-            model: this.withStopTokens(promptMapping.stopTokens),
-            promptMapping,
+    withPromptFormat(promptFormat) {
+        return new PromptFormatTextGenerationModel_js_1.PromptFormatTextGenerationModel({
+            model: this.withSettings({ stopSequences: promptFormat.stopSequences }),
+            promptFormat,
         });
     }
     withSettings(additionalSettings) {
         return new OpenAIChatModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.maxTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ maxTokens: maxCompletionTokens });
-    }
-    withStopTokens(stopTokens) {
-        return this.withSettings({ stop: stopTokens });
-    }
 }
 exports.OpenAIChatModel = OpenAIChatModel;
 const openAIChatResponseSchema = zod_1.default.object({

package/model-provider/openai/chat/OpenAIChatModel.d.ts

@@ -5,8 +5,8 @@ import { GenerateJsonModel } from "../../../model-function/generate-json/Generat
 import { GenerateJsonOrTextModel } from "../../../model-function/generate-json/GenerateJsonOrTextModel.js";
 import { DeltaEvent } from "../../../model-function/generate-text/DeltaEvent.js";
 import { TextGenerationModel, TextGenerationModelSettings } from "../../../model-function/generate-text/TextGenerationModel.js";
-import { PromptMapping } from "../../../prompt/PromptMapping.js";
-import { PromptMappingTextGenerationModel } from "../../../prompt/PromptMappingTextGenerationModel.js";
+import { PromptFormat } from "../../../prompt/PromptFormat.js";
+import { PromptFormatTextGenerationModel } from "../../../prompt/PromptFormatTextGenerationModel.js";
 import { ResponseHandler } from "../../../util/api/postToApi.js";
 import { OpenAIModelSettings } from "../OpenAIModelSettings.js";
 import { TikTokenTokenizer } from "../TikTokenTokenizer.js";
@@ -95,7 +95,7 @@ export interface OpenAIChatCallSettings {
     presencePenalty?: number;
     frequencyPenalty?: number;
 }
-export interface OpenAIChatSettings extends TextGenerationModelSettings, OpenAIModelSettings, OpenAIChatCallSettings {
+export interface OpenAIChatSettings extends TextGenerationModelSettings, OpenAIModelSettings, Omit<OpenAIChatCallSettings, "stop" | "maxTokens"> {
     isUserIdForwardingEnabled?: boolean;
 }
 /**
@@ -107,7 +107,7 @@ export interface OpenAIChatSettings extends TextGenerationModelSettings, OpenAIM
  * const model = new OpenAIChatModel({
  *   model: "gpt-3.5-turbo",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  * });
  *
  * const text = await generateText([
@@ -170,11 +170,8 @@ export declare class OpenAIChatModel extends AbstractModel<OpenAIChatSettings> i
      */
     generateJsonResponse(prompt: OpenAIChatSingleFunctionPrompt<unknown> | OpenAIChatAutoFunctionPrompt<Array<OpenAIFunctionDescription<unknown>>>, options?: FunctionOptions<OpenAIChatSettings> | undefined): PromiseLike<OpenAIChatResponse>;
     extractJson(response: OpenAIChatResponse): unknown;
-    mapPrompt<INPUT_PROMPT>(promptMapping: PromptMapping<INPUT_PROMPT, OpenAIChatMessage[]>): PromptMappingTextGenerationModel<INPUT_PROMPT, OpenAIChatMessage[], OpenAIChatResponse, OpenAIChatDelta, OpenAIChatSettings, this>;
+    withPromptFormat<INPUT_PROMPT>(promptFormat: PromptFormat<INPUT_PROMPT, OpenAIChatMessage[]>): PromptFormatTextGenerationModel<INPUT_PROMPT, OpenAIChatMessage[], OpenAIChatResponse, OpenAIChatDelta, OpenAIChatSettings, this>;
     withSettings(additionalSettings: Partial<OpenAIChatSettings>): this;
-    get maxCompletionTokens(): number | undefined;
-    withMaxCompletionTokens(maxCompletionTokens: number): this;
-    withStopTokens(stopTokens: string[]): this;
 }
 declare const openAIChatResponseSchema: z.ZodObject<{
     id: z.ZodString;

package/model-provider/openai/chat/OpenAIChatModel.js

@@ -1,7 +1,7 @@
 import SecureJSON from "secure-json-parse";
 import z from "zod";
 import { AbstractModel } from "../../../model-function/AbstractModel.js";
-import { PromptMappingTextGenerationModel } from "../../../prompt/PromptMappingTextGenerationModel.js";
+import { PromptFormatTextGenerationModel } from "../../../prompt/PromptFormatTextGenerationModel.js";
 import { callWithRetryAndThrottle } from "../../../util/api/callWithRetryAndThrottle.js";
 import { createJsonResponseHandler, postJsonToApi, } from "../../../util/api/postToApi.js";
 import { failedOpenAICallResponseHandler } from "../OpenAIError.js";
@@ -85,7 +85,7 @@ export const calculateOpenAIChatCostInMillicents = ({ model, response, }) => res
  * const model = new OpenAIChatModel({
  *   model: "gpt-3.5-turbo",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  * });
  *
  * const text = await generateText([
@@ -142,14 +142,20 @@ export class OpenAIChatModel extends AbstractModel {
     }
     async callAPI(messages, options) {
         const { run, settings, responseFormat } = options;
-        const callSettings = Object.assign({
+        const combinedSettings = {
+            ...this.settings,
+            ...settings,
+        };
+        const callSettings = {
             apiKey: this.apiKey,
             user: this.settings.isUserIdForwardingEnabled ? run?.userId : undefined,
-        }, this.settings, settings, {
+            ...combinedSettings,
+            stop: combinedSettings.stopSequences,
+            maxTokens: combinedSettings.maxCompletionTokens,
             abortSignal: run?.abortSignal,
             messages,
             responseFormat,
-        });
+        };
         return callWithRetryAndThrottle({
             retry: callSettings.retry,
             throttle: callSettings.throttle,
@@ -197,24 +203,15 @@ export class OpenAIChatModel extends AbstractModel {
         const jsonText = response.choices[0].message.function_call.arguments;
         return SecureJSON.parse(jsonText);
     }
-    mapPrompt(promptMapping) {
-        return new PromptMappingTextGenerationModel({
-            model: this.withStopTokens(promptMapping.stopTokens),
-            promptMapping,
+    withPromptFormat(promptFormat) {
+        return new PromptFormatTextGenerationModel({
+            model: this.withSettings({ stopSequences: promptFormat.stopSequences }),
+            promptFormat,
         });
     }
     withSettings(additionalSettings) {
         return new OpenAIChatModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.maxTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ maxTokens: maxCompletionTokens });
-    }
-    withStopTokens(stopTokens) {
-        return this.withSettings({ stop: stopTokens });
-    }
 }
 const openAIChatResponseSchema = z.object({
     id: z.string(),

