@librechat/agents 3.0.51 → 3.0.52

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (7) hide show
  1. package/dist/cjs/llm/vertexai/index.cjs +7 -8
  2. package/dist/cjs/llm/vertexai/index.cjs.map +1 -1
  3. package/dist/esm/llm/vertexai/index.mjs +7 -8
  4. package/dist/esm/llm/vertexai/index.mjs.map +1 -1
  5. package/package.json +1 -1
  6. package/src/llm/vertexai/index.ts +9 -10
  7. package/src/utils/llmConfig.ts +5 -0
package/dist/cjs/llm/vertexai/index.cjs CHANGED
@@ -6,9 +6,12 @@ var googleCommon = require('@langchain/google-common');
6
6
  class CustomChatConnection extends googleCommon.ChatConnection {
7
7
  async formatData(input, parameters) {
8
8
  const formattedData = (await super.formatData(input, parameters));
9
- if (formattedData.generationConfig?.thinkingConfig?.thinkingBudget === -1 &&
10
- formattedData.generationConfig.thinkingConfig.includeThoughts === false) {
11
- formattedData.generationConfig.thinkingConfig.includeThoughts = true;
9
+ if (formattedData.generationConfig?.thinkingConfig?.thinkingBudget === -1) {
10
+ // -1 means "let the model decide" - delete the property so the API doesn't receive an invalid value
11
+ if (formattedData.generationConfig.thinkingConfig.includeThoughts === false) {
12
+ formattedData.generationConfig.thinkingConfig.includeThoughts = true;
13
+ }
14
+ delete formattedData.generationConfig.thinkingConfig.thinkingBudget;
12
15
  }
13
16
  return formattedData;
14
17
  }
@@ -302,11 +305,7 @@ class ChatVertexAI extends googleGauth.ChatGoogle {
302
305
  return 'LibreChatVertexAI';
303
306
  }
304
307
  constructor(fields) {
305
- let dynamicThinkingBudget = false;
306
- if (fields?.thinkingBudget === -1) {
307
- dynamicThinkingBudget = true;
308
- fields.thinkingBudget = 1;
309
- }
308
+ const dynamicThinkingBudget = fields?.thinkingBudget === -1;
310
309
  super({
311
310
  ...fields,
312
311
  platformType: 'gcp',
package/dist/cjs/llm/vertexai/index.cjs.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"file":"index.cjs","sources":["../../../../src/llm/vertexai/index.ts"],"sourcesContent":["import { ChatGoogle } from '@langchain/google-gauth';\nimport { ChatConnection } from '@langchain/google-common';\nimport type {\n GeminiRequest,\n GoogleAIModelRequestParams,\n GoogleAbstractedClient,\n} from '@langchain/google-common';\nimport type { BaseMessage } from '@langchain/core/messages';\nimport type { VertexAIClientOptions } from '@/types';\n\nclass CustomChatConnection extends ChatConnection<VertexAIClientOptions> {\n async formatData(\n input: BaseMessage[],\n parameters: GoogleAIModelRequestParams\n ): Promise<unknown> {\n const formattedData = (await super.formatData(\n input,\n parameters\n )) as GeminiRequest;\n if (\n formattedData.generationConfig?.thinkingConfig?.thinkingBudget === -1 &&\n formattedData.generationConfig.thinkingConfig.includeThoughts === false\n ) {\n formattedData.generationConfig.thinkingConfig.includeThoughts = true;\n }\n return formattedData;\n }\n}\n\n/**\n * Integration with Google Vertex AI chat models.\n *\n * Setup:\n * Install `@langchain/google-vertexai` and set your stringified\n * Vertex AI credentials as an environment variable named `GOOGLE_APPLICATION_CREDENTIALS`.\n *\n * ```bash\n * npm install @langchain/google-vertexai\n * export GOOGLE_APPLICATION_CREDENTIALS=\"path/to/credentials\"\n * ```\n *\n * ## [Constructor args](https://api.js.langchain.com/classes/_langchain_google_vertexai.index.ChatVertexAI.html#constructor.new_ChatVertexAI)\n *\n * ## [Runtime args](https://api.js.langchain.com/interfaces/langchain_google_common_types.GoogleAIBaseLanguageModelCallOptions.html)\n *\n * Runtime args can be passed as the second argument to any of the base runnable methods `.invoke`. `.stream`, `.batch`, etc.\n * They can also be passed via `.withConfig`, or the second arg in `.bindTools`, like shown in the examples below:\n *\n * ```typescript\n * // When calling `.withConfig`, call options should be passed via the first argument\n * const llmWithArgsBound = llm.withConfig({\n * stop: [\"\\n\"],\n * tools: [...],\n * });\n *\n * // When calling `.bindTools`, call options should be passed via the second argument\n * const llmWithTools = llm.bindTools(\n * [...],\n * {\n * tool_choice: \"auto\",\n * }\n * );\n * ```\n *\n * ## Examples\n *\n * <details open>\n * <summary><strong>Instantiate</strong></summary>\n *\n * ```typescript\n * import { ChatVertexAI } from '@langchain/google-vertexai';\n *\n * const llm = new ChatVertexAI({\n * model: \"gemini-1.5-pro\",\n * temperature: 0,\n * // other params...\n * });\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Invoking</strong></summary>\n *\n * ```typescript\n * const input = `Translate \"I love programming\" into French.`;\n *\n * // Models also accept a list of chat messages or a formatted prompt\n * const result = await llm.invoke(input);\n * console.log(result);\n * ```\n *\n * ```txt\n * AIMessageChunk {\n * \"content\": \"\\\"J'adore programmer\\\" \\n\\nHere's why this is the best translation:\\n\\n* **J'adore** means \\\"I love\\\" and conveys a strong passion.\\n* **Programmer** is the French verb for \\\"to program.