modelfusion 0.65.0 → 0.66.1

package/model-function/generate-text/streamText.cjs

@@ -3,8 +3,11 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.streamText = void 0;
 const executeStreamCall_js_1 = require("../executeStreamCall.cjs");
 async function streamText(model, prompt, options) {
+    const shouldTrimWhitespace = model.settings.trimWhitespace ?? true;
     let accumulatedText = "";
     let lastFullDelta;
+    let isFirstDelta = true;
+    let trailingWhitespace = "";
     const fullResponse = await (0, executeStreamCall_js_1.executeStreamCall)({
         functionType: "stream-text",
         input: prompt,
@@ -13,8 +16,25 @@ async function streamText(model, prompt, options) {
         startStream: async (options) => model.doStreamText(prompt, options),
         processDelta: (delta) => {
             lastFullDelta = delta.fullDelta;
-            const textDelta = delta.valueDelta;
+            let textDelta = delta.valueDelta;
             if (textDelta != null && textDelta.length > 0) {
+                if (shouldTrimWhitespace) {
+                    if (isFirstDelta) {
+                        // remove leading whitespace:
+                        textDelta = textDelta.trimStart();
+                    }
+                    else {
+                        // restore trailing whitespace from previous chunk:
+                        textDelta = trailingWhitespace + textDelta;
+                    }
+                    // trim trailing whitespace and store it for the next chunk:
+                    const trailingWhitespaceMatch = textDelta.match(/\s+$/);
+                    trailingWhitespace = trailingWhitespaceMatch
+                        ? trailingWhitespaceMatch[0]
+                        : "";
+                    textDelta = textDelta.trimEnd();
+                }
+                isFirstDelta = false;
                 accumulatedText += textDelta;
                 return textDelta;
             }

package/model-function/generate-text/streamText.js

@@ -1,7 +1,10 @@
 import { executeStreamCall } from "../executeStreamCall.js";
 export async function streamText(model, prompt, options) {
+    const shouldTrimWhitespace = model.settings.trimWhitespace ?? true;
     let accumulatedText = "";
     let lastFullDelta;
+    let isFirstDelta = true;
+    let trailingWhitespace = "";
     const fullResponse = await executeStreamCall({
         functionType: "stream-text",
         input: prompt,
@@ -10,8 +13,25 @@ export async function streamText(model, prompt, options) {
         startStream: async (options) => model.doStreamText(prompt, options),
         processDelta: (delta) => {
             lastFullDelta = delta.fullDelta;
-            const textDelta = delta.valueDelta;
+            let textDelta = delta.valueDelta;
             if (textDelta != null && textDelta.length > 0) {
+                if (shouldTrimWhitespace) {
+                    if (isFirstDelta) {
+                        // remove leading whitespace:
+                        textDelta = textDelta.trimStart();
+                    }
+                    else {
+                        // restore trailing whitespace from previous chunk:
+                        textDelta = trailingWhitespace + textDelta;
+                    }
+                    // trim trailing whitespace and store it for the next chunk:
+                    const trailingWhitespaceMatch = textDelta.match(/\s+$/);
+                    trailingWhitespace = trailingWhitespaceMatch
+                        ? trailingWhitespaceMatch[0]
+                        : "";
+                    textDelta = textDelta.trimEnd();
+                }
+                isFirstDelta = false;
                 accumulatedText += textDelta;
                 return textDelta;
             }

