peerbench 0.0.1 → 0.0.2-alpha.0

package/dist/llm-DNj_tp2T.d.ts

@@ -0,0 +1,22 @@
+import { A as AbstractProvider, P as ProviderResponse } from './provider-BDjGp2y-.js';
+import { ResponseFormatText, ResponseFormatJSONSchema, ResponseFormatJSONObject } from 'openai/resources/shared';
+import { ChatCompletionMessageParam } from 'openai/resources/chat/completions';
+
+declare abstract class AbstractLLMProvider extends AbstractProvider {
+    abstract forward(args: LLMProviderForwardArgs): Promise<ChatResponse>;
+}
+type LLMProviderForwardArgs = {
+    messages: ChatCompletionMessageParam[];
+    model: string;
+    abortSignal?: AbortSignal;
+    temperature?: number;
+    responseFormat?: ResponseFormatText | ResponseFormatJSONSchema | ResponseFormatJSONObject;
+};
+type ChatResponse = ProviderResponse<string> & {
+    inputTokensUsed?: number;
+    outputTokensUsed?: number;
+    inputCost?: string;
+    outputCost?: string;
+};
+
+export { AbstractLLMProvider as A, type ChatResponse as C, type LLMProviderForwardArgs as L };

package/dist/llm-judge-DIG1f1Az.d.ts

@@ -0,0 +1,67 @@
+import { A as AbstractScorer, B as BaseScorerResult } from './abstract-Dec9Sc5O.js';
+import { A as AbstractLLMProvider } from './llm-DNj_tp2T.js';
+import { R as RateLimiter } from './rate-limiter-CSmVIRsM.js';
+
+type MCQScorerParams = {
+    response: string;
+    choices: Record<string, string>;
+    correctAnswers: string[];
+};
+declare class MCQScorer extends AbstractScorer {
+    readonly kind = "mcq";
+    private regexScorer;
+    score(params: MCQScorerParams): Promise<BaseScorerResult & {
+        extractedAnswers: string[];
+    }>;
+    private buildPatternsForAnswer;
+}
+
+declare class LLMJudgeScorer extends AbstractScorer {
+    readonly kind = "llmJudge";
+    private provider;
+    constructor(provider: AbstractLLMProvider);
+    score(params: LLMJudgeScorerParams): Promise<LLMJudgeScorerResult | null>;
+}
+type LLMJudgeCriterion = {
+    id: string;
+    description: string;
+    weight?: number;
+    scale?: {
+        min: number;
+        max: number;
+    };
+};
+type LLMJudgeScorerResult = BaseScorerResult & {
+    provider: string;
+    inputTokensUsed?: number;
+    outputTokensUsed?: number;
+    inputCost?: string;
+    outputCost?: string;
+    verdict?: "pass" | "borderline" | "fail";
+};
+type LLMJudgeScorerParams = {
+    task: string;
+    candidateAnswer: string;
+    referenceAnswer?: string;
+    model: string;
+    /**
+     * The rubric used for judging (defaults to a generic set).
+     */
+    criteria?: LLMJudgeCriterion[];
+    /**
+     * Optional extra context that the judge can use (constraints, references, etc.).
+     */
+    meta?: Record<string, unknown>;
+    /**
+     * Optional rate limiter wrapper for provider calls.
+     */
+    rateLimiter?: RateLimiter;
+    /**
+     * Optional prompt tweaks.
+     */
+    systemPrompt?: string;
+    promptPrefix?: string;
+    promptSuffix?: string;
+};
+
+export { LLMJudgeScorer as L, type MCQScorerParams as M, MCQScorer as a, type LLMJudgeCriterion as b, type LLMJudgeScorerResult as c, type LLMJudgeScorerParams as d };

package/dist/provider-BDjGp2y-.d.ts

@@ -0,0 +1,10 @@
+declare abstract class AbstractProvider {
+    abstract readonly kind: string;
+}
+type ProviderResponse<TData = unknown> = {
+    startedAt: number;
+    completedAt: number;
+    data: TData;
+};
+
+export { AbstractProvider as A, type ProviderResponse as P };

package/dist/providers/index.d.ts

@@ -0,0 +1,72 @@
+import { A as AbstractLLMProvider, L as LLMProviderForwardArgs, C as ChatResponse } from '../llm-DNj_tp2T.js';
+export { A as AbstractProvider, P as ProviderResponse } from '../provider-BDjGp2y-.js';
+import { R as RateLimiter } from '../rate-limiter-CSmVIRsM.js';
+import { ChatCompletionMessageParam } from 'openai/resources/chat/completions';
+import { ResponseFormatText, ResponseFormatJSONSchema, ResponseFormatJSONObject } from 'openai/resources/shared';
+import { ChatCompletionMessageParam as ChatCompletionMessageParam$1 } from 'openai/resources/index';
+
+declare class MastraProvider extends AbstractLLMProvider {
+    readonly kind = "mastra";
+    private readonly endpoint;
+    private readonly authToken;
+    private readonly agentName;
+    private client?;
+    private clientInitPromise?;
+    private warnedAboutSystemMessages;
+    private warnedAboutResponseFormat;
+    constructor(params: {
+        endpoint: string;
+        authToken: string;
+        agentName: string;
+    });
+    private getClient;
+    forward(args: LLMProviderForwardArgs): Promise<ChatResponse>;
+}
+
+declare class OpenAIProvider extends AbstractLLMProvider {
+    readonly kind = "openai";
+    private client;
+    private rateLimiter;
+    private maxRetries;
+    constructor(config: {
+        apiKey: string;
+        baseURL: string;
+        maxRetries?: number;
+        timeout?: number;
+        rateLimiter?: RateLimiter;
+    });
+    forward(args: {
+        messages: ChatCompletionMessageParam[];
+        model: string;
+        abortSignal?: AbortSignal;
+        temperature?: number;
+        responseFormat?: ResponseFormatText | ResponseFormatJSONSchema | ResponseFormatJSONObject;
+    }): Promise<ChatResponse>;
+}
+
+declare class OpenRouterProvider extends OpenAIProvider {
+    private models;
+    private modelsCachePromise;
+    private modelsUpdatedAt;
+    constructor(config: {
+        apiKey: string;
+        maxRetries?: number;
+        timeout?: number;
+        rateLimiter?: RateLimiter;
+    });
+    forward(args: {
+        messages: ChatCompletionMessageParam$1[];
+        model: string;
+        abortSignal?: AbortSignal;
+        temperature?: number;
+        responseFormat?: ResponseFormatText | ResponseFormatJSONSchema | ResponseFormatJSONObject;
+    }): Promise<ChatResponse>;
+    /**
+     * Updates the cache that holds information about OpenRouter models
+     * including pricing information. It will be valid for 24 hours as
+     * long as the instance of this Provider object is alive.
+     */
+    private updateModelsCache;
+}
+
+export { AbstractLLMProvider, ChatResponse, LLMProviderForwardArgs, MastraProvider, OpenAIProvider, OpenRouterProvider };

