@langchain/deepseek 1.0.7 → 1.0.8

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 (6) hide show
  1. package/CHANGELOG.md +22 -0
  2. package/dist/chat_models.cjs +1 -0
  3. package/dist/chat_models.cjs.map +1 -1
  4. package/dist/chat_models.js +1 -0
  5. package/dist/chat_models.js.map +1 -1
  6. package/package.json +4 -4
package/CHANGELOG.md CHANGED
@@ -1,5 +1,27 @@
1
1
  # @langchain/deepseek
2
2
 
3
+ ## 1.0.8
4
+
5
+ ### Patch Changes
6
+
7
+ - [#9900](https://github.com/langchain-ai/langchainjs/pull/9900) [`a9b5059`](https://github.com/langchain-ai/langchainjs/commit/a9b50597186002221aaa4585246e569fa44c27c8) Thanks [@hntrl](https://github.com/hntrl)! - Improved abort signal handling for chat models:
8
+ - Added `ModelAbortError` class in `@langchain/core/errors` that contains partial output when a model invocation is aborted mid-stream
9
+ - `invoke()` now throws `ModelAbortError` with accumulated `partialOutput` when aborted during streaming (when using streaming callback handlers)
10
+ - `stream()` throws a regular `AbortError` when aborted (since chunks are already yielded to the caller)
11
+ - All provider implementations now properly check and propagate abort signals in both `_generate()` and `_streamResponseChunks()` methods
12
+ - Added standard tests for abort signal behavior
13
+
14
+ - [#9900](https://github.com/langchain-ai/langchainjs/pull/9900) [`a9b5059`](https://github.com/langchain-ai/langchainjs/commit/a9b50597186002221aaa4585246e569fa44c27c8) Thanks [@hntrl](https://github.com/hntrl)! - fix(providers): add proper abort signal handling for invoke and stream operations
15
+ - Added early abort check (`signal.throwIfAborted()`) at the start of `_generate` methods to immediately throw when signal is already aborted
16
+ - Added abort signal checks inside streaming loops in `_streamResponseChunks` to return early when signal is aborted
17
+ - Propagated abort signals to underlying SDK calls where applicable (Google GenAI, Google Common/VertexAI, Cohere)
18
+ - Added standard tests for abort signal behavior in `@langchain/standard-tests`
19
+
20
+ This enables proper cancellation behavior for both invoke and streaming operations, and allows fallback chains to correctly proceed to the next runnable when the previous one is aborted.
21
+
22
+ - Updated dependencies [[`1fa865b`](https://github.com/langchain-ai/langchainjs/commit/1fa865b1cb8a30c2269b83cdb5fc84d374c3fca9), [`28efb57`](https://github.com/langchain-ai/langchainjs/commit/28efb57448933368094ca41c63d9262ac0f348a6), [`4e42452`](https://github.com/langchain-ai/langchainjs/commit/4e42452e4c020408bd6687667e931497b05aaff5), [`a9b5059`](https://github.com/langchain-ai/langchainjs/commit/a9b50597186002221aaa4585246e569fa44c27c8), [`a9b5059`](https://github.com/langchain-ai/langchainjs/commit/a9b50597186002221aaa4585246e569fa44c27c8)]:
23
+ - @langchain/openai@1.2.4
24
+
3
25
  ## 1.0.7
4
26
 
5
27
  ### Patch Changes
package/dist/chat_models.cjs CHANGED
@@ -398,6 +398,7 @@ var ChatDeepSeek = class extends __langchain_openai.ChatOpenAICompletions {
398
398
  let tokensBuffer = "";
399
399
  let isThinking = false;
400
400
  for await (const chunk of stream) {
401
+ if (options.signal?.aborted) return;
401
402
  if (chunk.message.additional_kwargs.reasoning_content) {
402
403
  yield chunk;
403
404
  continue;
package/dist/chat_models.cjs.