package/model-provider/openai/OpenAICompletionModel.cjs

@@ -212,6 +212,7 @@ class OpenAICompletionModel extends AbstractModel_js_1.AbstractModel {
             "frequencyPenalty",
             "bestOf",
             "logitBias",
+            "seed",
         ];
         return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
     }
@@ -266,22 +267,26 @@ class OpenAICompletionModel extends AbstractModel_js_1.AbstractModel {
 exports.OpenAICompletionModel = OpenAICompletionModel;
 const OpenAICompletionResponseSchema = zod_1.z.object({
     id: zod_1.z.string(),
-    object: zod_1.z.literal("text_completion"),
-    created: zod_1.z.number(),
-    model: zod_1.z.string(),
     choices: zod_1.z.array(zod_1.z.object({
-        text: zod_1.z.string(),
+        finish_reason: zod_1.z
+            .enum(["stop", "length", "content_filter"])
+            .optional()
+            .nullable(),
         index: zod_1.z.number(),
         logprobs: zod_1.z.nullable(zod_1.z.any()),
-        finish_reason: zod_1.z.string(),
+        text: zod_1.z.string(),
     })),
+    created: zod_1.z.number(),
+    model: zod_1.z.string(),
+    system_fingerprint: zod_1.z.string().optional(),
+    object: zod_1.z.literal("text_completion"),
     usage: zod_1.z.object({
         prompt_tokens: zod_1.z.number(),
         completion_tokens: zod_1.z.number(),
         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, user, }) {
+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;
@@ -301,6 +306,7 @@ async function callOpenAICompletionAPI({ api = new OpenAIApiConfiguration_js_1.O
             logprobs,
             echo,
             stop,
+            seed,
             presence_penalty: presencePenalty,
             frequency_penalty: frequencyPenalty,
             best_of: bestOf,
@@ -332,13 +338,17 @@ exports.OpenAITextResponseFormat = {
 const textResponseStreamEventSchema = zod_1.z.object({
     choices: zod_1.z.array(zod_1.z.object({
         text: zod_1.z.string(),
-        finish_reason: zod_1.z.enum(["stop", "length"]).nullable(),
+        finish_reason: zod_1.z
+            .enum(["stop", "length", "content_filter"])
+            .optional()
+            .nullable(),
         index: zod_1.z.number(),
     })),
     created: zod_1.z.number(),
     id: zod_1.z.string(),
     model: zod_1.z.string(),
-    object: zod_1.z.string(),
+    system_fingerprint: zod_1.z.string().optional(),
+    object: zod_1.z.literal("text_completion"),
 });
 async function createOpenAITextFullDeltaIterableQueue(stream) {
     const queue = new AsyncQueue_js_1.AsyncQueue();

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

@@ -112,6 +112,7 @@ export interface OpenAICompletionCallSettings {
     frequencyPenalty?: number;
     bestOf?: number;
     logitBias?: Record<number, number>;
+    seed?: number | null;
 }
 export interface OpenAICompletionModelSettings extends TextGenerationModelSettings, Omit<OpenAICompletionCallSettings, "stop" | "maxTokens"> {
     isUserIdForwardingEnabled?: boolean;
@@ -157,11 +158,12 @@ export declare class OpenAICompletionModel extends AbstractModel<OpenAICompletio
             id: string;
             choices: {
                 text: string;
-                finish_reason: string;
                 index: number;
+                finish_reason?: "length" | "stop" | "content_filter" | null | undefined;
                 logprobs?: any;
             }[];
             created: number;
+            system_fingerprint?: string | undefined;
         };
         text: string;
         usage: {
@@ -187,25 +189,26 @@ export declare class OpenAICompletionModel extends AbstractModel<OpenAICompletio
 }
 declare const OpenAICompletionResponseSchema: z.ZodObject<{
     id: z.ZodString;
-    object: z.ZodLiteral<"text_completion">;
-    created: z.ZodNumber;
-    model: z.ZodString;
     choices: z.ZodArray<z.ZodObject<{
-        text: z.ZodString;
+        finish_reason: z.ZodNullable<z.ZodOptional<z.ZodEnum<["stop", "length", "content_filter"]>>>;
         index: z.ZodNumber;
         logprobs: z.ZodNullable<z.ZodAny>;
-        finish_reason: z.ZodString;
+        text: z.ZodString;
     }, "strip", z.ZodTypeAny, {
         text: string;
-        finish_reason: string;
         index: number;
+        finish_reason?: "length" | "stop" | "content_filter" | null | undefined;
         logprobs?: any;
     }, {
         text: string;
-        finish_reason: string;
         index: number;
+        finish_reason?: "length" | "stop" | "content_filter" | null | undefined;
         logprobs?: any;
     }>, "many">;
+    created: z.ZodNumber;
+    model: z.ZodString;
+    system_fingerprint: z.ZodOptional<z.ZodString>;
+    object: z.ZodLiteral<"text_completion">;
     usage: z.ZodObject<{
         prompt_tokens: z.ZodNumber;
         completion_tokens: z.ZodNumber;
@@ -230,11 +233,12 @@ declare const OpenAICompletionResponseSchema: z.ZodObject<{
     id: string;
     choices: {
         text: string;
-        finish_reason: string;
         index: number;
+        finish_reason?: "length" | "stop" | "content_filter" | null | undefined;
         logprobs?: any;
     }[];
     created: number;
+    system_fingerprint?: string | undefined;
 }, {
     object: "text_completion";
     usage: {
@@ -246,11 +250,12 @@ declare const OpenAICompletionResponseSchema: z.ZodObject<{
     id: string;
     choices: {
         text: string;
-        finish_reason: string;
         index: number;
+        finish_reason?: "length" | "stop" | "content_filter" | null | undefined;
         logprobs?: any;
     }[];
     created: number;
+    system_fingerprint?: string | undefined;
 }>;
 export type OpenAICompletionResponse = z.infer<typeof OpenAICompletionResponseSchema>;
 export type OpenAITextResponseFormatType<T> = {
@@ -274,11 +279,12 @@ export declare const OpenAITextResponseFormat: {
             id: string;
             choices: {
                 text: string;
-                finish_reason: string;
                 index: number;
+                finish_reason?: "length" | "stop" | "content_filter" | null | undefined;
                 logprobs?: any;
             }[];
             created: number;
+            system_fingerprint?: string | undefined;
         }>;
     };
     /**