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "modelfusion",
   "description": "Build AI applications, chatbots, and agents with JavaScript and TypeScript.",
-  "version": "0.13.0",
+  "version": "0.15.0",
   "author": "Lars Grammel",
   "license": "MIT",
   "keywords": [
@@ -67,11 +67,11 @@
     "eslint": "^8.45.0",
     "eslint-config-prettier": "9.0.0",
     "husky": "^8.0.3",
-    "lint-staged": "14.0.0",
+    "lint-staged": "14.0.1",
     "prettier": "3.0.2",
     "rimraf": "5.0.1",
     "typescript": "5.1.6",
-    "zod": "3.22.1",
+    "zod": "3.22.2",
     "zod-to-json-schema": "3.21.4"
   },
   "peerDependencies": {

package/prompt/{AlpacaPromptMapping.cjs → AlpacaPromptFormat.cjs}

@@ -1,19 +1,19 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.InstructionToAlpacaPromptMapping = void 0;
+exports.AlpacaInstructionPromptFormat = void 0;
 const DEFAULT_SYSTEM_PROMPT_INPUT = "Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.";
 const DEFAULT_SYSTEM_PROMPT_NO_INPUT = "Below is an instruction that describes a task. Write a response that appropriately completes the request.";
 /**
- * Maps an instruction prompt to the Alpaca prompt format.
+ * Formats an instruction prompt as an Alpaca prompt.
  *
  * If the instruction has a system prompt, it overrides the default system prompt
  * (which can impact the results, because the model may be trained on the default system prompt).
  *
  * @see https://github.com/tatsu-lab/stanford_alpaca#data-release
  */
-const InstructionToAlpacaPromptMapping = () => ({
-    stopTokens: [],
-    map: (instruction) => {
+const AlpacaInstructionPromptFormat = () => ({
+    stopSequences: [],
+    format: (instruction) => {
         let text = instruction.system ??
             (instruction.input != null
                 ? DEFAULT_SYSTEM_PROMPT_INPUT
@@ -30,4 +30,4 @@ const InstructionToAlpacaPromptMapping = () => ({
         return text;
     },
 });
-exports.InstructionToAlpacaPromptMapping = InstructionToAlpacaPromptMapping;
+exports.AlpacaInstructionPromptFormat = AlpacaInstructionPromptFormat;

package/prompt/{AlpacaPromptMapping.d.ts → AlpacaPromptFormat.d.ts}

@@ -1,11 +1,11 @@
 import { InstructionPrompt } from "./InstructionPrompt.js";
-import { PromptMapping } from "./PromptMapping.js";
+import { PromptFormat } from "./PromptFormat.js";
 /**
- * Maps an instruction prompt to the Alpaca prompt format.
+ * Formats an instruction prompt as an Alpaca prompt.
  *
  * If the instruction has a system prompt, it overrides the default system prompt
  * (which can impact the results, because the model may be trained on the default system prompt).
  *
  * @see https://github.com/tatsu-lab/stanford_alpaca#data-release
  */
-export declare const InstructionToAlpacaPromptMapping: () => PromptMapping<InstructionPrompt, string>;
+export declare const AlpacaInstructionPromptFormat: () => PromptFormat<InstructionPrompt, string>;

package/prompt/{AlpacaPromptMapping.js → AlpacaPromptFormat.js}

@@ -1,16 +1,16 @@
 const DEFAULT_SYSTEM_PROMPT_INPUT = "Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.";
 const DEFAULT_SYSTEM_PROMPT_NO_INPUT = "Below is an instruction that describes a task. Write a response that appropriately completes the request.";
 /**
- * Maps an instruction prompt to the Alpaca prompt format.
+ * Formats an instruction prompt as an Alpaca prompt.
  *
  * If the instruction has a system prompt, it overrides the default system prompt
  * (which can impact the results, because the model may be trained on the default system prompt).
  *
  * @see https://github.com/tatsu-lab/stanford_alpaca#data-release
  */
-export const InstructionToAlpacaPromptMapping = () => ({
-    stopTokens: [],
-    map: (instruction) => {
+export const AlpacaInstructionPromptFormat = () => ({
+    stopSequences: [],
+    format: (instruction) => {
         let text = instruction.system ??
             (instruction.input != null
                 ? DEFAULT_SYSTEM_PROMPT_INPUT