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"file":"chat_models.cjs","names":["ChatOpenAICompletions","fields?: Partial<ChatDeepSeekInput>","delta: Record<string, any>","rawResponse: OpenAIClient.ChatCompletionChunk","defaultRole?:\n | \"function\"\n | \"user\"\n | \"system\"\n | \"developer\"\n | \"assistant\"\n | \"tool\"","messages: BaseMessage[]","options: this[\"ParsedCallOptions\"]","runManager?: CallbackManagerForLLMRun","ChatGenerationChunk","AIMessageChunk","message: OpenAIClient.ChatCompletionMessage","rawResponse: OpenAIClient.ChatCompletion","PROFILES","outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>","config?: StructuredOutputMethodOptions<boolean>"],"sources":["../src/chat_models.ts"],"sourcesContent":["import {\n BaseLanguageModelInput,\n StructuredOutputMethodOptions,\n} from \"@langchain/core/language_models/base\";\nimport { ModelProfile } from \"@langchain/core/language_models/profile\";\nimport { BaseMessage, AIMessageChunk } from \"@langchain/core/messages\";\nimport { Runnable } from \"@langchain/core/runnables\";\nimport { getEnvironmentVariable } from \"@langchain/core/utils/env\";\nimport { InteropZodType } from \"@langchain/core/utils/types\";\nimport {\n ChatOpenAICallOptions,\n ChatOpenAICompletions,\n ChatOpenAIFields,\n OpenAIClient,\n} from \"@langchain/openai\";\nimport { ChatGenerationChunk } from \"@langchain/core/outputs\";\nimport { CallbackManagerForLLMRun } from \"@langchain/core/callbacks/manager\";\nimport PROFILES from \"./profiles.js\";\n\nexport interface ChatDeepSeekCallOptions extends ChatOpenAICallOptions {\n headers?: Record<string, string>;\n}\n\nexport interface ChatDeepSeekInput extends ChatOpenAIFields {\n /**\n * The Deepseek API key to use for requests.\n * @default process.env.DEEPSEEK_API_KEY\n */\n apiKey?: string;\n /**\n * The name of the model to use.\n */\n model?: string;\n /**\n * Up to 4 sequences where the API will stop generating further tokens. The\n * returned text will not contain the stop sequence.\n * Alias for `stopSequences`\n */\n stop?: Array<string>;\n /**\n * Up to 4 sequences where the API will stop generating further tokens. The\n * returned text will not contain the stop sequence.\n */\n stopSequences?: Array<string>;\n /**\n * Whether or not to stream responses.\n */\n streaming?: boolean;\n /**\n * The temperature to use for sampling.\n */\n temperature?: number;\n /**\n * The maximum number of tokens that the model can process in a single response.\n * This limits ensures computational efficiency and resource management.\n */\n maxTokens?: number;\n}\n\n/**\n * Deepseek chat model integration.\n *\n * The Deepseek API is compatible to the OpenAI API with some limitations.\n *\n * Setup:\n * Install `@langchain/deepseek` and set an environment variable named `DEEPSEEK_API_KEY`.\n *\n * ```bash\n * npm install @langchain/deepseek\n * export DEEPSEEK_API_KEY=\"your-api-key\"\n * ```\n *\n * ## [Constructor args](https://api.js.langchain.com/classes/_langchain_deepseek.ChatDeepSeek.html#constructor)\n *\n * ## [Runtime args](https://api.js.langchain.com/interfaces/_langchain_deepseek.ChatDeepSeekCallOptions.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 { ChatDeepSeek } from '@langchain/deepseek';\n *\n * const llm = new ChatDeepSeek({\n * model: \"deepseek-reasoner\",\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 * AIMessage {\n * \"content\": \"The French translation of \\\"I love programming\\\" is \\\"J'aime programmer\\\". In this sentence, \\\"J'aime\\\" is the first person singular conjugation of the French verb \\\"aimer\\\" which means \\\"to love\\\", and \\\"programmer\\\" is the French infinitive for \\\"to program\\\". I hope this helps! Let me know if you have any other questions.\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"tokenUsage\": {\n * \"completionTokens\": 82,\n * \"promptTokens\": 20,\n * \"totalTokens\": 102\n * },\n * \"finish_reason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"invalid_tool_calls\": []\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 * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"The\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" French\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" translation\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" of\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" \\\"\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"I\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" love\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * ...\n * AIMessageChunk {\n * \"content\": \".\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\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\": \"The French translation of \\\"I love programming\\\" is \\\"J'aime programmer\\\". In this sentence, \\\"J'aime\\\" is the first person singular conjugation of the French verb \\\"aimer\\\" which means \\\"to love\\\", and \\\"programmer\\\" is the French infinitive for \\\"to program\\\". I hope this helps! Let me know if you have any other questions.\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\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 llmForToolCalling = new ChatDeepSeek({\n * model: \"deepseek-chat\",\n * temperature: 0,\n * // other params...\n * });\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 = llmForToolCalling.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: 'GetWeather',\n * args: { location: 'Los Angeles, CA' },\n * type: 'tool_call',\n * id: 'call_cd34'\n * },\n * {\n * name: 'GetWeather',\n * args: { location: 'New York, NY' },\n * type: 'tool_call',\n * id: 'call_68rf'\n * },\n * {\n * name: 'GetPopulation',\n * args: { location: 'Los Angeles, CA' },\n * type: 'tool_call',\n * id: 'call_f81z'\n * },\n * {\n * name: 'GetPopulation',\n * args: { location: 'New York, NY' },\n * type: 'tool_call',\n * id: 'call_8byt'\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 = llmForToolCalling.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: \"Why don't cats play poker in the wild?\",\n * punchline: 'Because there are too many cheetahs.'\n * }\n * ```\n * </details>\n *\n * <br />\n */\nexport class ChatDeepSeek extends ChatOpenAICompletions<ChatDeepSeekCallOptions> {\n static lc_name() {\n return \"ChatDeepSeek\";\n }\n\n _llmType() {\n return \"deepseek\";\n }\n\n get lc_secrets(): { [key: string]: string } | undefined {\n return {\n apiKey: \"DEEPSEEK_API_KEY\",\n };\n }\n\n lc_serializable = true;\n\n lc_namespace = [\"langchain\", \"chat_models\", \"deepseek\"];\n\n constructor(fields?: Partial<ChatDeepSeekInput>) {\n const apiKey = fields?.apiKey || getEnvironmentVariable(\"DEEPSEEK_API_KEY\");\n if (!apiKey) {\n throw new Error(\n `Deepseek API key not found. Please set the DEEPSEEK_API_KEY environment variable or pass the key into \"apiKey\" field.`\n );\n }\n\n super({\n ...fields,\n apiKey,\n configuration: {\n baseURL: \"https://api.deepseek.com\",\n ...fields?.configuration,\n },\n });\n }\n\n protected override _convertCompletionsDeltaToBaseMessageChunk(\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n delta: Record<string, any>,\n rawResponse: OpenAIClient.ChatCompletionChunk,\n defaultRole?:\n | \"function\"\n | \"user\"\n | \"system\"\n | \"developer\"\n | \"assistant\"\n | \"tool\"\n ) {\n const messageChunk = super._convertCompletionsDeltaToBaseMessageChunk(\n delta,\n rawResponse,\n defaultRole\n );\n messageChunk.additional_kwargs.reasoning_content = delta.reasoning_content;\n // Override model_provider for DeepSeek-specific block translation\n messageChunk.response_metadata = {\n ...messageChunk.response_metadata,\n model_provider: \"deepseek\",\n };\n return messageChunk;\n }\n\n async *_streamResponseChunks(\n messages: BaseMessage[],\n options: this[\"ParsedCallOptions\"],\n runManager?: CallbackManagerForLLMRun\n ): AsyncGenerator<ChatGenerationChunk> {\n const stream = super._streamResponseChunks(messages, options, runManager);\n\n // State for parsing <think> tags\n let tokensBuffer = \"\";\n let isThinking = false;\n\n for await (const chunk of stream) {\n // If the model already provided reasoning_content natively, just yield it\n if (chunk.message.additional_kwargs.reasoning_content) {\n yield chunk;\n continue;\n }\n\n const text = chunk.text;\n if (!text) {\n yield chunk;\n continue;\n }\n\n // Append text to buffer to handle split tags\n tokensBuffer += text;\n\n // Check for <think> start tag\n if (!isThinking && tokensBuffer.includes(\"<think>\")) {\n isThinking = true;\n const thinkIndex = tokensBuffer.indexOf(\"<think>\");\n const beforeThink = tokensBuffer.substring(0, thinkIndex);\n const afterThink = tokensBuffer.substring(\n thinkIndex + \"<think>\".length\n );\n\n // We consumed up to <think>, so buffer becomes what's after\n tokensBuffer = afterThink || \"\"; // might be empty or part of thought\n\n if (beforeThink) {\n // Send the content before the tag\n const newChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: beforeThink,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: beforeThink,\n generationInfo: chunk.generationInfo,\n });\n yield newChunk;\n }\n }\n\n // Check for </think> end tag\n if (isThinking && tokensBuffer.includes(\"</think>\")) {\n isThinking = false;\n const thinkEndIndex = tokensBuffer.indexOf(\"</think>\");\n const thoughtContent = tokensBuffer.substring(0, thinkEndIndex);\n const afterThink = tokensBuffer.substring(\n thinkEndIndex + \"</think>\".length\n );\n\n // Yield the reasoning content\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: {\n ...chunk.message.additional_kwargs,\n reasoning_content: thoughtContent,\n },\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: \"\",\n generationInfo: chunk.generationInfo,\n });\n yield reasoningChunk;\n\n // Reset buffer to what's after </think>\n tokensBuffer = afterThink || \"\";\n\n // Yield the rest as normal content if any\n if (tokensBuffer) {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: tokensBuffer,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: tokensBuffer,\n generationInfo: chunk.generationInfo,\n });\n yield contentChunk;\n tokensBuffer = \"\"; // consumed\n }\n } else if (isThinking) {\n // We are inside thinking block.\n // Check partial </think> match\n const possibleEndTag = \"</think>\";\n let splitIndex = -1;\n\n // Check if buffer ends with a prefix of </think> - Greedy check (longest first)\n for (let i = possibleEndTag.length - 1; i >= 1; i--) {\n if (tokensBuffer.endsWith(possibleEndTag.substring(0, i))) {\n splitIndex = tokensBuffer.length - i;\n break;\n }\n }\n\n if (splitIndex !== -1) {\n const safeToYield = tokensBuffer.substring(0, splitIndex);\n if (safeToYield) {\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: {\n ...chunk.message.additional_kwargs,\n reasoning_content: safeToYield,\n },\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: \"\",\n generationInfo: chunk.generationInfo,\n });\n yield reasoningChunk;\n }\n tokensBuffer = tokensBuffer.substring(splitIndex); // keep partial tag\n } else {\n // content is safe to yield as reasoning\n if (tokensBuffer) {\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: {\n ...chunk.message.additional_kwargs,\n reasoning_content: tokensBuffer,\n },\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: \"\",\n generationInfo: chunk.generationInfo,\n });\n yield reasoningChunk;\n tokensBuffer = \"\";\n }\n }\n } else {\n // NOT thinking.\n // Check partial start tag \"<think>\" - Greedy check (longest first)\n const possibleStartTag = \"<think>\";\n let splitIndex = -1;\n for (let i = possibleStartTag.length - 1; i >= 1; i--) {\n if (tokensBuffer.endsWith(possibleStartTag.substring(0, i))) {\n splitIndex = tokensBuffer.length - i;\n break;\n }\n }\n\n if (splitIndex !== -1) {\n // Yield safe content\n const safeToYield = tokensBuffer.substring(0, splitIndex);\n if (safeToYield) {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: safeToYield,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: safeToYield,\n generationInfo: chunk.generationInfo,\n });\n yield contentChunk;\n }\n tokensBuffer = tokensBuffer.substring(splitIndex); // keep partial tag\n } else {\n // Yield all\n if (tokensBuffer) {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: tokensBuffer,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: tokensBuffer,\n generationInfo: chunk.generationInfo,\n });\n yield contentChunk;\n tokensBuffer = \"\";\n }\n }\n }\n }\n\n // Flush remaining buffer at end of stream\n if (tokensBuffer) {\n // If we were thinking, it's unclosed thought.