package/dist/providers/index.js

@@ -0,0 +1,263 @@
+import {
+  RateLimiter
+} from "../chunk-DUBKY73H.js";
+import "../chunk-PZ5AY32C.js";
+
+// src/providers/abstract/provider.ts
+var AbstractProvider = class {
+};
+
+// src/providers/abstract/llm.ts
+var AbstractLLMProvider = class extends AbstractProvider {
+};
+
+// src/providers/mastra.ts
+var MastraProvider = class extends AbstractLLMProvider {
+  kind = "mastra";
+  endpoint;
+  authToken;
+  agentName;
+  client;
+  clientInitPromise;
+  warnedAboutSystemMessages = false;
+  warnedAboutResponseFormat = false;
+  constructor(params) {
+    super();
+    this.endpoint = params.endpoint;
+    this.authToken = params.authToken;
+    this.agentName = params.agentName;
+  }
+  async getClient() {
+    if (this.client) return this.client;
+    if (this.clientInitPromise) return this.clientInitPromise;
+    this.clientInitPromise = import("@mastra/client-js").then((mod) => {
+      const MastraClient = mod?.MastraClient;
+      if (!MastraClient) {
+        throw new Error(`@mastra/client-js did not export MastraClient`);
+      }
+      this.client = new MastraClient({
+        baseUrl: this.endpoint,
+        headers: {
+          Authorization: `Bearer ${this.authToken}`
+        }
+      });
+      return this.client;
+    }).catch((err) => {
+      this.clientInitPromise = void 0;
+      throw new Error(
+        `MastraProvider requires optional peer dependency "@mastra/client-js" to be installed. Install it in your host application: npm i @mastra/client-js. Underlying error: ${err instanceof Error ? err.message : String(err)}`
+      );
+    });
+    return this.clientInitPromise;
+  }
+  async forward(args) {
+    const startedAt = Date.now();
+    if (!this.warnedAboutSystemMessages && args.messages.some((m) => m.role === "system")) {
+      this.warnedAboutSystemMessages = true;
+      console.warn(
+        `Mastra provider: system messages are ignored (agent "${this.agentName}" has baked-in prompts).`
+      );
+    }
+    if (!this.warnedAboutResponseFormat && args.responseFormat) {
+      this.warnedAboutResponseFormat = true;
+      console.warn(
+        `Mastra provider: responseFormat is ignored (configure structured output in the Mastra agent).`
+      );
+    }
+    const apiMessages = args.messages.filter((m) => m.role === "user" || m.role === "assistant").map((m) => ({
+      role: m.role,
+      content: String(m.content ?? "")
+    }));
+    const generateOptions = { messages: apiMessages };
+    const parts = String(args.model ?? "").split("/");
+    if (parts.length === 2) {
+      const [providerId, modelId] = parts;
+      generateOptions.runtimeContext = {
+        "provider-id": providerId,
+        "model-id": modelId
+      };
+    } else if (args.model && args.model !== this.agentName) {
+      generateOptions.runtimeContext = {
+        "model-id": args.model
+      };
+    }
+    const client = await this.getClient();
+    const agent = client.getAgent(this.agentName);
+    const requestOptions = args.abortSignal ? { signal: args.abortSignal } : void 0;
+    const response = await agent.generate(generateOptions, requestOptions);
+    const completedAt = Date.now();
+    const text = response?.text ?? "";
+    return {
+      data: String(text),
+      startedAt,
+      completedAt
+    };
+  }
+};
+
+// src/providers/openai.ts
+import OpenAI, { APIError } from "openai";
+var OpenAIProvider = class extends AbstractLLMProvider {
+  kind = "openai";
+  client;
+  rateLimiter;
+  maxRetries;
+  constructor(config) {
+    super();
+    this.maxRetries = config.maxRetries ?? 3;
+    this.rateLimiter = config.rateLimiter ?? new RateLimiter({
+      maxWeight: 20,
+      timeWindow: 3e3
+    });
+    this.client = new OpenAI({
+      baseURL: config.baseURL,
+      apiKey: config.apiKey,
+      timeout: config.timeout,
+      dangerouslyAllowBrowser: true
+    });
+  }
+  async forward(args) {
+    let retryCount = this.maxRetries;
+    while (retryCount > 0) {
+      let startedAt = /* @__PURE__ */ new Date();
+      try {
+        const response = await this.rateLimiter.execute(
+          async () => {
+            startedAt = /* @__PURE__ */ new Date();
+            return await this.client.chat.completions.create(
+              {
+                model: args.model,
+                messages: args.messages,
+                temperature: args.temperature,
+                response_format: args.responseFormat
+              },
+              // Signal for request
+              { signal: args.abortSignal }
+            );
+          },
+          // Signal for rate limiting
+          { signal: args.abortSignal }
+        );
+        if ("error" in response) {
+          const err = response.error;
+          throw new Error(
+            `${err.message} - Code ${err.code} - ${JSON.stringify(err)}`
+          );
+        }
+        if (!response?.choices?.[0]?.message?.content) {
+          throw new Error("No content returned from the model");
+        }
+        return {
+          data: response.choices[0].message.content,
+          inputTokensUsed: response?.usage?.prompt_tokens,
+          outputTokensUsed: response?.usage?.completion_tokens,
+          startedAt: startedAt.getTime(),
+          completedAt: Date.now()
+        };
+      } catch (err) {
+        if (err instanceof APIError && err.status === 401) {
+          throw new Error(`Invalid credentials provided`, { cause: err });
+        }
+        retryCount--;
+        if (err instanceof SyntaxError) {
+          console.debug(err);
+          continue;
+        }
+        if (retryCount !== 0) {
+          continue;
+        }
+        throw new Error(
+          `Failed to forward prompt to the model: ${err instanceof Error ? err.message : err}`,
+          { cause: err }
+        );
+      }
+    }
+    throw new Error(
+      `Failed to forward prompt to the model: Max retries reached`,
+      { cause: new Error("Max retries reached") }
+    );
+  }
+};
+
+// src/providers/openrouter.ts
+import axios from "axios";
+import Decimal from "decimal.js";
+var baseURL = "https://openrouter.ai/api/v1";
+var MODELS_CACHE_TTL = 1e3 * 60 * 60 * 24;
+var OpenRouterProvider = class extends OpenAIProvider {
+  models = void 0;
+  modelsCachePromise = Promise.resolve(void 0);
+  modelsUpdatedAt = 0;
+  constructor(config) {
+    super({
+      baseURL,
+      apiKey: config.apiKey,
+      maxRetries: config.maxRetries,
+      timeout: config.timeout,
+      rateLimiter: config.rateLimiter
+    });
+  }
+  async forward(args) {
+    const [response] = await Promise.all([
+      super.forward(args),
+      this.updateModelsCache().catch(() => {
+      })
+    ]);
+    const modelInfo = this.models?.data.find((m) => m.id === args.model);
+    let inputCost = void 0;
+    let outputCost = void 0;
+    if (modelInfo !== void 0) {
+      if (response.inputTokensUsed !== void 0) {
+        inputCost = new Decimal(modelInfo.pricing.prompt).mul(response.inputTokensUsed).toFixed(10);
+      }
+      if (response.outputTokensUsed !== void 0) {
+        outputCost = new Decimal(modelInfo.pricing.completion).mul(response.outputTokensUsed).toFixed(10);
+      }
+    }
+    return {
+      ...response,
+      inputCost,
+      outputCost
+    };
+  }
+  /**
+   * Updates the cache that holds information about OpenRouter models
+   * including pricing information. It will be valid for 24 hours as
+   * long as the instance of this Provider object is alive.
+   */
+  async updateModelsCache() {
+    this.modelsCachePromise = this.modelsCachePromise.then(async () => {
+      if (
+        // The data presented in the cache
+        this.models !== void 0 && // The cache is still valid
+        Date.now() - this.modelsUpdatedAt < MODELS_CACHE_TTL
+      ) {
+        return this.models;
+      }
+      return axios.get(`${baseURL}/models`).then((res) => res.data).then((data) => {
+        data = {
+          data: data.data.filter(
+            (m) => m.architecture.input_modalities.includes("text") && m.architecture.output_modalities.includes("text") && // These models are "fast apply model" and don't support multi turn conversations so don't include them
+            ![
+              "morph/morph-v3-large",
+              "morph/morph-v3-fast",
+              "relace/relace-apply-3"
+            ].includes(m.id)
+          )
+        };
+        this.models = data;
+        this.modelsUpdatedAt = Date.now();
+        return data;
+      });
+    }).catch(() => void 0);
+    await this.modelsCachePromise;
+  }
+};
+export {
+  AbstractLLMProvider,
+  AbstractProvider,
+  MastraProvider,
+  OpenAIProvider,
+  OpenRouterProvider
+};
+//# sourceMappingURL=index.js.map