\\\"\\n\\nThis translation is natural and idiomatic in French. \\n\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": [],\n * \"usage_metadata\": {\n * \"input_tokens\": 9,\n * \"output_tokens\": 63,\n * \"total_tokens\": 72\n * }\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Streaming Chunks</strong></summary>\n *\n * ```typescript\n * for await (const chunk of await llm.stream(input)) {\n * console.log(chunk);\n * }\n * ```\n *\n * ```txt\n * AIMessageChunk {\n * \"content\": \"\\\"\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"J'adore programmer\\\" \\n\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": [],\n * \"usage_metadata\": {\n * \"input_tokens\": 9,\n * \"output_tokens\": 8,\n * \"total_tokens\": 17\n * }\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Aggregate Streamed Chunks</strong></summary>\n *\n * ```typescript\n * import { AIMessageChunk } from '@langchain/core/messages';\n * import { concat } from '@langchain/core/utils/stream';\n *\n * const stream = await llm.stream(input);\n * let full: AIMessageChunk | undefined;\n * for await (const chunk of stream) {\n * full = !full ? chunk : concat(full, chunk);\n * }\n * console.log(full);\n * ```\n *\n * ```txt\n * AIMessageChunk {\n * \"content\": \"\\\"J'adore programmer\\\" \\n\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": [],\n * \"usage_metadata\": {\n * \"input_tokens\": 9,\n * \"output_tokens\": 8,\n * \"total_tokens\": 17\n * }\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Bind tools</strong></summary>\n *\n * ```typescript\n * import { z } from 'zod';\n *\n * const GetWeather = {\n * name: \"GetWeather\",\n * description: \"Get the current weather in a given location\",\n * schema: z.object({\n * location: z.string().describe(\"The city and state, e.g. San Francisco, CA\")\n * }),\n * }\n *\n * const GetPopulation = {\n * name: \"GetPopulation\",\n * description: \"Get the current population in a given location\",\n * schema: z.object({\n * location: z.string().describe(\"The city and state, e.g. San Francisco, CA\")\n * }),\n * }\n *\n * const llmWithTools = llm.bindTools([GetWeather, GetPopulation]);\n * const aiMsg = await llmWithTools.invoke(\n * \"Which city is hotter today and which is bigger: LA or NY?\"\n * );\n * console.log(aiMsg.tool_calls);\n * ```\n *\n * ```txt\n * [\n * {\n * name: 'GetPopulation',\n * args: { location: 'New York City, NY' },\n * id: '33c1c1f47e2f492799c77d2800a43912',\n * type: 'tool_call'\n * }\n * ]\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Structured Output</strong></summary>\n *\n * ```typescript\n * import { z } from 'zod';\n *\n * const Joke = z.object({\n * setup: z.string().describe(\"The setup of the joke\"),\n * punchline: z.string().describe(\"The punchline to the joke\"),\n * rating: z.number().optional().describe(\"How funny the joke is, from 1 to 10\")\n * }).describe('Joke to tell user.');\n *\n * const structuredLlm = llm.withStructuredOutput(Joke, { name: \"Joke\" });\n * const jokeResult = await structuredLlm.invoke(\"Tell me a joke about cats\");\n * console.log(jokeResult);\n * ```\n *\n * ```txt\n * {\n * setup: 'What do you call a cat that loves to bowl?',\n * punchline: 'An alley cat!'\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Usage Metadata</strong></summary>\n *\n * ```typescript\n * const aiMsgForMetadata = await llm.invoke(input);\n * console.log(aiMsgForMetadata.usage_metadata);\n * ```\n *\n * ```txt\n * { input_tokens: 9, output_tokens: 8, total_tokens: 17 }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Stream Usage Metadata</strong></summary>\n *\n * ```typescript\n * const streamForMetadata = await llm.stream(\n * input,\n * {\n * streamUsage: true\n * }\n * );\n * let fullForMetadata: AIMessageChunk | undefined;\n * for await (const chunk of streamForMetadata) {\n * fullForMetadata = !fullForMetadata ? chunk : concat(fullForMetadata, chunk);\n * }\n * console.log(fullForMetadata?.usage_metadata);\n * ```\n *\n * ```txt\n * { input_tokens: 9, output_tokens: 8, total_tokens: 17 }\n * ```\n * </details>\n *\n * <br />\n */\nexport class ChatVertexAI extends ChatGoogle {\n lc_namespace = ['langchain', 'chat_models', 'vertexai'];\n dynamicThinkingBudget = false;\n\n static lc_name(): 'LibreChatVertexAI' {\n return 'LibreChatVertexAI';\n }\n\n constructor(fields?: VertexAIClientOptions) {\n let dynamicThinkingBudget = false;\n if (fields?.thinkingBudget === -1) {\n dynamicThinkingBudget = true;\n fields.thinkingBudget = 1;\n }\n super({\n ...fields,\n platformType: 'gcp',\n });\n this.dynamicThinkingBudget = dynamicThinkingBudget;\n }\n invocationParams(\n options?: this['ParsedCallOptions'] | undefined\n ): GoogleAIModelRequestParams {\n const params = super.invocationParams(options);\n if (this.dynamicThinkingBudget) {\n params.maxReasoningTokens = -1;\n }\n return params;\n }\n\n buildConnection(\n fields: VertexAIClientOptions,\n client: GoogleAbstractedClient\n ): void {\n this.connection = new CustomChatConnection(\n { ...fields, ...this },\n this.caller,\n client,\n false\n );\n\n this.streamedConnection = new CustomChatConnection(\n { ...fields, ...this },\n this.caller,\n client,\n true\n );\n }\n}\n"],"names":["ChatConnection","ChatGoogle"],"mappings":";;;;;AAUA,MAAM,oBAAqB,SAAQA,2BAAqC,CAAA;AACtE,IAAA,MAAM,UAAU,CACd,KAAoB,EACpB,UAAsC,EAAA;AAEtC,QAAA,MAAM,aAAa,IAAI,MAAM,KAAK,CAAC,UAAU,CAC3C,KAAK,EACL,UAAU,CACX,CAAkB;QACnB,IACE,aAAa,CAAC,gBAAgB,EAAE,cAAc,EAAE,cAAc,KAAK,EAAE;YACrE,aAAa,CAAC,gBAAgB,CAAC,cAAc,CAAC,eAAe,KAAK,KAAK,EACvE;YACA,aAAa,CAAC,gBAAgB,CAAC,cAAc,CAAC,eAAe,GAAG,IAAI;;AAEtE,QAAA,OAAO,aAAa;;AAEvB;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAyRG;AACG,MAAO,YAAa,SAAQC,sBAAU,CAAA;IAC1C,YAAY,GAAG,CAAC,WAAW,EAAE,aAAa,EAAE,UAAU,CAAC;IACvD,qBAAqB,GAAG,KAAK;AAE7B,IAAA,OAAO,OAAO,GAAA;AACZ,QAAA,OAAO,mBAAmB;;AAG5B,IAAA,WAAA,CAAY,MAA8B,EAAA;QACxC,IAAI,qBAAqB,GAAG,KAAK;AACjC,QAAA,IAAI,MAAM,EAAE,cAAc,KAAK,EAAE,EAAE;YACjC,qBAAqB,GAAG,IAAI;AAC5B,YAAA,MAAM,CAAC,cAAc,GAAG,CAAC;;AAE3B,QAAA,KAAK,CAAC;AACJ,YAAA,GAAG,MAAM;AACT,YAAA,YAAY,EAAE,KAAK;AACpB,SAAA,CAAC;AACF,QAAA,IAAI,CAAC,qBAAqB,GAAG,qBAAqB;;AAEpD,IAAA,gBAAgB,CACd,OAA+C,EAAA;QAE/C,MAAM,MAAM,GAAG,KAAK,CAAC,gBAAgB,CAAC,OAAO,CAAC;AAC9C,QAAA,IAAI,IAAI,CAAC,qBAAqB,EAAE;AAC9B,YAAA,MAAM,CAAC,kBAAkB,GAAG,EAAE;;AAEhC,QAAA,OAAO,MAAM;;IAGf,eAAe,CACb,MAA6B,EAC7B,MAA8B,EAAA;QAE9B,IAAI,CAAC,UAAU,GAAG,IAAI,oBAAoB,CACxC,EAAE,GAAG,MAAM,EAAE,GAAG,IAAI,EAAE,EACtB,IAAI,CAAC,MAAM,EACX,MAAM,EACN,KAAK,CACN;QAED,IAAI,CAAC,kBAAkB,GAAG,IAAI,oBAAoB,CAChD,EAAE,GAAG,MAAM,EAAE,GAAG,IAAI,EAAE,EACtB,IAAI,CAAC,MAAM,EACX,MAAM,EACN,IAAI,CACL;;AAEJ;;;;"}