package/model-provider/openai/OpenAICompletionModel.js

@@ -206,6 +206,7 @@ export class OpenAICompletionModel extends AbstractModel {
             "frequencyPenalty",
             "bestOf",
             "logitBias",
+            "seed",
         ];
         return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
     }
@@ -259,22 +260,26 @@ export class OpenAICompletionModel extends AbstractModel {
 }
 const OpenAICompletionResponseSchema = z.object({
     id: z.string(),
-    object: z.literal("text_completion"),
-    created: z.number(),
-    model: z.string(),
     choices: z.array(z.object({
-        text: z.string(),
+        finish_reason: z
+            .enum(["stop", "length", "content_filter"])
+            .optional()
+            .nullable(),
         index: z.number(),
         logprobs: z.nullable(z.any()),
-        finish_reason: z.string(),
+        text: z.string(),
     })),
+    created: z.number(),
+    model: z.string(),
+    system_fingerprint: z.string().optional(),
+    object: z.literal("text_completion"),
     usage: z.object({
         prompt_tokens: z.number(),
         completion_tokens: z.number(),
         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, user, }) {
+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;
@@ -294,6 +299,7 @@ async function callOpenAICompletionAPI({ api = new OpenAIApiConfiguration(), abo
             logprobs,
             echo,
             stop,
+            seed,
             presence_penalty: presencePenalty,
             frequency_penalty: frequencyPenalty,
             best_of: bestOf,
@@ -325,13 +331,17 @@ export const OpenAITextResponseFormat = {
 const textResponseStreamEventSchema = z.object({
     choices: z.array(z.object({
         text: z.string(),
-        finish_reason: z.enum(["stop", "length"]).nullable(),
+        finish_reason: z
+            .enum(["stop", "length", "content_filter"])
+            .optional()
+            .nullable(),
         index: z.number(),
     })),
     created: z.number(),
     id: z.string(),
     model: z.string(),
-    object: z.string(),
+    system_fingerprint: z.string().optional(),
+    object: z.literal("text_completion"),
 });
 async function createOpenAITextFullDeltaIterableQueue(stream) {
     const queue = new AsyncQueue();