package/dist/providers/index.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../src/providers/abstract/provider.ts","../../src/providers/abstract/llm.ts","../../src/providers/mastra.ts","../../src/providers/openai.ts","../../src/providers/openrouter.ts"],"sourcesContent":["export abstract class AbstractProvider {\n  abstract readonly kind: string;\n}\n\nexport type ProviderResponse<TData = unknown> = {\n  startedAt: number;\n  completedAt: number;\n  data: TData;\n};\n","import { AbstractProvider, ProviderResponse } from \"./provider\";\nimport {\n  ResponseFormatJSONObject,\n  ResponseFormatJSONSchema,\n  ResponseFormatText,\n} from \"openai/resources/shared\";\nimport { ChatCompletionMessageParam } from \"openai/resources/chat/completions\";\n\nexport abstract class AbstractLLMProvider extends AbstractProvider {\n  abstract forward(args: LLMProviderForwardArgs): Promise<ChatResponse>;\n}\n\nexport type LLMProviderForwardArgs = {\n  messages: ChatCompletionMessageParam[];\n  model: string;\n  abortSignal?: AbortSignal;\n  temperature?: number;\n  responseFormat?:\n    | ResponseFormatText\n    | ResponseFormatJSONSchema\n    | ResponseFormatJSONObject;\n};\n\nexport type ChatResponse = ProviderResponse<string> & {\n  inputTokensUsed?: number;\n  outputTokensUsed?: number;\n  inputCost?: string;\n  outputCost?: string;\n};\n","import {\n  AbstractLLMProvider,\n  type ChatResponse,\n  type LLMProviderForwardArgs,\n} from \"./abstract/llm\";\n\nexport class MastraProvider extends AbstractLLMProvider {\n  override readonly kind = \"mastra\";\n\n  private readonly endpoint: string;\n  private readonly authToken: string;\n  private readonly agentName: string;\n  private client?: any;\n  private clientInitPromise?: Promise<any>;\n  private warnedAboutSystemMessages = false;\n  private warnedAboutResponseFormat = false;\n\n  constructor(params: { endpoint: string; authToken: string; agentName: string }) {\n    super();\n    this.endpoint = params.endpoint;\n    this.authToken = params.authToken;\n    this.agentName = params.agentName;\n  }\n\n  private async getClient(): Promise<any> {\n    if (this.client) return this.client;\n    if (this.clientInitPromise) return this.clientInitPromise;\n\n    this.clientInitPromise = import(\"@mastra/client-js\")\n      .then((mod) => {\n        const MastraClient = (mod as any)?.MastraClient;\n        if (!MastraClient) {\n          throw new Error(`@mastra/client-js did not export MastraClient`);\n        }\n\n        this.client = new MastraClient({\n          baseUrl: this.endpoint,\n          headers: {\n            Authorization: `Bearer ${this.authToken}`,\n          },\n        });\n\n        return this.client;\n      })\n      .catch((err) => {\n        this.clientInitPromise = undefined;\n        throw new Error(\n          `MastraProvider requires optional peer dependency \"@mastra/client-js\" to be installed. ` +\n            `Install it in your host application: npm i @mastra/client-js. ` +\n            `Underlying error: ${err instanceof Error ? err.message : String(err)}`\n        );\n      });\n\n    return this.clientInitPromise;\n  }\n\n  override async forward(args: LLMProviderForwardArgs): Promise<ChatResponse> {\n    const startedAt = Date.now();\n\n    if (!this.warnedAboutSystemMessages && args.messages.some((m) => m.role === \"system\")) {\n      this.warnedAboutSystemMessages = true;\n      console.warn(\n        `Mastra provider: system messages are ignored (agent \"${this.agentName}\" has baked-in prompts).`\n      );\n    }\n\n    if (!this.warnedAboutResponseFormat && args.responseFormat) {\n      this.warnedAboutResponseFormat = true;\n      console.warn(\n        `Mastra provider: responseFormat is ignored (configure structured output in the Mastra agent).