1
+ {"version":3,"file":"index.cjs","sources":["../../../../src/llm/vertexai/index.ts"],"sourcesContent":["import { ChatGoogle } from '@langchain/google-gauth';\nimport { ChatConnection } from '@langchain/google-common';\nimport type {\n GeminiRequest,\n GoogleAIModelRequestParams,\n GoogleAbstractedClient,\n} from '@langchain/google-common';\nimport type { BaseMessage } from '@langchain/core/messages';\nimport type { VertexAIClientOptions } from '@/types';\n\nclass CustomChatConnection extends ChatConnection<VertexAIClientOptions> {\n async formatData(\n input: BaseMessage[],\n parameters: GoogleAIModelRequestParams\n ): Promise<unknown> {\n const formattedData = (await super.formatData(\n input,\n parameters\n )) as GeminiRequest;\n if (formattedData.generationConfig?.thinkingConfig?.thinkingBudget === -1) {\n // -1 means \"let the model decide\" - delete the property so the API doesn't receive an invalid value\n if (\n formattedData.generationConfig.thinkingConfig.includeThoughts === false\n ) {\n formattedData.generationConfig.thinkingConfig.includeThoughts = true;\n }\n delete formattedData.generationConfig.thinkingConfig.thinkingBudget;\n }\n return formattedData;\n }\n}\n\n/**\n * Integration with Google Vertex AI chat models.\n *\n * Setup:\n * Install `@langchain/google-vertexai` and set your stringified\n * Vertex AI credentials as an environment variable named `GOOGLE_APPLICATION_CREDENTIALS`.\n *\n * ```bash\n * npm install @langchain/google-vertexai\n * export GOOGLE_APPLICATION_CREDENTIALS=\"path/to/credentials\"\n * ```\n *\n * ## [Constructor args](https://api.js.langchain.com/classes/_langchain_google_vertexai.index.ChatVertexAI.html#constructor.new_ChatVertexAI)\n *\n * ## [Runtime args](https://api.js.langchain.com/interfaces/langchain_google_common_types.GoogleAIBaseLanguageModelCallOptions.html)\n *\n * Runtime args can be passed as the second argument to any of the base runnable methods `.invoke`. `.stream`, `.batch`, etc.\n * They can also be passed via `.withConfig`, or the second arg in `.bindTools`, like shown in the examples below:\n *\n * ```typescript\n * // When calling `.withConfig`, call options should be passed via the first argument\n * const llmWithArgsBound = llm.withConfig({\n * stop: [\"\\n\"],\n * tools: [...],\n * });\n *\n * // When calling `.bindTools`, call options should be passed via the second argument\n * const llmWithTools = llm.bindTools(\n * [...],\n * {\n * tool_choice: \"auto\",\n * }\n * );\n * ```\n *\n * ## Examples\n *\n * <details open>\n * <summary><strong>Instantiate</strong></summary>\n *\n * ```typescript\n * import { ChatVertexAI } from '@langchain/google-vertexai';\n *\n * const llm = new ChatVertexAI({\n * model: \"gemini-1.5-pro\",\n * temperature: 0,\n * // other params...\n * });\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Invoking</strong></summary>\n *\n * ```typescript\n * const input = `Translate \"I love programming\" into French.`;\n *\n * // Models also accept a list of chat messages or a formatted prompt\n * const result = await llm.invoke(input);\n * console.log(result);\n * ```\n *\n * ```txt\n * AIMessageChunk {\n * \"content\": \"\\\"J'adore programmer\\\" \\n\\nHere's why this is the best translation:\\n\\n* **J'adore** means \\\"I love\\\" and conveys a strong passion.\\n* **Programmer** is the French verb for \\\"to program.\\\"\\n\\nThis translation is natural and idiomatic in French. \\n\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": [],\n * \"usage_metadata\": {\n * \"input_tokens\": 9,\n * \"output_tokens\": 63,\n * \"total_tokens\": 72\n * }\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Streaming Chunks</strong></summary>\n *\n * ```typescript\n * for await (const chunk of await llm.stream(input)) {\n * console.log(chunk);\n * }\n * ```\n *\n * ```txt\n * AIMessageChunk {\n * \"content\": \"\\\"\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"J'adore programmer\\\" \\n\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": [],\n * \"usage_metadata\": {\n * \"input_tokens\": 9,\n * \"output_tokens\": 8,\n * \"total_tokens\": 17\n * }\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Aggregate Streamed Chunks</strong></summary>\n *\n * ```typescript\n * import { AIMessageChunk } from '@langchain/core/messages';\n * import { concat } from '@langchain/core/utils/stream';\n *\n * const stream = await llm.stream(input);\n * let full: AIMessageChunk | undefined;\n * for await (const chunk of stream) {\n * full = !full ? chunk : concat(full, chunk);\n * }\n * console.log(full);\n * ```\n *\n * ```txt\n * AIMessageChunk {\n * \"content\": \"\\\"J'adore programmer\\\" \\n\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": [],\n * \"usage_metadata\": {\n * \"input_tokens\": 9,\n * \"output_tokens\": 8,\n * \"total_tokens\": 17\n * }\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Bind tools</strong></summary>\n *\n * ```typescript\n * import { z } from 'zod';\n *\n * const GetWeather = {\n * name: \"GetWeather\",\n * description: \"Get the current weather in a given location\",\n * schema: z.object({\n * location: z.string().describe(\"The city and state, e.g. San Francisco, CA\")\n * }),\n * }\n *\n * const GetPopulation = {\n * name: \"GetPopulation\",\n * description: \"Get the current population in a given location\",\n * schema: z.object({\n * location: z.string().describe(\"The city and state, e.g. San Francisco, CA\")\n * }),\n * }\n *\n * const llmWithTools = llm.bindTools([GetWeather, GetPopulation]);\n * const aiMsg = await llmWithTools.invoke(\n * \"Which city is hotter today and which is bigger: LA or NY?