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

@@ -39,6 +39,8 @@ exports.OPENAI_CHAT_MODELS = {
         contextWindowSize: 8192,
         promptTokenCostInMillicents: 3,
         completionTokenCostInMillicents: 6,
+        fineTunedPromptTokenCostInMillicents: null,
+        fineTunedCompletionTokenCostInMillicents: null,
     },
     "gpt-4-1106-preview": {
         contextWindowSize: 128000,
@@ -115,7 +117,7 @@ function getOpenAIChatModelInformation(model) {
     // Extract the base model from the fine-tuned model:
     // eslint-disable-next-line @typescript-eslint/no-unused-vars
     const [_, baseModel, ___, ____, _____] = model.split(":");
-    if (["gpt-3.5-turbo", "gpt-3.5-turbo-0613"].includes(baseModel)) {
+    if (["gpt-3.5-turbo", "gpt-3.5-turbo-0613", "gpt-4-0613"].includes(baseModel)) {
         const baseModelInformation = exports.OPENAI_CHAT_MODELS[baseModel];
         return {
             baseModel: baseModel,
@@ -133,11 +135,14 @@ const isOpenAIChatModel = (model) => model in exports.OPENAI_CHAT_MODELS ||
     model.startsWith("ft:gpt-3.5-turbo:");
 exports.isOpenAIChatModel = isOpenAIChatModel;
 const calculateOpenAIChatCostInMillicents = ({ model, response, }) => {
-    const modelInformation = getOpenAIChatModelInformation(model);
-    return (response.usage.prompt_tokens *
-        modelInformation.promptTokenCostInMillicents +
-        response.usage.completion_tokens *
-            modelInformation.completionTokenCostInMillicents);
+    const { promptTokenCostInMillicents, completionTokenCostInMillicents } = getOpenAIChatModelInformation(model);
+    // null: when cost is unknown, e.g. for fine-tuned models where the price is not yet known
+    if (promptTokenCostInMillicents == null ||
+        completionTokenCostInMillicents == null) {
+        return null;
+    }
+    return (response.usage.prompt_tokens * promptTokenCostInMillicents +
+        response.usage.completion_tokens * completionTokenCostInMillicents);
 };
 exports.calculateOpenAIChatCostInMillicents = calculateOpenAIChatCostInMillicents;
 /**
@@ -234,6 +239,8 @@ class OpenAIChatModel extends AbstractModel_js_1.AbstractModel {
             "presencePenalty",
             "frequencyPenalty",
             "logitBias",
+            "seed",
+            "responseFormat",
         ];
         return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
     }
@@ -388,10 +395,6 @@ class OpenAIChatModel extends AbstractModel_js_1.AbstractModel {
 exports.OpenAIChatModel = OpenAIChatModel;
 const openAIChatResponseSchema = zod_1.z.object({
     id: zod_1.z.string(),
-    object: zod_1.z.literal("chat.completion"),
-    created: zod_1.z.number(),
-    model: zod_1.z.string(),
-    system_fingerprint: zod_1.z.string(),
     choices: zod_1.z.array(zod_1.z.object({
         message: zod_1.z.object({
             role: zod_1.z.literal("assistant"),
@@ -405,14 +408,21 @@ const openAIChatResponseSchema = zod_1.z.object({
         }),
         index: zod_1.z.number(),
         logprobs: zod_1.z.nullable(zod_1.z.any()),
-        finish_reason: zod_1.z.enum([
+        finish_reason: zod_1.z
+            .enum([
             "stop",
             "length",
             "tool_calls",
             "content_filter",
             "function_call",
-        ]),
+        ])
+            .optional()
+            .nullable(),
     })),
+    created: zod_1.z.number(),
+    model: zod_1.z.string(),
+    system_fingerprint: zod_1.z.string().optional(),
+    object: zod_1.z.literal("chat.completion"),
     usage: zod_1.z.object({
         prompt_tokens: zod_1.z.number(),
         completion_tokens: zod_1.z.number(),