`\n      );\n    }\n\n    const apiMessages = args.messages\n      .filter((m) => m.role === \"user\" || m.role === \"assistant\")\n      .map((m) => ({\n        role: m.role as \"user\" | \"assistant\",\n        content: String((m as any).content ?? \"\"),\n      }));\n\n    const generateOptions: any = { messages: apiMessages };\n\n    // Allow runtime model override via runtimeContext.\n    // Model format: \"provider/modelId\" (e.g., \"openai/gpt-4o\", \"anthropic/claude-3.5-sonnet\").\n    const parts = String(args.model ?? \"\").split(\"/\");\n    if (parts.length === 2) {\n      const [providerId, modelId] = parts;\n      generateOptions.runtimeContext = {\n        \"provider-id\": providerId,\n        \"model-id\": modelId,\n      };\n    } else if (args.model && args.model !== this.agentName) {\n      generateOptions.runtimeContext = {\n        \"model-id\": args.model,\n      };\n    }\n\n    const client = await this.getClient();\n    const agent = client.getAgent(this.agentName);\n    const requestOptions = args.abortSignal ? ({ signal: args.abortSignal } as any) : undefined;\n    const response = await agent.generate(generateOptions, requestOptions);\n\n    const completedAt = Date.now();\n    const text = (response as any)?.text ?? \"\";\n\n    return {\n      data: String(text),\n      startedAt,\n      completedAt,\n    };\n  }\n}\n","import { RateLimiter } from \"@/utils\";\nimport { ChatCompletionMessageParam } from \"openai/resources/chat/completions\";\nimport {\n  ResponseFormatJSONObject,\n  ResponseFormatJSONSchema,\n  ResponseFormatText,\n} from \"openai/resources/shared\";\nimport OpenAI, { APIError } from \"openai\";\nimport { AbstractLLMProvider, ChatResponse } from \"./abstract/llm\";\n\nexport class OpenAIProvider extends AbstractLLMProvider {\n  override readonly kind = \"openai\";\n\n  private client: OpenAI;\n  private rateLimiter: RateLimiter;\n  private maxRetries: number;\n\n  constructor(config: {\n    apiKey: string;\n    baseURL: string;\n    maxRetries?: number;\n    timeout?: number;\n    rateLimiter?: RateLimiter;\n  }) {\n    super();\n    this.maxRetries = config.maxRetries ?? 3;\n    this.rateLimiter =\n      config.rateLimiter ??\n      new RateLimiter({\n        maxWeight: 20,\n        timeWindow: 3_000,\n      });\n\n    this.client = new OpenAI({\n      baseURL: config.baseURL,\n      apiKey: config.apiKey,\n      timeout: config.timeout,\n      dangerouslyAllowBrowser: true,\n    });\n  }\n\n  async forward(args: {\n    messages: ChatCompletionMessageParam[];\n    model: string;\n    abortSignal?: AbortSignal;\n    temperature?: number;\n    responseFormat?:\n      | ResponseFormatText\n      | ResponseFormatJSONSchema\n      | ResponseFormatJSONObject;\n  }): Promise<ChatResponse> {\n    let retryCount = this.maxRetries;\n    while (retryCount > 0) {\n      let startedAt: Date = new Date();\n\n      try {\n        const response = await this.rateLimiter.execute(\n          async () => {\n            // Capture the start time of the request\n            startedAt = new Date();\n            return await this.client.chat.completions.create(\n              {\n                model: args.model,\n                messages: args.messages,\n                temperature: args.temperature,\n                response_format: args.responseFormat,\n              },\n              // Signal for request\n              { signal: args.abortSignal }\n            );\n          },\n          // Signal for rate limiting\n          { signal: args.abortSignal }\n        );\n\n        if (\"error\" in response) {\n          const err = response.error as any;\n          throw new Error(\n            `${err.message} - Code ${err.code} - ${JSON.stringify(err)}`\n          );\n        }\n\n        if (!response?.choices?.