\"\n * );\n * console.log(aiMsg.tool_calls);\n * ```\n *\n * ```txt\n * [\n * {\n * name: 'GetPopulation',\n * args: { location: 'New York City, NY' },\n * id: '33c1c1f47e2f492799c77d2800a43912',\n * type: 'tool_call'\n * }\n * ]\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Structured Output</strong></summary>\n *\n * ```typescript\n * import { z } from 'zod';\n *\n * const Joke = z.object({\n * setup: z.string().describe(\"The setup of the joke\"),\n * punchline: z.string().describe(\"The punchline to the joke\"),\n * rating: z.number().optional().describe(\"How funny the joke is, from 1 to 10\")\n * }).describe('Joke to tell user.');\n *\n * const structuredLlm = llm.withStructuredOutput(Joke, { name: \"Joke\" });\n * const jokeResult = await structuredLlm.invoke(\"Tell me a joke about cats\");\n * console.log(jokeResult);\n * ```\n *\n * ```txt\n * {\n * setup: 'What do you call a cat that loves to bowl?',\n * punchline: 'An alley cat!'\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Usage Metadata</strong></summary>\n *\n * ```typescript\n * const aiMsgForMetadata = await llm.invoke(input);\n * console.log(aiMsgForMetadata.usage_metadata);\n * ```\n *\n * ```txt\n * { input_tokens: 9, output_tokens: 8, total_tokens: 17 }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Stream Usage Metadata</strong></summary>\n *\n * ```typescript\n * const streamForMetadata = await llm.stream(\n * input,\n * {\n * streamUsage: true\n * }\n * );\n * let fullForMetadata: AIMessageChunk | undefined;\n * for await (const chunk of streamForMetadata) {\n * fullForMetadata = !fullForMetadata ? chunk : concat(fullForMetadata, chunk);\n * }\n * console.log(fullForMetadata?.usage_metadata);\n * ```\n *\n * ```txt\n * { input_tokens: 9, output_tokens: 8, total_tokens: 17 }\n * ```\n * </details>\n *\n * <br />\n */\nexport class ChatVertexAI extends ChatGoogle {\n lc_namespace = ['langchain', 'chat_models', 'vertexai'];\n dynamicThinkingBudget = false;\n\n static lc_name(): 'LibreChatVertexAI' {\n return 'LibreChatVertexAI';\n }\n\n constructor(fields?: VertexAIClientOptions) {\n const dynamicThinkingBudget = fields?.thinkingBudget === -1;\n super({\n ...fields,\n platformType: 'gcp',\n });\n this.dynamicThinkingBudget = dynamicThinkingBudget;\n }\n invocationParams(\n options?: this['ParsedCallOptions'] | undefined\n ): GoogleAIModelRequestParams {\n const params = super.invocationParams(options);\n if (this.dynamicThinkingBudget) {\n params.maxReasoningTokens = -1;\n }\n return params;\n }\n\n buildConnection(\n fields: VertexAIClientOptions,\n client: GoogleAbstractedClient\n ): void {\n this.connection = new CustomChatConnection(\n { ...fields, ...this },\n this.caller,\n client,\n false\n );\n\n this.streamedConnection = new CustomChatConnection(\n { ...fields, ...this },\n this.caller,\n client,\n true\n );\n }\n}\n"],"names":["ChatConnection","ChatGoogle"],"mappings":";;;;;AAUA,MAAM,oBAAqB,SAAQA,2BAAqC,CAAA;AACtE,IAAA,MAAM,UAAU,CACd,KAAoB,EACpB,UAAsC,EAAA;AAEtC,QAAA,MAAM,aAAa,IAAI,MAAM,KAAK,CAAC,UAAU,CAC3C,KAAK,EACL,UAAU,CACX,CAAkB;QACnB,IAAI,aAAa,CAAC,gBAAgB,EAAE,cAAc,EAAE,cAAc,KAAK,EAAE,EAAE;;YAEzE,IACE,aAAa,CAAC,gBAAgB,CAAC,cAAc,CAAC,eAAe,KAAK,KAAK,EACvE;gBACA,aAAa,CAAC,gBAAgB,CAAC,cAAc,CAAC,eAAe,GAAG,IAAI;;AAEtE,YAAA,OAAO,aAAa,CAAC,gBAAgB,CAAC,cAAc,CAAC,cAAc;;AAErE,QAAA,OAAO,aAAa;;AAEvB;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAyRG;AACG,MAAO,YAAa,SAAQC,sBAAU,CAAA;IAC1C,YAAY,GAAG,CAAC,WAAW,EAAE,aAAa,EAAE,UAAU,CAAC;IACvD,qBAAqB,GAAG,KAAK;AAE7B,IAAA,OAAO,OAAO,GAAA;AACZ,QAAA,OAAO,mBAAmB;;AAG5B,IAAA,WAAA,CAAY,MAA8B,EAAA;QACxC,MAAM,qBAAqB,GAAG,MAAM,EAAE,cAAc,KAAK,EAAE;AAC3D,QAAA,KAAK,CAAC;AACJ,YAAA,GAAG,MAAM;AACT,YAAA,YAAY,EAAE,KAAK;AACpB,SAAA,CAAC;AACF,QAAA,IAAI,CAAC,qBAAqB,GAAG,qBAAqB;;AAEpD,IAAA,gBAAgB,CACd,OAA+C,EAAA;QAE/C,MAAM,MAAM,GAAG,KAAK,CAAC,gBAAgB,CAAC,OAAO,CAAC;AAC9C,QAAA,IAAI,IAAI,CAAC,qBAAqB,EAAE;AAC9B,YAAA,MAAM,CAAC,kBAAkB,GAAG,EAAE;;AAEhC,QAAA,OAAO,MAAM;;IAGf,eAAe,CACb,MAA6B,EAC7B,MAA8B,EAAA;QAE9B,IAAI,CAAC,UAAU,GAAG,IAAI,oBAAoB,CACxC,EAAE,GAAG,MAAM,EAAE,GAAG,IAAI,EAAE,EACtB,IAAI,CAAC,MAAM,EACX,MAAM,EACN,KAAK,CACN;QAED,IAAI,CAAC,kBAAkB,GAAG,IAAI,oBAAoB,CAChD,EAAE,GAAG,MAAM,EAAE,GAAG,IAAI,EAAE,EACtB,IAAI,CAAC,MAAM,EACX,MAAM,EACN,IAAI,CACL;;AAEJ;;;;"}
package/dist/esm/llm/vertexai/index.mjs CHANGED
@@ -4,9 +4,12 @@ import { ChatConnection } from '@langchain/google-common';
4
4
  class CustomChatConnection extends ChatConnection {
5
5
  async formatData(input, parameters) {
6
6
  const formattedData = (await super.formatData(input, parameters));
7
- if (formattedData.generationConfig?.thinkingConfig?.thinkingBudget === -1 &&
8
- formattedData.generationConfig.thinkingConfig.includeThoughts === false) {
9
- formattedData.generationConfig.thinkingConfig.includeThoughts = true;
7
+ if (formattedData.generationConfig?.thinkingConfig?.thinkingBudget === -1) {
8
+ // -1 means "let the model decide" - delete the property so the API doesn't receive an invalid value
9
+ if (formattedData.generationConfig.thinkingConfig.includeThoughts === false) {
10
+ formattedData.generationConfig.thinkingConfig.includeThoughts = true;
11
+ }
12