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

@@ -27,6 +27,8 @@ export declare const OPENAI_CHAT_MODELS: {
         contextWindowSize: number;
         promptTokenCostInMillicents: number;
         completionTokenCostInMillicents: number;
+        fineTunedPromptTokenCostInMillicents: null;
+        fineTunedCompletionTokenCostInMillicents: null;
     };
     "gpt-4-1106-preview": {
         contextWindowSize: number;
@@ -92,10 +94,10 @@ export declare function getOpenAIChatModelInformation(model: OpenAIChatModelType
     baseModel: OpenAIChatBaseModelType;
     isFineTuned: boolean;
     contextWindowSize: number;
-    promptTokenCostInMillicents: number;
-    completionTokenCostInMillicents: number;
+    promptTokenCostInMillicents: number | null;
+    completionTokenCostInMillicents: number | null;
 };
-type FineTuneableOpenAIChatModelType = `gpt-3.5-turbo` | `gpt-3.5-turbo-0613`;
+type FineTuneableOpenAIChatModelType = `gpt-3.5-turbo` | `gpt-3.5-turbo-0613` | `gpt-4-0613`;
 type FineTunedOpenAIChatModelType = `ft:${FineTuneableOpenAIChatModelType}:${string}:${string}:${string}`;
 export type OpenAIChatBaseModelType = keyof typeof OPENAI_CHAT_MODELS;
 export type OpenAIChatModelType = OpenAIChatBaseModelType | FineTunedOpenAIChatModelType;
@@ -103,7 +105,7 @@ export declare const isOpenAIChatModel: (model: string) => model is OpenAIChatMo
 export declare const calculateOpenAIChatCostInMillicents: ({ model, response, }: {
     model: OpenAIChatModelType;
     response: OpenAIChatResponse;
-}) => number;
+}) => number | null;
 export interface OpenAIChatCallSettings {
     api?: ApiConfiguration;
     model: OpenAIChatModelType;
@@ -193,12 +195,12 @@ export declare class OpenAIChatModel extends AbstractModel<OpenAIChatSettings> i
                         arguments: string;
                     } | undefined;
                 };
-                finish_reason: "length" | "stop" | "function_call" | "tool_calls" | "content_filter";
                 index: number;
                 logprobs?: any;
+                finish_reason?: "length" | "stop" | "function_call" | "tool_calls" | "content_filter" | null | undefined;
             }[];
             created: number;
-            system_fingerprint: string;
+            system_fingerprint?: string | undefined;
         };
         text: string;
         usage: {
@@ -234,12 +236,12 @@ export declare class OpenAIChatModel extends AbstractModel<OpenAIChatSettings> i
                         arguments: string;
                     } | undefined;
                 };
-                finish_reason: "length" | "stop" | "function_call" | "tool_calls" | "content_filter";
                 index: number;
                 logprobs?: any;
+                finish_reason?: "length" | "stop" | "function_call" | "tool_calls" | "content_filter" | null | undefined;
             }[];
             created: number;
-            system_fingerprint: string;
+            system_fingerprint?: string | undefined;
         };
         valueText: string;
         value: any;
@@ -269,12 +271,12 @@ export declare class OpenAIChatModel extends AbstractModel<OpenAIChatSettings> i
                         arguments: string;
                     } | undefined;
                 };
-                finish_reason: "length" | "stop" | "function_call" | "tool_calls" | "content_filter";
                 index: number;
                 logprobs?: any;
+                finish_reason?: "length" | "stop" | "function_call" | "tool_calls" | "content_filter" | null | undefined;
             }[];
             created: number;
-            system_fingerprint: string;
+            system_fingerprint?: string | undefined;
         };
         structureAndText: {
             structure: null;
@@ -306,12 +308,12 @@ export declare class OpenAIChatModel extends AbstractModel<OpenAIChatSettings> i
                         arguments: string;
                     } | undefined;
                 };
-                finish_reason: "length" | "stop" | "function_call" | "tool_calls" | "content_filter";
                 index: number;
                 logprobs?: any;
+                finish_reason?: "length" | "stop" | "function_call" | "tool_calls" | "content_filter" | null | undefined;