[0]?.message?.content) {\n          throw new Error(\"No content returned from the model\");\n        }\n\n        return {\n          data: response.choices[0].message.content,\n\n          inputTokensUsed: response?.usage?.prompt_tokens,\n          outputTokensUsed: response?.usage?.completion_tokens,\n\n          startedAt: startedAt.getTime(),\n          completedAt: Date.now(),\n        };\n      } catch (err) {\n        if (err instanceof APIError && err.status === 401) {\n          throw new Error(`Invalid credentials provided`, { cause: err });\n        }\n\n        retryCount--;\n\n        // More likely an empty HTTP response returned by the Provider\n        // and it couldn't be parsed as JSON by the OpenAI SDK. We need to retry the request\n        // More info can be found in the following links:\n        // https://www.reddit.com/r/SillyTavernAI/comments/1ik95vr/deepseek_r1_on_openrouter_returning_blank_messages/\n        // https://github.com/cline/cline/issues/60\n        if (err instanceof SyntaxError) {\n          console.debug(err);\n          continue;\n        }\n\n        // If it was another error, just continue until we run out of retries\n        if (retryCount !== 0) {\n          continue;\n        }\n\n        throw new Error(\n          `Failed to forward prompt to the model: ${err instanceof Error ? err.message : err}`,\n          { cause: err }\n        );\n      }\n    }\n\n    throw new Error(\n      `Failed to forward prompt to the model: Max retries reached`,\n      { cause: new Error(\"Max retries reached\") }\n    );\n  }\n}\n","import { ChatResponse } from \"./abstract/llm\";\nimport { RateLimiter } from \"@/utils\";\nimport { OpenAIProvider } from \"./openai\";\nimport { ChatCompletionMessageParam } from \"openai/resources/index\";\nimport {\n  ResponseFormatText,\n  ResponseFormatJSONSchema,\n  ResponseFormatJSONObject,\n} from \"openai/resources/shared\";\nimport axios from \"axios\";\nimport Decimal from \"decimal.js\";\n\nconst baseURL = \"https://openrouter.ai/api/v1\";\nconst MODELS_CACHE_TTL = 1000 * 60 * 60 * 24; // 24 hours\n\nexport class OpenRouterProvider extends OpenAIProvider {\n  private models: ModelsResponse | undefined = undefined;\n  private modelsCachePromise: Promise<ModelsResponse | undefined> =\n    Promise.resolve(undefined);\n  private modelsUpdatedAt = 0;\n\n  constructor(config: {\n    apiKey: string;\n    maxRetries?: number;\n    timeout?: number;\n    rateLimiter?: RateLimiter;\n  }) {\n    super({\n      baseURL,\n      apiKey: config.apiKey,\n      maxRetries: config.maxRetries,\n      timeout: config.timeout,\n      rateLimiter: config.rateLimiter,\n    });\n  }\n\n  override async forward(args: {\n    messages: ChatCompletionMessageParam[];\n    model: string;\n    abortSignal?: AbortSignal;\n    temperature?: number;\n    responseFormat?:\n      | ResponseFormatText\n      | ResponseFormatJSONSchema\n      | ResponseFormatJSONObject;\n  }): Promise<ChatResponse> {\n    // Update models cache concurrently (non-blocking)\n    const [response] = await Promise.all([\n      super.forward(args),\n      this.updateModelsCache().catch(() => {\n        // Silently fail if cache update fails\n      }),\n    ]);\n\n    // Get the model info from the cache\n    const modelInfo = this.models?.data.find((m) => m.id === args.model);\n    let inputCost: string | undefined = undefined;\n    let outputCost: string | undefined = undefined;\n\n    if (modelInfo !== undefined) {\n      // Use Decimal.js for more accurate calculation\n      if (response.inputTokensUsed !== undefined) {\n        inputCost = new Decimal(modelInfo.pricing.prompt)\n          .mul(response.inputTokensUsed)\n          .toFixed(10);\n      }\n      if (response.outputTokensUsed !== undefined) {\n        outputCost = new Decimal(modelInfo.pricing.completion)\n          .mul(response.outputTokensUsed)\n          .toFixed(10);\n      }\n    }\n\n    return {\n      ...response,\n      inputCost,\n      outputCost,\n    };\n  }\n\n  /**\n   * Updates the cache that holds information about OpenRouter models\n   * including pricing information. It will be valid for 24 hours as\n   * long as the instance of this Provider object is alive.\n   */\n  private async updateModelsCache() {\n    // Chain each update method call to the promise.