+ delete formattedData.generationConfig.thinkingConfig.thinkingBudget;
10
13
  }
11
14
  return formattedData;
12
15
  }
@@ -300,11 +303,7 @@ class ChatVertexAI extends ChatGoogle {
300
303
  return 'LibreChatVertexAI';
301
304
  }
302
305
  constructor(fields) {
303
- let dynamicThinkingBudget = false;
304
- if (fields?.thinkingBudget === -1) {
305
- dynamicThinkingBudget = true;
306
- fields.thinkingBudget = 1;
307
- }
306
+ const dynamicThinkingBudget = fields?.thinkingBudget === -1;
308
307
  super({
309
308
  ...fields,
310
309
  platformType: 'gcp',
package/dist/esm/llm/vertexai/index.mjs.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"file":"index.mjs","sources":["../../../../src/llm/vertexai/index.ts"],"sourcesContent":["import { ChatGoogle } from '@langchain/google-gauth';\nimport { ChatConnection } from '@langchain/google-common';\nimport type {\n GeminiRequest,\n GoogleAIModelRequestParams,\n GoogleAbstractedClient,\n} from '@langchain/google-common';\nimport type { BaseMessage } from '@langchain/core/messages';\nimport type { VertexAIClientOptions } from '@/types';\n\nclass CustomChatConnection extends ChatConnection<VertexAIClientOptions> {\n async formatData(\n input: BaseMessage[],\n parameters: GoogleAIModelRequestParams\n ): Promise<unknown> {\n const formattedData = (await super.formatData(\n input,\n parameters\n )) as GeminiRequest;\n if (\n formattedData.generationConfig?.thinkingConfig?.thinkingBudget === -1 &&\n formattedData.generationConfig.thinkingConfig.includeThoughts === false\n ) {\n formattedData.generationConfig.thinkingConfig.includeThoughts = true;\n }\n return formattedData;\n }\n}\n\n/**\n * Integration with Google Vertex AI chat models.\n *\n * Setup:\n * Install `@langchain/google-vertexai` and set your stringified\n * Vertex AI credentials as an environment variable named `GOOGLE_APPLICATION_CREDENTIALS`.\n *\n * ```bash\n * npm install @langchain/google-vertexai\n * export GOOGLE_APPLICATION_CREDENTIALS=\"path/to/credentials\"\n * ```\n *\n * ## [Constructor args](https://api.js.langchain.com/classes/_langchain_google_vertexai.index.ChatVertexAI.html#constructor.new_ChatVertexAI)\n *\n * ## [Runtime args](https://api.js.langchain.com/interfaces/langchain_google_common_types.GoogleAIBaseLanguageModelCallOptions.html)\n *\n * Runtime args can be passed as the second argument to any of the base runnable methods `.invoke`. `.stream`, `.batch`, etc.\n * They can also be passed via `.withConfig`, or the second arg in `.bindTools`, like shown in the examples below:\n *\n * ```typescript\n * // When calling `.withConfig`, call options should be passed via the first argument\n * const llmWithArgsBound = llm.withConfig({\n * stop: [\"\\n\"],\n * tools: [...],\n * });\n *\n * // When calling `.bindTools`, call options should be passed via the second argument\n * const llmWithTools = llm.bindTools(\n * [...],\n * {\n * tool_choice: \"auto\",\n * }\n * );\n * ```\n *\n * ## Examples\n *\n * <details open>\n * <summary><strong>Instantiate</strong></summary>\n *\n * ```typescript\n * import { ChatVertexAI } from '@langchain/google-vertexai';\n *\n * const llm = new ChatVertexAI({\n * model: \"gemini-1.5-pro\",\n * temperature: 0,\n * // other params...\n * });\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Invoking</strong></summary>\n *\n * ```typescript\n * const input = `Translate \"I love programming\" into French.`;\n *\n * // Models also accept a list of chat messages or a formatted prompt\n * const result = await llm.invoke(input);\n * console.log(result);\n * ```\n *\n * ```txt\n * AIMessageChunk {\n * \"content\": \"\\\"J'adore programmer\\\" \\n\\nHere's why this is the best translation:\\n\\n* **J'adore** means \\\"I love\\\" and conveys a strong passion.\\n* **Programmer** is the French verb for \\\"to program.\\\"\\n\\nThis translation is natural and idiomatic in French. \\n\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": [],\n * \"usage_metadata\": {\n * \"input_tokens\": 9,\n * \"output_tokens\": 63,\n * \"total_tokens\": 72\n * }\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Streaming Chunks</strong></summary>\n *\n * ```typescript\n * for await (const chunk of await llm.stream(input)) {\n * console.log(chunk);\n * }\n * ```\n *\n * ```txt\n * AIMessageChunk {\n * \"content\": \"\\\"\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"J'adore programmer\\\" \\n\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": [],\n * \"usage_metadata\": {\n * \"input_tokens\": 9,\n * \"output_tokens\": 8,\n * \"total_tokens\": 17\n * }\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Aggregate Streamed Chunks</strong></summary>\n *\n * ```typescript\n * import { AIMessageChunk } from '@langchain/core/messages';\n * import { concat } from '@langchain/core/utils/stream';\n *\n * const stream = await llm.stream(input);\n * let full: AIMessageChunk | undefined;\n * for await (const chunk of stream) {\n * full = !full ? chunk : concat(full, chunk);\n * }\n * console.log(full);\n * ```\n *\n * ```txt\n * AIMessageChunk {\n * \"content\": \"\\\"J'adore programmer\\\" \\n\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": [],\n * \"usage_metadata\": {\n * \"input_tokens\": 9,\n * \"output_tokens\": 8,\n * \"total_tokens\": 17\n * }\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Bind tools</strong></summary>\n *\n * ```typescript\n * import { z } from 'zod';\n *\n * const GetWeather = {\n * name: \"GetWeather\",\n * description: \"Get the current weather in a given location\",\n * schema: z.object({\n * location: z.string().describe(\"The city and state, e.g. San Francisco, CA\")\n * }),\n * }\n *\n * const GetPopulation = {\n * name: \"GetPopulation\",\n * description: \"Get the current population in a given location\",\n * schema: z.object({\n * location: z.string().describe(\"The city and state, e.g. San Francisco, CA\")\n * }),\n * }\n *\n * const llmWithTools = llm.bindTools([GetWeather, GetPopulation]);\n * const aiMsg = await llmWithTools.invoke(\n * \"Which city is hotter today and which is bigger: LA or NY?