\n if (isThinking) {\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: { reasoning_content: tokensBuffer },\n }),\n text: \"\",\n });\n yield reasoningChunk;\n } else {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: tokensBuffer,\n }),\n text: tokensBuffer,\n });\n yield contentChunk;\n }\n }\n }\n\n protected override _convertCompletionsMessageToBaseMessage(\n message: OpenAIClient.ChatCompletionMessage,\n rawResponse: OpenAIClient.ChatCompletion\n ) {\n const langChainMessage = super._convertCompletionsMessageToBaseMessage(\n message,\n rawResponse\n );\n langChainMessage.additional_kwargs.reasoning_content =\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n (message as any).reasoning_content;\n // Override model_provider for DeepSeek-specific block translation\n langChainMessage.response_metadata = {\n ...langChainMessage.response_metadata,\n model_provider: \"deepseek\",\n };\n return langChainMessage;\n }\n\n /**\n * Return profiling information for the model.\n *\n * Provides information about the model's capabilities and constraints,\n * including token limits, multimodal support, and advanced features like\n * tool calling and structured output.\n *\n * @returns {ModelProfile} An object describing the model's capabilities and constraints\n *\n * @example\n * ```typescript\n * const model = new ChatDeepSeek({ model: \"deepseek-chat\" });\n * const profile = model.profile;\n * console.log(profile.maxInputTokens); // 128000\n * console.log(profile.imageInputs); // false\n * ```\n */\n get profile(): ModelProfile {\n return PROFILES[this.model] ?? {};\n }\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<false>\n ): Runnable<BaseLanguageModelInput, RunOutput>;\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<true>\n ): Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<boolean>\n ):\n | Runnable<BaseLanguageModelInput, RunOutput>\n | Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<boolean>\n ):\n | Runnable<BaseLanguageModelInput, RunOutput>\n | Runnable<\n BaseLanguageModelInput,\n { raw: BaseMessage; parsed: RunOutput }\n > {\n const ensuredConfig = { ...config };\n // Deepseek does not support json schema yet\n if (ensuredConfig?.method === undefined) {\n ensuredConfig.method = \"functionCalling\";\n }\n return super.withStructuredOutput<RunOutput>(outputSchema, ensuredConfig);\n }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAyZA,IAAa,eAAb,cAAkCA,yCAA+C;CAC/E,OAAO,UAAU;AACf,SAAO;CACR;CAED,WAAW;AACT,SAAO;CACR;CAED,IAAI,aAAoD;AACtD,SAAO,EACL,QAAQ,mBACT;CACF;CAED,kBAAkB;CAElB,eAAe;EAAC;EAAa;EAAe;CAAW;CAEvD,YAAYC,QAAqC;EAC/C,MAAM,SAAS,QAAQ,iEAAiC,mBAAmB;AAC3E,MAAI,CAAC,OACH,OAAM,IAAI,MACR,CAAC,qHAAqH,CAAC;EAI3H,MAAM;GACJ,GAAG;GACH;GACA,eAAe;IACb,SAAS;IACT,GAAG,QAAQ;GACZ;EACF,EAAC;CACH;CAED,AAAmB,2CAEjBC,OACAC,aACAC,aAOA;EACA,MAAM,eAAe,MAAM,2CACzB,OACA,aACA,YACD;EACD,aAAa,kBAAkB,oBAAoB,MAAM;EAEzD,aAAa,oBAAoB;GAC/B,GAAG,aAAa;GAChB,gBAAgB;EACjB;AACD,SAAO;CACR;CAED,OAAO,sBACLC,UACAC,SACAC,YACqC;EACrC,MAAM,SAAS,MAAM,sBAAsB,UAAU,SAAS,WAAW;EAGzE,IAAI,eAAe;EACnB,IAAI,aAAa;AAEjB,aAAW,MAAM,SAAS,QAAQ;AAEhC,OAAI,MAAM,QAAQ,kBAAkB,mBAAmB;IACrD,MAAM;AACN;GACD;GAED,MAAM,OAAO,MAAM;AACnB,OAAI,CAAC,MAAM;IACT,MAAM;AACN;GACD;GAGD,gBAAgB;AAGhB,OAAI,CAAC,cAAc,aAAa,SAAS,UAAU,EAAE;IACnD,aAAa;IACb,MAAM,aAAa,aAAa,QAAQ,UAAU;IAClD,MAAM,cAAc,aAAa,UAAU,GAAG,WAAW;IACzD,MAAM,aAAa,aAAa,UAC9B,aAAa,EACd;IAGD,eAAe,cAAc;AAE7B,QAAI,aAAa;KAEf,MAAM,WAAW,IAAIC,6CAAoB;MACvC,SAAS,IAAIC,yCAAe;OAC1B,SAAS;OACT,mBAAmB,MAAM,QAAQ;OACjC,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;IACP;GACF;AAGD,OAAI,cAAc,aAAa,SAAS,WAAW,EAAE;IACnD,aAAa;IACb,MAAM,gBAAgB,aAAa,QAAQ,WAAW;IACtD,MAAM,iBAAiB,aAAa,UAAU,GAAG,cAAc;IAC/D,MAAM,aAAa,aAAa,UAC9B,gBAAgB,EACjB;IAGD,MAAM,iBAAiB,IAAID,6CAAoB;KAC7C,SAAS,IAAIC,yCAAe;MAC1B,SAAS;MACT,mBAAmB;OACjB,GAAG,MAAM,QAAQ;OACjB,mBAAmB;MACpB;MACD,mBAAmB,MAAM,QAAQ;MACjC,YAAY,MAAM,QAAQ;MAC1B,kBAAkB,MAAM,QAAQ;MAChC,IAAI,MAAM,QAAQ;KACnB;KACD,MAAM;KACN,gBAAgB,MAAM;IACvB;IACD,MAAM;IAGN,eAAe,cAAc;AAG7B,QAAI,cAAc;KAChB,MAAM,eAAe,IAAID,6CAAoB;MAC3C,SAAS,IAAIC,yCAAe;OAC1B,SAAS;OACT,mBAAmB,MAAM,QAAQ;OACjC,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;KACN,eAAe;IAChB;GACF,WAAU,YAAY;IAGrB,MAAM,iBAAiB;IACvB,IAAI,aAAa;AAGjB,SAAK,IAAI,IAAI,GAA2B,KAAK,GAAG,IAC9C,KAAI,aAAa,SAAS,eAAe,UAAU,GAAG,EAAE,CAAC,EAAE;KACzD,aAAa,aAAa,SAAS;AACnC;IACD;AAGH,QAAI,eAAe,IAAI;KACrB,MAAM,cAAc,aAAa,UAAU,GAAG,WAAW;AACzD,SAAI,aAAa;MACf,MAAM,iBAAiB,IAAID,6CAAoB;OAC7C,SAAS,IAAIC,yCAAe;QAC1B,SAAS;QACT,mBAAmB;SACjB,GAAG,MAAM,QAAQ;SACjB,mBAAmB;QACpB;QACD,mBAAmB,MAAM,QAAQ;QACjC,YAAY,MAAM,QAAQ;QAC1B,kBAAkB,MAAM,QAAQ;QAChC,IAAI,MAAM,QAAQ;OACnB;OACD,MAAM;OACN,gBAAgB,MAAM;MACvB;MACD,MAAM;KACP;KACD,eAAe,aAAa,UAAU,WAAW;IAClD,WAEK,cAAc;KAChB,MAAM,iBAAiB,IAAID,6CAAoB;MAC7C,SAAS,IAAIC,yCAAe;OAC1B,SAAS;OACT,mBAAmB;QACjB,GAAG,MAAM,QAAQ;QACjB,mBAAmB;OACpB;OACD,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;KACN,eAAe;IAChB;GAEJ,OAAM;IAGL,MAAM,mBAAmB;IACzB,IAAI,aAAa;AACjB,SAAK,IAAI,IAAI,GAA6B,KAAK,GAAG,IAChD,KAAI,aAAa,SAAS,iBAAiB,UAAU,GAAG,EAAE,CAAC,EAAE;KAC3D,aAAa,aAAa,SAAS;AACnC;IACD;AAGH,QAAI,eAAe,IAAI;KAErB,MAAM,cAAc,aAAa,UAAU,GAAG,WAAW;AACzD,SAAI,aAAa;MACf,MAAM,eAAe,IAAID,6CAAoB;OAC3C,SAAS,IAAIC,yCAAe;QAC1B,SAAS;QACT,mBAAmB,MAAM,QAAQ;QACjC,mBAAmB,MAAM,QAAQ;QACjC,YAAY,MAAM,QAAQ;QAC1B,kBAAkB,MAAM,QAAQ;QAChC,IAAI,MAAM,QAAQ;OACnB;OACD,MAAM;OACN,gBAAgB,MAAM;MACvB;MACD,MAAM;KACP;KACD,eAAe,aAAa,UAAU,WAAW;IAClD,WAEK,cAAc;KAChB,MAAM,eAAe,IAAID,6CAAoB;MAC3C,SAAS,IAAIC,yCAAe;OAC1B,SAAS;OACT,mBAAmB,MAAM,QAAQ;OACjC,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;KACN,eAAe;IAChB;GAEJ;EACF;AAGD,MAAI,aAEF,KAAI,YAAY;GACd,MAAM,iBAAiB,IAAID,6CAAoB;IAC7C,SAAS,IAAIC,yCAAe;KAC1B,SAAS;KACT,mBAAmB,EAAE,mBAAmB,aAAc;IACvD;IACD,MAAM;GACP;GACD,MAAM;EACP,OAAM;GACL,MAAM,eAAe,IAAID,6CAAoB;IAC3C,SAAS,IAAIC,yCAAe,EAC1B,SAAS,aACV;IACD,MAAM;GACP;GACD,MAAM;EACP;CAEJ;CAED,AAAmB,wCACjBC,SACAC,aACA;EACA,MAAM,mBAAmB,MAAM,wCAC7B,SACA,YACD;EACD,iBAAiB,kBAAkB,oBAEhC,QAAgB;EAEnB,iBAAiB,oBAAoB;GACnC,GAAG,iBAAiB;GACpB,gBAAgB;EACjB;AACD,SAAO;CACR;;;;;;;;;;;;;;;;;;CAmBD,IAAI,UAAwB;AAC1B,SAAOC,yBAAS,KAAK,UAAU,CAAE;CAClC;CAqCD,qBAIEC,cAIAC,QAMI;EACJ,MAAM,gBAAgB,EAAE,GAAG,OAAQ;AAEnC,MAAI,eAAe,WAAW,QAC5B,cAAc,SAAS;AAEzB,SAAO,MAAM,qBAAgC,cAAc,cAAc;CAC1E;AACF"}
1
+ {"version":3,"file":"chat_models.cjs","names":["ChatOpenAICompletions","fields?: Partial<ChatDeepSeekInput>","delta: Record<string, any>","rawResponse: OpenAIClient.ChatCompletionChunk","defaultRole?:\n | \"function\"\n | \"user\"\n | \"system\"\n | \"developer\"\n | \"assistant\"\n | \"tool\"","messages: BaseMessage[]","options: this[\"ParsedCallOptions\"]","runManager?: CallbackManagerForLLMRun","ChatGenerationChunk","AIMessageChunk","message: OpenAIClient.ChatCompletionMessage","rawResponse: OpenAIClient.ChatCompletion","PROFILES","outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>","config?: StructuredOutputMethodOptions<boolean>"],"sources":["../src/chat_models.ts"],"sourcesContent":["import {\n BaseLanguageModelInput,\n StructuredOutputMethodOptions,\n} from \"@langchain/core/language_models/base\";\nimport { ModelProfile } from \"@langchain/core/language_models/profile\";\nimport { BaseMessage, AIMessageChunk } from \"@langchain/core/messages\";\nimport { Runnable } from \"@langchain/core/runnables\";\nimport { getEnvironmentVariable } from \"@langchain/core/utils/env\";\nimport { InteropZodType } from \"@langchain/core/utils/types\";\nimport {\n ChatOpenAICallOptions,\n ChatOpenAICompletions,\n ChatOpenAIFields,\n OpenAIClient,\n} from \"@langchain/openai\";\nimport { ChatGenerationChunk } from \"@langchain/core/outputs\";\nimport { CallbackManagerForLLMRun } from \"@langchain/core/callbacks/manager\";\nimport PROFILES from \"./profiles.js\";\n\nexport interface ChatDeepSeekCallOptions extends ChatOpenAICallOptions {\n headers?: Record<string, string>;\n}\n\nexport interface ChatDeepSeekInput extends ChatOpenAIFields {\n /**\n * The Deepseek API key to use for requests.