             }[];
             created: number;
-            system_fingerprint: string;
+            system_fingerprint?: string | undefined;
         };
         structureAndText: {
             structure: string;
@@ -343,10 +345,6 @@ export declare class OpenAIChatModel extends AbstractModel<OpenAIChatSettings> i
 }
 declare const openAIChatResponseSchema: z.ZodObject<{
     id: z.ZodString;
-    object: z.ZodLiteral<"chat.completion">;
-    created: z.ZodNumber;
-    model: z.ZodString;
-    system_fingerprint: z.ZodString;
     choices: z.ZodArray<z.ZodObject<{
         message: z.ZodObject<{
             role: z.ZodLiteral<"assistant">;
@@ -378,7 +376,7 @@ declare const openAIChatResponseSchema: z.ZodObject<{
         }>;
         index: z.ZodNumber;
         logprobs: z.ZodNullable<z.ZodAny>;
-        finish_reason: z.ZodEnum<["stop", "length", "tool_calls", "content_filter", "function_call"]>;
+        finish_reason: z.ZodNullable<z.ZodOptional<z.ZodEnum<["stop", "length", "tool_calls", "content_filter", "function_call"]>>>;
     }, "strip", z.ZodTypeAny, {
         message: {
             content: string | null;
@@ -388,9 +386,9 @@ declare const openAIChatResponseSchema: z.ZodObject<{
                 arguments: string;
             } | undefined;
         };
-        finish_reason: "length" | "stop" | "function_call" | "tool_calls" | "content_filter";
         index: number;
         logprobs?: any;
+        finish_reason?: "length" | "stop" | "function_call" | "tool_calls" | "content_filter" | null | undefined;
     }, {
         message: {
             content: string | null;
@@ -400,10 +398,14 @@ declare const openAIChatResponseSchema: z.ZodObject<{
                 arguments: string;
             } | undefined;
         };
-        finish_reason: "length" | "stop" | "function_call" | "tool_calls" | "content_filter";
         index: number;
         logprobs?: any;
+        finish_reason?: "length" | "stop" | "function_call" | "tool_calls" | "content_filter" | null | undefined;
     }>, "many">;
+    created: z.ZodNumber;
+    model: z.ZodString;
+    system_fingerprint: z.ZodOptional<z.ZodString>;
+    object: z.ZodLiteral<"chat.completion">;
     usage: z.ZodObject<{
         prompt_tokens: z.ZodNumber;
         completion_tokens: z.ZodNumber;
@@ -435,12 +437,12 @@ declare const openAIChatResponseSchema: z.ZodObject<{
                 arguments: string;
             } | undefined;
         };
-        finish_reason: "length" | "stop" | "function_call" | "tool_calls" | "content_filter";
         index: number;
         logprobs?: any;
+        finish_reason?: "length" | "stop" | "function_call" | "tool_calls" | "content_filter" | null | undefined;
     }[];
     created: number;
-    system_fingerprint: string;
+    system_fingerprint?: string | undefined;
 }, {
     object: "chat.completion";
     usage: {
@@ -459,12 +461,12 @@ declare const openAIChatResponseSchema: z.ZodObject<{
                 arguments: string;
             } | undefined;
         };
-        finish_reason: "length" | "stop" | "function_call" | "tool_calls" | "content_filter";
         index: number;
         logprobs?: any;
+        finish_reason?: "length" | "stop" | "function_call" | "tool_calls" | "content_filter" | null | undefined;
     }[];
     created: number;
-    system_fingerprint: string;
+    system_fingerprint?: string | undefined;
 }>;
 export type OpenAIChatResponse = z.infer<typeof openAIChatResponseSchema>;
 export type OpenAIChatResponseFormatType<T> = {
@@ -495,12 +497,12 @@ export declare const OpenAIChatResponseFormat: {
                         arguments: string;
                     } | undefined;
                 };
-                finish_reason: "length" | "stop" | "function_call" | "tool_calls" | "content_filter";
                 index: number;
                 logprobs?: any;
+                finish_reason?: "length" | "stop" | "function_call" | "tool_calls" | "content_filter" | null | undefined;
             }[];
             created: number;
-            system_fingerprint: string;
+            system_fingerprint?: string | undefined;
         }>;
     };
     /**