\"\n * );\n * console.log(aiMsg.tool_calls);\n * ```\n *\n * ```txt\n * [\n * {\n * name: 'GetPopulation',\n * args: { location: 'New York City, NY' },\n * id: '33c1c1f47e2f492799c77d2800a43912',\n * type: 'tool_call'\n * }\n * ]\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Structured Output</strong></summary>\n *\n * ```typescript\n * import { z } from 'zod';\n *\n * const Joke = z.object({\n * setup: z.string().describe(\"The setup of the joke\"),\n * punchline: z.string().describe(\"The punchline to the joke\"),\n * rating: z.number().optional().describe(\"How funny the joke is, from 1 to 10\")\n * }).describe('Joke to tell user.');\n *\n * const structuredLlm = llm.withStructuredOutput(Joke, { name: \"Joke\" });\n * const jokeResult = await structuredLlm.invoke(\"Tell me a joke about cats\");\n * console.log(jokeResult);\n * ```\n *\n * ```txt\n * {\n * setup: 'What do you call a cat that loves to bowl?',\n * punchline: 'An alley cat!'\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Usage Metadata</strong></summary>\n *\n * ```typescript\n * const aiMsgForMetadata = await llm.invoke(input);\n * console.log(aiMsgForMetadata.usage_metadata);\n * ```\n *\n * ```txt\n * { input_tokens: 9, output_tokens: 8, total_tokens: 17 }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Stream Usage Metadata</strong></summary>\n *\n * ```typescript\n * const streamForMetadata = await llm.stream(\n * input,\n * {\n * streamUsage: true\n * }\n * );\n * let fullForMetadata: AIMessageChunk | undefined;\n * for await (const chunk of streamForMetadata) {\n * fullForMetadata = !fullForMetadata ? chunk : concat(fullForMetadata, chunk);\n * }\n * console.log(fullForMetadata?.usage_metadata);\n * ```\n *\n * ```txt\n * { input_tokens: 9, output_tokens: 8, total_tokens: 17 }\n * ```\n * </details>\n *\n * <br />\n */\nexport class ChatVertexAI extends ChatGoogle {\n lc_namespace = ['langchain', 'chat_models', 'vertexai'];\n dynamicThinkingBudget = false;\n\n static lc_name(): 'LibreChatVertexAI' {\n return 'LibreChatVertexAI';\n }\n\n constructor(fields?: VertexAIClientOptions) {\n let dynamicThinkingBudget = false;\n if (fields?.thinkingBudget === -1) {\n dynamicThinkingBudget = true;\n fields.thinkingBudget = 1;\n }\n super({\n ...fields,\n platformType: 'gcp',\n });\n this.dynamicThinkingBudget = dynamicThinkingBudget;\n }\n invocationParams(\n options?: this['ParsedCallOptions'] | undefined\n ): GoogleAIModelRequestParams {\n const params = super.invocationParams(options);\n if (this.dynamicThinkingBudget) {\n params.maxReasoningTokens = -1;\n }\n return params;\n }\n\n buildConnection(\n fields: VertexAIClientOptions,\n client: GoogleAbstractedClient\n ): void {\n this.connection = new CustomChatConnection(\n { ...fields, ...this },\n this.caller,\n client,\n false\n );\n\n this.streamedConnection = new CustomChatConnection(\n { ...fields, ...this },\n this.caller,\n client,\n true\n );\n }\n}\n"],"names":[],"mappings":";;;AAUA,MAAM,oBAAqB,SAAQ,cAAqC,CAAA;AACtE,IAAA,MAAM,UAAU,CACd,KAAoB,EACpB,UAAsC,EAAA;AAEtC,QAAA,MAAM,aAAa,IAAI,MAAM,KAAK,CAAC,UAAU,CAC3C,KAAK,EACL,UAAU,CACX,CAAkB;QACnB,IACE,aAAa,CAAC,gBAAgB,EAAE,cAAc,EAAE,cAAc,KAAK,EAAE;YACrE,aAAa,CAAC,gBAAgB,CAAC,cAAc,CAAC,eAAe,KAAK,KAAK,EACvE;YACA,aAAa,CAAC,gBAAgB,CAAC,cAAc,CAAC,eAAe,GAAG,IAAI;;AAEtE,QAAA,OAAO,aAAa;;AAEvB;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAyRG;AACG,MAAO,YAAa,SAAQ,UAAU,CAAA;IAC1C,YAAY,GAAG,CAAC,WAAW,EAAE,aAAa,EAAE,UAAU,CAAC;IACvD,qBAAqB,GAAG,KAAK;AAE7B,IAAA,OAAO,OAAO,GAAA;AACZ,QAAA,OAAO,mBAAmB;;AAG5B,IAAA,WAAA,CAAY,MAA8B,EAAA;QACxC,IAAI,qBAAqB,GAAG,KAAK;AACjC,QAAA,IAAI,MAAM,EAAE,cAAc,KAAK,EAAE,EAAE;YACjC,qBAAqB,GAAG,IAAI;AAC5B,YAAA,MAAM,CAAC,cAAc,GAAG,CAAC;;AAE3B,QAAA,KAAK,CAAC;AACJ,YAAA,GAAG,MAAM;AACT,YAAA,YAAY,EAAE,KAAK;AACpB,SAAA,CAAC;AACF,QAAA,IAAI,CAAC,qBAAqB,GAAG,qBAAqB;;AAEpD,IAAA,gBAAgB,CACd,OAA+C,EAAA;QAE/C,MAAM,MAAM,GAAG,KAAK,CAAC,gBAAgB,CAAC,OAAO,CAAC;AAC9C,QAAA,IAAI,IAAI,CAAC,qBAAqB,EAAE;AAC9B,YAAA,MAAM,CAAC,kBAAkB,GAAG,EAAE;;AAEhC,QAAA,OAAO,MAAM;;IAGf,eAAe,CACb,MAA6B,EAC7B,MAA8B,EAAA;QAE9B,IAAI,CAAC,UAAU,GAAG,IAAI,oBAAoB,CACxC,EAAE,GAAG,MAAM,EAAE,GAAG,IAAI,EAAE,EACtB,IAAI,CAAC,MAAM,EACX,MAAM,EACN,KAAK,CACN;QAED,IAAI,CAAC,kBAAkB,GAAG,IAAI,oBAAoB,CAChD,EAAE,GAAG,MAAM,EAAE,GAAG,IAAI,EAAE,EACtB,IAAI,CAAC,MAAM,EACX,MAAM,EACN,IAAI,CACL;;AAEJ;;;;"}
1
+ {"version":3,"file":"index.mjs","sources":["../../../../src/llm/vertexai/index.ts"],"sourcesContent":["import { ChatGoogle } from '@langchain/google-gauth';\nimport { ChatConnection } from '@langchain/google-common';\nimport type {\n GeminiRequest,\n GoogleAIModelRequestParams,\n GoogleAbstractedClient,\n} from '@langchain/google-common';\nimport type { BaseMessage } from '@langchain/core/messages';\nimport type { VertexAIClientOptions } from '@/types';\n\nclass CustomChatConnection extends ChatConnection<VertexAIClientOptions> {\n async formatData(\n input: BaseMessage[],\n parameters: GoogleAIModelRequestParams\n ): Promise<unknown> {\n const formattedData = (await super.formatData(\n input,\n parameters\n )) as GeminiRequest;\n if (formattedData.generationConfig?.thinkingConfig?.thinkingBudget === -1) {\n // -1 means \"let the model decide\" - delete the property so the API doesn't receive an invalid value\n if (\n formattedData.generationConfig.thinkingConfig.includeThoughts === false\n ) {\n formattedData.generationConfig.thinkingConfig.includeThoughts = true;\n }\n delete formattedData.generationConfig.thinkingConfig.thinkingBudget;\n }\n return formattedData;\n }\n}\n\n/**\n * Integration with Google Vertex AI chat models.