\n * @default process.env.DEEPSEEK_API_KEY\n */\n apiKey?: string;\n /**\n * The name of the model to use.\n */\n model?: string;\n /**\n * Up to 4 sequences where the API will stop generating further tokens. The\n * returned text will not contain the stop sequence.\n * Alias for `stopSequences`\n */\n stop?: Array<string>;\n /**\n * Up to 4 sequences where the API will stop generating further tokens. The\n * returned text will not contain the stop sequence.\n */\n stopSequences?: Array<string>;\n /**\n * Whether or not to stream responses.\n */\n streaming?: boolean;\n /**\n * The temperature to use for sampling.\n */\n temperature?: number;\n /**\n * The maximum number of tokens that the model can process in a single response.\n * This limits ensures computational efficiency and resource management.\n */\n maxTokens?: number;\n}\n\n/**\n * Deepseek chat model integration.\n *\n * The Deepseek API is compatible to the OpenAI API with some limitations.\n *\n * Setup:\n * Install `@langchain/deepseek` and set an environment variable named `DEEPSEEK_API_KEY`.\n *\n * ```bash\n * npm install @langchain/deepseek\n * export DEEPSEEK_API_KEY=\"your-api-key\"\n * ```\n *\n * ## [Constructor args](https://api.js.langchain.com/classes/_langchain_deepseek.ChatDeepSeek.html#constructor)\n *\n * ## [Runtime args](https://api.js.langchain.com/interfaces/_langchain_deepseek.ChatDeepSeekCallOptions.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 { ChatDeepSeek } from '@langchain/deepseek';\n *\n * const llm = new ChatDeepSeek({\n * model: \"deepseek-reasoner\",\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 * AIMessage {\n * \"content\": \"The French translation of \\\"I love programming\\\" is \\\"J'aime programmer\\\". In this sentence, \\\"J'aime\\\" is the first person singular conjugation of the French verb \\\"aimer\\\" which means \\\"to love\\\", and \\\"programmer\\\" is the French infinitive for \\\"to program\\\". I hope this helps! Let me know if you have any other questions.\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"tokenUsage\": {\n * \"completionTokens\": 82,\n * \"promptTokens\": 20,\n * \"totalTokens\": 102\n * },\n * \"finish_reason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"invalid_tool_calls\": []\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 * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"The\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" French\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" translation\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" of\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" \\\"\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"I\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" love\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * ...\n * AIMessageChunk {\n * \"content\": \".\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\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\": \"The French translation of \\\"I love programming\\\" is \\\"J'aime programmer\\\". In this sentence, \\\"J'aime\\\" is the first person singular conjugation of the French verb \\\"aimer\\\" which means \\\"to love\\\", and \\\"programmer\\\" is the French infinitive for \\\"to program\\\". I hope this helps! Let me know if you have any other questions.\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\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 llmForToolCalling = new ChatDeepSeek({\n * model: \"deepseek-chat\",\n * temperature: 0,\n * // other params...\n * });\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 = llmForToolCalling.