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

@@ -33,6 +33,8 @@ export const OPENAI_CHAT_MODELS = {
         contextWindowSize: 8192,
         promptTokenCostInMillicents: 3,
         completionTokenCostInMillicents: 6,
+        fineTunedPromptTokenCostInMillicents: null,
+        fineTunedCompletionTokenCostInMillicents: null,
     },
     "gpt-4-1106-preview": {
         contextWindowSize: 128000,
@@ -109,7 +111,7 @@ export function getOpenAIChatModelInformation(model) {
     // Extract the base model from the fine-tuned model:
     // eslint-disable-next-line @typescript-eslint/no-unused-vars
     const [_, baseModel, ___, ____, _____] = model.split(":");
-    if (["gpt-3.5-turbo", "gpt-3.5-turbo-0613"].includes(baseModel)) {
+    if (["gpt-3.5-turbo", "gpt-3.5-turbo-0613", "gpt-4-0613"].includes(baseModel)) {
         const baseModelInformation = OPENAI_CHAT_MODELS[baseModel];
         return {
             baseModel: baseModel,
@@ -125,11 +127,14 @@ export const isOpenAIChatModel = (model) => model in OPENAI_CHAT_MODELS ||
     model.startsWith("ft:gpt-3.5-turbo-0613:") ||
     model.startsWith("ft:gpt-3.5-turbo:");
 export const calculateOpenAIChatCostInMillicents = ({ model, response, }) => {
-    const modelInformation = getOpenAIChatModelInformation(model);
-    return (response.usage.prompt_tokens *
-        modelInformation.promptTokenCostInMillicents +
-        response.usage.completion_tokens *
-            modelInformation.completionTokenCostInMillicents);
+    const { promptTokenCostInMillicents, completionTokenCostInMillicents } = getOpenAIChatModelInformation(model);
+    // null: when cost is unknown, e.g. for fine-tuned models where the price is not yet known
+    if (promptTokenCostInMillicents == null ||
+        completionTokenCostInMillicents == null) {
+        return null;
+    }
+    return (response.usage.prompt_tokens * promptTokenCostInMillicents +
+        response.usage.completion_tokens * completionTokenCostInMillicents);
 };
 /**
  * Create a text generation model that calls the OpenAI chat completion API.
@@ -225,6 +230,8 @@ export class OpenAIChatModel extends AbstractModel {
             "presencePenalty",
             "frequencyPenalty",
             "logitBias",
+            "seed",
+            "responseFormat",
         ];
         return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
     }
@@ -378,10 +385,6 @@ export class OpenAIChatModel extends AbstractModel {
 }
 const openAIChatResponseSchema = z.object({
     id: z.string(),
-    object: z.literal("chat.completion"),
-    created: z.number(),
-    model: z.string(),
-    system_fingerprint: z.string(),
     choices: z.array(z.object({
         message: z.object({
             role: z.literal("assistant"),
@@ -395,14 +398,21 @@ const openAIChatResponseSchema = z.object({
         }),
         index: z.number(),
         logprobs: z.nullable(z.any()),
-        finish_reason: z.enum([
+        finish_reason: z
+            .enum([
             "stop",
             "length",
             "tool_calls",
             "content_filter",
             "function_call",
-        ]),
+        ])
+            .optional()
+            .nullable(),
     })),
+    created: z.number(),
+    model: z.string(),
+    system_fingerprint: z.string().optional(),
+    object: z.literal("chat.completion"),
     usage: z.object({
         prompt_tokens: z.number(),
         completion_tokens: z.number(),

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

@@ -6,6 +6,7 @@ const AsyncQueue_js_1 = require("../../../util/AsyncQueue.cjs");
 const parseEventSourceStream_js_1 = require("../../../util/streaming/parseEventSourceStream.cjs");
 const parseJSON_js_1 = require("../../../util/parseJSON.cjs");
 const chatResponseStreamEventSchema = zod_1.z.object({
+    id: zod_1.z.string(),
     choices: zod_1.z.array(zod_1.z.object({
         delta: zod_1.z.object({
             role: zod_1.z.enum(["assistant", "user"]).optional(),
@@ -30,9 +31,9 @@ const chatResponseStreamEventSchema = zod_1.z.object({
         index: zod_1.z.number(),
     })),
     created: zod_1.z.number(),
-    id: zod_1.z.string(),
     model: zod_1.z.string(),
-    object: zod_1.z.string(),
+    system_fingerprint: zod_1.z.string().optional(),
+    object: zod_1.z.literal("chat.completion.chunk"),
 });
 async function createOpenAIChatDeltaIterableQueue(stream, extractDeltaValue) {
     const queue = new AsyncQueue_js_1.AsyncQueue();

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

@@ -3,6 +3,7 @@ import { AsyncQueue } from "../../../util/AsyncQueue.js";
 import { parseEventSourceStream } from "../../../util/streaming/parseEventSourceStream.js";
 import { safeParseJsonWithZod } from "../../../util/parseJSON.js";
 const chatResponseStreamEventSchema = z.object({
+    id: z.string(),
     choices: z.array(z.object({
         delta: z.object({
             role: z.enum(["assistant", "user"]).optional(),
@@ -27,9 +28,9 @@ const chatResponseStreamEventSchema = z.object({
         index: z.number(),
     })),
     created: z.number(),
-    id: z.string(),
     model: z.string(),
-    object: z.string(),
+    system_fingerprint: z.string().optional(),
+    object: z.literal("chat.completion.chunk"),
 });
 export async function createOpenAIChatDeltaIterableQueue(stream, extractDeltaValue) {
     const queue = new AsyncQueue();

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "modelfusion",
   "description": "Build multimodal applications, chatbots, and agents with JavaScript and TypeScript.",
-  "version": "0.65.0",
+  "version": "0.66.1",
   "author": "Lars Grammel",
   "license": "MIT",
   "keywords": [