\n *\n * Setup:\n * Install `@langchain/google-vertexai` and set your stringified\n * Vertex AI credentials as an environment variable named `GOOGLE_APPLICATION_CREDENTIALS`.\n *\n * ```bash\n * npm install @langchain/google-vertexai\n * export GOOGLE_APPLICATION_CREDENTIALS=\"path/to/credentials\"\n * ```\n *\n * ## [Constructor args](https://api.js.langchain.com/classes/_langchain_google_vertexai.index.ChatVertexAI.html#constructor.new_ChatVertexAI)\n *\n * ## [Runtime args](https://api.js.langchain.com/interfaces/langchain_google_common_types.GoogleAIBaseLanguageModelCallOptions.html)\n *\n * Runtime args can be passed as the second argument to any of the base runnable methods `.invoke`. `.stream`, `.batch`, etc.\n * They can also be passed via `.withConfig`, or the second arg in `.bindTools`, like shown in the examples below:\n *\n * ```typescript\n * // When calling `.withConfig`, call options should be passed via the first argument\n * const llmWithArgsBound = llm.withConfig({\n * stop: [\"\\n\"],\n * tools: [...],\n * });\n *\n * // When calling `.bindTools`, call options should be passed via the second argument\n * const llmWithTools = llm.bindTools(\n * [...],\n * {\n * tool_choice: \"auto\",\n * }\n * );\n * ```\n *\n * ## Examples\n *\n * <details open>\n * <summary><strong>Instantiate</strong></summary>\n *\n * ```typescript\n * import { ChatVertexAI } from '@langchain/google-vertexai';\n *\n * const llm = new ChatVertexAI({\n * model: \"gemini-1.5-pro\",\n * temperature: 0,\n * // other params...\n * });\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Invoking</strong></summary>\n *\n * ```typescript\n * const input = `Translate \"I love programming\" into French.`;\n *\n * // Models also accept a list of chat messages or a formatted prompt\n * const result = await llm.invoke(input);\n * console.log(result);\n * ```\n *\n * ```txt\n * AIMessageChunk {\n * \"content\": \"\\\"J'adore programmer\\\" \\n\\nHere's why this is the best translation:\\n\\n* **J'adore** means \\\"I love\\\" and conveys a strong passion.\\n* **Programmer** is the French verb for \\\"to program.\\\"\\n\\nThis translation is natural and idiomatic in French. \\n\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": [],\n * \"usage_metadata\": {\n * \"input_tokens\": 9,\n * \"output_tokens\": 63,\n * \"total_tokens\": 72\n * }\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Streaming Chunks</strong></summary>\n *\n * ```typescript\n * for await (const chunk of await llm.stream(input)) {\n * console.log(chunk);\n * }\n * ```\n *\n * ```txt\n * AIMessageChunk {\n * \"content\": \"\\\"\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"J'adore programmer\\\" \\n\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {},\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": [],\n * \"usage_metadata\": {\n * \"input_tokens\": 9,\n * \"output_tokens\": 8,\n * \"total_tokens\": 17\n * }\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Aggregate Streamed Chunks</strong></summary>\n *\n * ```typescript\n * import { AIMessageChunk } from '@langchain/core/messages';\n * import { concat } from '@langchain/core/utils/stream';\n *\n * const stream = await llm.stream(input);\n * let full: AIMessageChunk | undefined;\n * for await (const chunk of stream) {\n * full = !full ? chunk : concat(full, chunk);\n * }\n * console.log(full);\n * ```\n *\n * ```txt\n * AIMessageChunk {\n * \"content\": \"\\\"J'adore programmer\\\" \\n\",\n * \"additional_kwargs\": {},\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": [],\n * \"usage_metadata\": {\n * \"input_tokens\": 9,\n * \"output_tokens\": 8,\n * \"total_tokens\": 17\n * }\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Bind tools</strong></summary>\n *\n * ```typescript\n * import { z } from 'zod';\n *\n * const GetWeather = {\n * name: \"GetWeather\",\n * description: \"Get the current weather in a given location\",\n * schema: z.object({\n * location: z.string().describe(\"The city and state, e.g. San Francisco, CA\")\n * }),\n * }\n *\n * const GetPopulation = {\n * name: \"GetPopulation\",\n * description: \"Get the current population in a given location\",\n * schema: z.object({\n * location: z.string().describe(\"The city and state, e.g. San Francisco, CA\")\n * }),\n * }\n *\n * const llmWithTools = llm.bindTools([GetWeather, GetPopulation]);\n * const aiMsg = await llmWithTools.invoke(\n * \"Which city is hotter today and which is bigger: LA or NY?\"\n * );\n * console.log(aiMsg.tool_calls);\n * ```\n *\n * ```txt\n * [\n * {\n * name: 'GetPopulation',\n * args: { location: 'New York City, NY' },\n * id: '33c1c1f47e2f492799c77d2800a43912',\n * type: 'tool_call'\n * }\n * ]\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Structured Output</strong></summary>\n *\n * ```typescript\n * import { z } from 'zod';\n *\n * const Joke = z.object({\n * setup: z.string().describe(\"The setup of the joke\"),\n * punchline: z.string().describe(\"The punchline to the joke\"),\n * rating: z.number().optional().describe(\"How funny the joke is, from 1 to 10\")\n * }).describe('Joke to tell user.');\n *\n * const structuredLlm = llm.withStructuredOutput(Joke, { name: \"Joke\" });\n * const jokeResult = await structuredLlm.invoke(\"Tell me a joke about cats\");\n * console.log(jokeResult);\n * ```\n *\n * ```txt\n * {\n * setup: 'What do you call a cat that loves to bowl?',\n * punchline: 'An alley cat!'