\n    // This approach prevents race conditions between multiple calls.\n    // Since each call is chained to the end of the previous one,\n    // each promise makes a request only if the models cache is not updated\n    // in the last call. Otherwise it simply resolves to the cached value.\n    this.modelsCachePromise = this.modelsCachePromise\n      .then(async () => {\n        if (\n          // The data presented in the cache\n          this.models !== undefined &&\n          // The cache is still valid\n          Date.now() - this.modelsUpdatedAt < MODELS_CACHE_TTL\n        ) {\n          return this.models;\n        }\n\n        // If the cache is not valid, update it\n        return axios\n          .get<ModelsResponse>(`${baseURL}/models`)\n          .then((res) => res.data)\n          .then((data) => {\n            // Only get the models that supports text input and output\n            data = {\n              data: data.data.filter(\n                (m) =>\n                  m.architecture.input_modalities.includes(\"text\") &&\n                  m.architecture.output_modalities.includes(\"text\") &&\n                  // These models are \"fast apply model\" and don't support multi turn conversations so don't include them\n                  ![\n                    \"morph/morph-v3-large\",\n                    \"morph/morph-v3-fast\",\n                    \"relace/relace-apply-3\",\n                  ].includes(m.id)\n              ),\n            };\n\n            this.models = data;\n            this.modelsUpdatedAt = Date.now();\n\n            return data;\n          });\n      })\n      .catch(() => undefined);\n\n    // Wait for the promise chain to resolve\n    await this.modelsCachePromise;\n  }\n}\n\ntype PutModality = \"text\" | \"image\" | \"file\" | \"audio\";\ntype Modality = \"text->text\" | \"text+image->text\" | \"text+image->text+image\";\ntype ModelsResponse = {\n  data: {\n    readonly id: string;\n    readonly canonical_slug: string;\n    readonly hugging_face_id: null | string;\n    readonly name: string;\n    readonly created: number;\n    readonly description: string;\n    readonly context_length: number;\n    readonly architecture: {\n      readonly modality: Modality;\n      readonly input_modalities: PutModality[];\n      readonly output_modalities: PutModality[];\n      readonly instruct_type: null | string;\n    };\n    readonly pricing: {\n      readonly prompt: string;\n      readonly completion: string;\n      readonly request?: string;\n      readonly image?: string;\n      readonly web_search?: string;\n      readonly internal_reasoning?: string;\n      readonly input_cache_read?: string;\n      readonly input_cache_write?: string;\n      readonly audio?: string;\n    };\n  }[];\n};\n"],"mappings":";;;;;;AAAO,IAAe,mBAAf,MAAgC;AAEvC;;;ACMO,IAAe,sBAAf,cAA2C,iBAAiB;AAEnE;;;ACJO,IAAM,iBAAN,cAA6B,oBAAoB;AAAA,EACpC,OAAO;AAAA,EAER;AAAA,EACA;AAAA,EACA;AAAA,EACT;AAAA,EACA;AAAA,EACA,4BAA4B;AAAA,EAC5B,4BAA4B;AAAA,EAEpC,YAAY,QAAoE;AAC9E,UAAM;AACN,SAAK,WAAW,OAAO;AACvB,SAAK,YAAY,OAAO;AACxB,SAAK,YAAY,OAAO;AAAA,EAC1B;AAAA,EAEA,MAAc,YAA0B;AACtC,QAAI,KAAK,OAAQ,QAAO,KAAK;AAC7B,QAAI,KAAK,kBAAmB,QAAO,KAAK;AAExC,SAAK,oBAAoB,OAAO,mBAAmB,EAChD,KAAK,CAAC,QAAQ;AACb,YAAM,eAAgB,KAAa;AACnC,UAAI,CAAC,cAAc;AACjB,cAAM,IAAI,MAAM,+CAA+C;AAAA,MACjE;AAEA,WAAK,SAAS,IAAI,aAAa;AAAA,QAC7B,SAAS,KAAK;AAAA,QACd,SAAS;AAAA,UACP,eAAe,UAAU,KAAK,SAAS;AAAA,QACzC;AAAA,MACF,CAAC;AAED,aAAO,KAAK;AAAA,IACd,CAAC,EACA,MAAM,CAAC,QAAQ;AACd,WAAK,oBAAoB;AACzB,YAAM,IAAI;AAAA,QACR,yKAEuB,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC;AAAA,MACzE;AAAA,IACF,CAAC;AAEH,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAe,QAAQ,MAAqD;AAC1E,UAAM,YAAY,KAAK,IAAI;AAE3B,QAAI,CAAC,KAAK,6BAA6B,KAAK,SAAS,KAAK,CAAC,MAAM,EAAE,SAAS,QAAQ,GAAG;AACrF,WAAK,4BAA4B;AACjC,cAAQ;AAAA,QACN,wDAAwD,KAAK,SAAS;AAAA,MACxE;AAAA,IACF;AAEA,QAAI,CAAC,KAAK,6BAA6B,KAAK,gBAAgB;AAC1D,WAAK,4BAA4B;AACjC,cAAQ;AAAA,QACN;AAAA,MACF;AAAA,IACF;AAEA,UAAM,cAAc,KAAK,SACtB,OAAO,CAAC,MAAM,EAAE,SAAS,UAAU,EAAE,SAAS,WAAW,EACzD,IAAI,CAAC,OAAO;AAAA,MACX,MAAM,EAAE;AAAA,MACR,SAAS,OAAQ,EAAU,WAAW,EAAE;AAAA,IAC1C,EAAE;AAEJ,UAAM,kBAAuB,EAAE,UAAU,YAAY;AAIrD,UAAM,QAAQ,OAAO,KAAK,SAAS,EAAE,EAAE,MAAM,GAAG;AAChD,QAAI,MAAM,WAAW,GAAG;AACtB,YAAM,CAAC,YAAY,OAAO,IAAI;AAC9B,sBAAgB,iBAAiB;AAAA,QAC/B,eAAe;AAAA,QACf,YAAY;AAAA,MACd;AAAA,IACF,WAAW,KAAK,SAAS,KAAK,UAAU,KAAK,WAAW;AACtD,sBAAgB,iBAAiB;AAAA,QAC/B,YAAY,KAAK;AAAA,MACnB;AAAA,IACF;AAEA,UAAM,SAAS,MAAM,KAAK,UAAU;AACpC,UAAM,QAAQ,OAAO,SAAS,KAAK,SAAS;AAC5C,UAAM,iBAAiB,KAAK,cAAe,EAAE,QAAQ,KAAK,YAAY,IAAY;AAClF,UAAM,WAAW,MAAM,MAAM,SAAS,iBAAiB,cAAc;AAErE,UAAM,cAAc,KAAK,IAAI;AAC7B,UAAM,OAAQ,UAAkB,QAAQ;AAExC,WAAO;AAAA,MACL,MAAM,OAAO,IAAI;AAAA,MACjB;AAAA,MACA;AAAA,IACF;AAAA,EACF;AACF;;;ACxGA,OAAO,UAAU,gBAAgB;AAG1B,IAAM,iBAAN,cAA6B,oBAAoB;AAAA,EACpC,OAAO;AAAA,EAEjB;AAAA,EACA;AAAA,EACA;AAAA,EAER,YAAY,QAMT;AACD,UAAM;AACN,SAAK,aAAa,OAAO,cAAc;AACvC,SAAK,cACH,OAAO,eACP,IAAI,YAAY;AAAA,MACd,WAAW;AAAA,MACX,YAAY;AAAA,IACd,CAAC;AAEH,SAAK,SAAS,IAAI,OAAO;AAAA,MACvB,SAAS,OAAO;AAAA,MAChB,QAAQ,OAAO;AAAA,MACf,SAAS,OAAO;AAAA,MAChB,yBAAyB;AAAA,IAC3B,CAAC;AAAA,EACH;AAAA,EAEA,MAAM,QAAQ,MASY;AACxB,QAAI,aAAa,KAAK;AACtB,WAAO,aAAa,GAAG;AACrB,UAAI,YAAkB,oBAAI,KAAK;AAE/B,UAAI;AACF,cAAM,WAAW,MAAM,KAAK,YAAY;AAAA,UACtC,YAAY;AAEV,wBAAY,oBAAI,KAAK;AACrB,mBAAO,MAAM,KAAK,OAAO,KAAK,YAAY;AAAA,cACxC;AAAA,gBACE,OAAO,KAAK;AAAA,gBACZ,UAAU,KAAK;AAAA,gBACf,aAAa,KAAK;AAAA,gBAClB,iBAAiB,KAAK;AAAA,cACxB;AAAA;AAAA,cAEA,EAAE,QAAQ,KAAK,YAAY;AAAA,YAC7B;AAAA,UACF;AAAA;AAAA,UAEA,EAAE,QAAQ,KAAK,YAAY;AAAA,QAC7B;AAEA,YAAI,WAAW,UAAU;AACvB,gBAAM,MAAM,SAAS;AACrB,gBAAM,IAAI;AAAA,YACR,GAAG,IAAI,OAAO,WAAW,IAAI,IAAI,MAAM,KAAK,UAAU,GAAG,CAAC;AAAA,UAC5D;AAAA,QACF;AAEA,YAAI,CAAC,UAAU,UAAU,CAAC,GAAG,SAAS,SAAS;AAC7C,gBAAM,IAAI,MAAM,oCAAoC;AAAA,QACtD;AAEA,eAAO;AAAA,UACL,MAAM,SAAS,QAAQ,CAAC,EAAE,QAAQ;AAAA,UAElC,iBAAiB,UAAU,OAAO;AAAA,UAClC,kBAAkB,UAAU,OAAO;AAAA,UAEnC,WAAW,UAAU,QAAQ;AAAA,UAC7B,aAAa,KAAK,IAAI;AAAA,QACxB;AAAA,MACF,SAAS,KAAK;AACZ,YAAI,eAAe,YAAY,IAAI,WAAW,KAAK;AACjD,gBAAM,IAAI,MAAM,gCAAgC,EAAE,OAAO,IAAI,CAAC;AAAA,QAChE;AAEA;AAOA,YAAI,eAAe,aAAa;AAC9B,kBAAQ,MAAM,GAAG;AACjB;AAAA,QACF;AAGA,YAAI,eAAe,GAAG;AACpB;AAAA,QACF;AAEA,cAAM,IAAI;AAAA,UACR,0CAA0C,eAAe,QAAQ,IAAI,UAAU,GAAG;AAAA,UAClF,EAAE,OAAO,IAAI;AAAA,QACf;AAAA,MACF;AAAA,IACF;AAEA,UAAM,IAAI;AAAA,MACR;AAAA,MACA,EAAE,OAAO,IAAI,MAAM,qBAAqB,EAAE;AAAA,IAC5C;AAAA,EACF;AACF;;;ACxHA,OAAO,WAAW;AAClB,OAAO,aAAa;AAEpB,IAAM,UAAU;AAChB,IAAM,mBAAmB,MAAO,KAAK,KAAK;AAEnC,IAAM,qBAAN,cAAiC,eAAe;AAAA,EAC7C,SAAqC;AAAA,EACrC,qBACN,QAAQ,QAAQ,MAAS;AAAA,EACnB,kBAAkB;AAAA,EAE1B,YAAY,QAKT;AACD,UAAM;AAAA,MACJ;AAAA,MACA,QAAQ,OAAO;AAAA,MACf,YAAY,OAAO;AAAA,MACnB,SAAS,OAAO;AAAA,MAChB,aAAa,OAAO;AAAA,IACtB,CAAC;AAAA,EACH;AAAA,EAEA,MAAe,QAAQ,MASG;AAExB,UAAM,CAAC,QAAQ,IAAI,MAAM,QAAQ,IAAI;AAAA,MACnC,MAAM,QAAQ,IAAI;AAAA,MAClB,KAAK,kBAAkB,EAAE,MAAM,MAAM;AAAA,MAErC,CAAC;AAAA,IACH,CAAC;AAGD,UAAM,YAAY,KAAK,QAAQ,KAAK,KAAK,CAAC,MAAM,EAAE,OAAO,KAAK,KAAK;AACnE,QAAI,YAAgC;AACpC,QAAI,aAAiC;AAErC,QAAI,cAAc,QAAW;AAE3B,UAAI,SAAS,oBAAoB,QAAW;AAC1C,oBAAY,IAAI,QAAQ,UAAU,QAAQ,MAAM,EAC7C,IAAI,SAAS,eAAe,EAC5B,QAAQ,EAAE;AAAA,MACf;AACA,UAAI,SAAS,qBAAqB,QAAW;AAC3C,qBAAa,IAAI,QAAQ,UAAU,QAAQ,UAAU,EAClD,IAAI,SAAS,gBAAgB,EAC7B,QAAQ,EAAE;AAAA,MACf;AAAA,IACF;AAEA,WAAO;AAAA,MACL,GAAG;AAAA,MACH;AAAA,MACA;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,MAAc,oBAAoB;AAMhC,SAAK,qBAAqB,KAAK,mBAC5B,KAAK,YAAY;AAChB;AAAA;AAAA,QAEE,KAAK,WAAW;AAAA,QAEhB,KAAK,IAAI,IAAI,KAAK,kBAAkB;AAAA,QACpC;AACA,eAAO,KAAK;AAAA,MACd;AAGA,aAAO,MACJ,IAAoB,GAAG,OAAO,SAAS,EACvC,KAAK,CAAC,QAAQ,IAAI,IAAI,EACtB,KAAK,CAAC,SAAS;AAEd,eAAO;AAAA,UACL,MAAM,KAAK,KAAK;AAAA,YACd,CAAC,MACC,EAAE,aAAa,iBAAiB,SAAS,MAAM,KAC/C,EAAE,aAAa,kBAAkB,SAAS,MAAM;AAAA,YAEhD,CAAC;AAAA,cACC;AAAA,cACA;AAAA,cACA;AAAA,YACF,EAAE,SAAS,EAAE,EAAE;AAAA,UACnB;AAAA,QACF;AAEA,aAAK,SAAS;AACd,aAAK,kBAAkB,KAAK,IAAI;AAEhC,eAAO;AAAA,MACT,CAAC;AAAA,IACL,CAAC,EACA,MAAM,MAAM,MAAS;AAGxB,UAAM,KAAK;AAAA,EACb;AACF;","names":[]}