\n * }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Usage Metadata</strong></summary>\n *\n * ```typescript\n * const aiMsgForMetadata = await llm.invoke(input);\n * console.log(aiMsgForMetadata.usage_metadata);\n * ```\n *\n * ```txt\n * { input_tokens: 9, output_tokens: 8, total_tokens: 17 }\n * ```\n * </details>\n *\n * <br />\n *\n * <details>\n * <summary><strong>Stream Usage Metadata</strong></summary>\n *\n * ```typescript\n * const streamForMetadata = await llm.stream(\n * input,\n * {\n * streamUsage: true\n * }\n * );\n * let fullForMetadata: AIMessageChunk | undefined;\n * for await (const chunk of streamForMetadata) {\n * fullForMetadata = !fullForMetadata ? chunk : concat(fullForMetadata, chunk);\n * }\n * console.log(fullForMetadata?.usage_metadata);\n * ```\n *\n * ```txt\n * { input_tokens: 9, output_tokens: 8, total_tokens: 17 }\n * ```\n * </details>\n *\n * <br />\n */\nexport class ChatVertexAI extends ChatGoogle {\n lc_namespace = ['langchain', 'chat_models', 'vertexai'];\n dynamicThinkingBudget = false;\n\n static lc_name(): 'LibreChatVertexAI' {\n return 'LibreChatVertexAI';\n }\n\n constructor(fields?: VertexAIClientOptions) {\n const dynamicThinkingBudget = fields?.thinkingBudget === -1;\n super({\n ...fields,\n platformType: 'gcp',\n });\n this.dynamicThinkingBudget = dynamicThinkingBudget;\n }\n invocationParams(\n options?: this['ParsedCallOptions'] | undefined\n ): GoogleAIModelRequestParams {\n const params = super.invocationParams(options);\n if (this.dynamicThinkingBudget) {\n params.maxReasoningTokens = -1;\n }\n return params;\n }\n\n buildConnection(\n fields: VertexAIClientOptions,\n client: GoogleAbstractedClient\n ): void {\n this.connection = new CustomChatConnection(\n { ...fields, ...this },\n this.caller,\n client,\n false\n );\n\n this.streamedConnection = new CustomChatConnection(\n { ...fields, ...this },\n this.caller,\n client,\n true\n );\n }\n}\n"],"names":[],"mappings":";;;AAUA,MAAM,oBAAqB,SAAQ,cAAqC,CAAA;AACtE,IAAA,MAAM,UAAU,CACd,KAAoB,EACpB,UAAsC,EAAA;AAEtC,QAAA,MAAM,aAAa,IAAI,MAAM,KAAK,CAAC,UAAU,CAC3C,KAAK,EACL,UAAU,CACX,CAAkB;QACnB,IAAI,aAAa,CAAC,gBAAgB,EAAE,cAAc,EAAE,cAAc,KAAK,EAAE,EAAE;;YAEzE,IACE,aAAa,CAAC,gBAAgB,CAAC,cAAc,CAAC,eAAe,KAAK,KAAK,EACvE;gBACA,aAAa,CAAC,gBAAgB,CAAC,cAAc,CAAC,eAAe,GAAG,IAAI;;AAEtE,YAAA,OAAO,aAAa,CAAC,gBAAgB,CAAC,cAAc,CAAC,cAAc;;AAErE,QAAA,OAAO,aAAa;;AAEvB;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAyRG;AACG,MAAO,YAAa,SAAQ,UAAU,CAAA;IAC1C,YAAY,GAAG,CAAC,WAAW,EAAE,aAAa,EAAE,UAAU,CAAC;IACvD,qBAAqB,GAAG,KAAK;AAE7B,IAAA,OAAO,OAAO,GAAA;AACZ,QAAA,OAAO,mBAAmB;;AAG5B,IAAA,WAAA,CAAY,MAA8B,EAAA;QACxC,MAAM,qBAAqB,GAAG,MAAM,EAAE,cAAc,KAAK,EAAE;AAC3D,QAAA,KAAK,CAAC;AACJ,YAAA,GAAG,MAAM;AACT,YAAA,YAAY,EAAE,KAAK;AACpB,SAAA,CAAC;AACF,QAAA,IAAI,CAAC,qBAAqB,GAAG,qBAAqB;;AAEpD,IAAA,gBAAgB,CACd,OAA+C,EAAA;QAE/C,MAAM,MAAM,GAAG,KAAK,CAAC,gBAAgB,CAAC,OAAO,CAAC;AAC9C,QAAA,IAAI,IAAI,CAAC,qBAAqB,EAAE;AAC9B,YAAA,MAAM,CAAC,kBAAkB,GAAG,EAAE;;AAEhC,QAAA,OAAO,MAAM;;IAGf,eAAe,CACb,MAA6B,EAC7B,MAA8B,EAAA;QAE9B,IAAI,CAAC,UAAU,GAAG,IAAI,oBAAoB,CACxC,EAAE,GAAG,MAAM,EAAE,GAAG,IAAI,EAAE,EACtB,IAAI,CAAC,MAAM,EACX,MAAM,EACN,KAAK,CACN;QAED,IAAI,CAAC,kBAAkB,GAAG,IAAI,oBAAoB,CAChD,EAAE,GAAG,MAAM,EAAE,GAAG,IAAI,EAAE,EACtB,IAAI,CAAC,MAAM,EACX,MAAM,EACN,IAAI,CACL;;AAEJ;;;;"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@librechat/agents",
3
- "version": "3.0.51",
3
+ "version": "3.0.52",
4
4
  "main": "./dist/cjs/main.cjs",
5
5
  "module": "./dist/esm/main.mjs",
6
6
  "types": "./dist/types/index.d.ts",
package/src/llm/vertexai/index.ts CHANGED
@@ -17,11 +17,14 @@ class CustomChatConnection extends ChatConnection<VertexAIClientOptions> {
17
17
  input,
18
18
  parameters
19
19
  )) as GeminiRequest;
20
- if (
21
- formattedData.generationConfig?.thinkingConfig?.thinkingBudget === -1 &&
22
- formattedData.generationConfig.thinkingConfig.includeThoughts === false
23
- ) {
24
- formattedData.generationConfig.thinkingConfig.includeThoughts = true;
20
+ if (formattedData.generationConfig?.thinkingConfig?.thinkingBudget === -1) {
21
+ // -1 means "let the model decide" - delete the property so the API doesn't receive an invalid value
22
+ if (
23
+ formattedData.generationConfig.thinkingConfig.includeThoughts === false
24
+ ) {
25
+ formattedData.generationConfig.thinkingConfig.includeThoughts = true;
26
+ }
27
+ delete formattedData.generationConfig.thinkingConfig.thinkingBudget;
25
28
  }
26
29
  return formattedData;
27
30
  }
@@ -318,11 +321,7 @@ export class ChatVertexAI extends ChatGoogle {
318
321
  }
319
322
 
320
323
  constructor(fields?: VertexAIClientOptions) {
321
- let dynamicThinkingBudget = false;
322
- if (fields?.thinkingBudget === -1) {
323
- dynamicThinkingBudget = true;
324
- fields.thinkingBudget = 1;
325
- }
324
+ const dynamicThinkingBudget = fields?.thinkingBudget === -1;
326
325
  super({
327
326
  ...fields,
328
327
  platformType: 'gcp',
package/src/utils/llmConfig.ts CHANGED
@@ -147,9 +147,14 @@ export const llmConfigs: Record<string, t.LLMConfig | undefined> = {
147
147
  [Providers.VERTEXAI]: {
148
148
  provider: Providers.VERTEXAI,
149
149
  model: 'gemini-2.5-flash',
150
+ // model: 'gemini-2.5-pro',
150
151
  streaming: true,
151
152
  streamUsage: true,
152
153
  keyFile: process.env.VERTEXAI_KEY_FILE,
154
+ // maxRetries: 2,
155
+ // location: 'global',
156
+ // thinkingBudget: -1,
157
+ // includeThoughts: true,
153
158
  } as t.VertexAIClientOptions & t.LLMConfig,
154
159
  [Providers.GOOGLE]: {
155
160
  provider: Providers.GOOGLE,