package/dist/rate-limiter-CSmVIRsM.d.ts

@@ -0,0 +1,60 @@
+interface RateLimiterOptions {
+    /**
+     * Maximum weight of calls allowed within a time window. Each call has a weight of 1 by default.
+     * Rate limiting is disabled when this is 0 or negative.
+     * @default 25
+     */
+    maxWeight?: number;
+    /**
+     * Time window in milliseconds for rate limiting
+     * @default 1000
+     */
+    timeWindow?: number;
+}
+interface RateLimiterCallOptions {
+    /**
+     * Weight of the call. Rate limiting is applied based on the total weight of the calls.
+     * @default 1
+     */
+    weight?: number;
+    /**
+     * Abort signal for cancellation
+     */
+    signal?: AbortSignal;
+}
+/**
+ * Generic rate limiter. It can be used to limit async function calls
+ * by a given number of calls within a time window.
+ */
+declare class RateLimiter {
+    maxWeight: number;
+    timeWindow: number;
+    private timestamps;
+    constructor(options?: RateLimiterOptions);
+    /**
+     * Checks if rate limiting is disabled
+     */
+    isDisabled(): boolean;
+    /**
+     * Disables rate limiting. Set `maxWeight` to re-enable it.
+     */
+    disable(): void;
+    /**
+     * Returns how many weight of calls are there in the current time window
+     */
+    getCurrentCalls(): number;
+    /**
+     * Executes the given function with rate limiting applied
+     */
+    execute<T = unknown>(func: () => Promise<T>, { weight, signal }?: RateLimiterCallOptions): Promise<T>;
+    /**
+     * Waits until rate limit allows the specified weight of calls
+     */
+    private waitForRateLimit;
+    /**
+     * Resets the rate limited weight of calls
+     */
+    reset(): void;
+}
+
+export { RateLimiter as R, type RateLimiterOptions as a, type RateLimiterCallOptions as b };

package/dist/schemas/extensions/index.d.ts

@@ -0,0 +1,14 @@
+import * as z from 'zod';
+
+declare const ScoreExtensions: {
+    ExtensionAIScorerFieldsV1: {
+        scorerAIProvider: z.ZodOptional<z.ZodString>;
+        scorerAIModelSlug: z.ZodOptional<z.ZodString>;
+        scorerAIInputTokensUsed: z.ZodOptional<z.ZodNumber>;
+        scorerAIOutputTokensUsed: z.ZodOptional<z.ZodNumber>;
+        scorerAIInputCost: z.ZodOptional<z.ZodString>;
+        scorerAIOutputCost: z.ZodOptional<z.ZodString>;
+    };
+};
+
+export { ScoreExtensions };

package/dist/schemas/extensions/index.js

@@ -0,0 +1,13 @@
+import {
+  ExtensionLLMAsAJudgeScorerFieldsV1
+} from "../../chunk-GVF4YZF3.js";
+import "../../chunk-PZ5AY32C.js";
+
+// src/schemas/extensions/index.ts
+var ScoreExtensions = {
+  ExtensionAIScorerFieldsV1: ExtensionLLMAsAJudgeScorerFieldsV1
+};
+export {
+  ScoreExtensions
+};
+//# sourceMappingURL=index.js.map

package/dist/schemas/extensions/index.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../../src/schemas/extensions/index.ts"],"sourcesContent":["import { ExtensionLLMAsAJudgeScorerFieldsV1 } from \"./score/llm-as-a-judge-scorer\";\n\nexport const ScoreExtensions = {\n  ExtensionAIScorerFieldsV1: ExtensionLLMAsAJudgeScorerFieldsV1,\n};\n"],"mappings":";;;;;;AAEO,IAAM,kBAAkB;AAAA,EAC7B,2BAA2B;AAC7B;","names":[]}