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: 'GetWeather',\n * args: { location: 'Los Angeles, CA' },\n * type: 'tool_call',\n * id: 'call_cd34'\n * },\n * {\n * name: 'GetWeather',\n * args: { location: 'New York, NY' },\n * type: 'tool_call',\n * id: 'call_68rf'\n * },\n * {\n * name: 'GetPopulation',\n * args: { location: 'Los Angeles, CA' },\n * type: 'tool_call',\n * id: 'call_f81z'\n * },\n * {\n * name: 'GetPopulation',\n * args: { location: 'New York, NY' },\n * type: 'tool_call',\n * id: 'call_8byt'\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 = llmForToolCalling.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: \"Why don't cats play poker in the wild?\",\n * punchline: 'Because there are too many cheetahs.'\n * }\n * ```\n * </details>\n *\n * <br />\n */\nexport class ChatDeepSeek extends ChatOpenAICompletions<ChatDeepSeekCallOptions> {\n static lc_name() {\n return \"ChatDeepSeek\";\n }\n\n _llmType() {\n return \"deepseek\";\n }\n\n get lc_secrets(): { [key: string]: string } | undefined {\n return {\n apiKey: \"DEEPSEEK_API_KEY\",\n };\n }\n\n lc_serializable = true;\n\n lc_namespace = [\"langchain\", \"chat_models\", \"deepseek\"];\n\n constructor(fields?: Partial<ChatDeepSeekInput>) {\n const apiKey = fields?.apiKey || getEnvironmentVariable(\"DEEPSEEK_API_KEY\");\n if (!apiKey) {\n throw new Error(\n `Deepseek API key not found. Please set the DEEPSEEK_API_KEY environment variable or pass the key into \"apiKey\" field.`\n );\n }\n\n super({\n ...fields,\n apiKey,\n configuration: {\n baseURL: \"https://api.deepseek.com\",\n ...fields?.configuration,\n },\n });\n }\n\n protected override _convertCompletionsDeltaToBaseMessageChunk(\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n delta: Record<string, any>,\n rawResponse: OpenAIClient.ChatCompletionChunk,\n defaultRole?:\n | \"function\"\n | \"user\"\n | \"system\"\n | \"developer\"\n | \"assistant\"\n | \"tool\"\n ) {\n const messageChunk = super._convertCompletionsDeltaToBaseMessageChunk(\n delta,\n rawResponse,\n defaultRole\n );\n messageChunk.additional_kwargs.reasoning_content = delta.reasoning_content;\n // Override model_provider for DeepSeek-specific block translation\n messageChunk.response_metadata = {\n ...messageChunk.response_metadata,\n model_provider: \"deepseek\",\n };\n return messageChunk;\n }\n\n async *_streamResponseChunks(\n messages: BaseMessage[],\n options: this[\"ParsedCallOptions\"],\n runManager?: CallbackManagerForLLMRun\n ): AsyncGenerator<ChatGenerationChunk> {\n const stream = super._streamResponseChunks(messages, options, runManager);\n\n // State for parsing <think> tags\n let tokensBuffer = \"\";\n let isThinking = false;\n\n for await (const chunk of stream) {\n if (options.signal?.aborted) {\n return;\n }\n // If the model already provided reasoning_content natively, just yield it\n if (chunk.message.additional_kwargs.reasoning_content) {\n yield chunk;\n continue;\n }\n\n const text = chunk.text;\n if (!text) {\n yield chunk;\n continue;\n }\n\n // Append text to buffer to handle split tags\n tokensBuffer += text;\n\n // Check for <think> start tag\n if (!isThinking && tokensBuffer.includes(\"<think>\")) {\n isThinking = true;\n const thinkIndex = tokensBuffer.indexOf(\"<think>\");\n const beforeThink = tokensBuffer.substring(0, thinkIndex);\n const afterThink = tokensBuffer.substring(\n thinkIndex + \"<think>\".length\n );\n\n // We consumed up to <think>, so buffer becomes what's after\n tokensBuffer = afterThink || \"\"; // might be empty or part of thought\n\n if (beforeThink) {\n // Send the content before the tag\n const newChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: beforeThink,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: beforeThink,\n generationInfo: chunk.generationInfo,\n });\n yield newChunk;\n }\n }\n\n // Check for </think> end tag\n if (isThinking && tokensBuffer.includes(\"</think>\")) {\n isThinking = false;\n const thinkEndIndex = tokensBuffer.indexOf(\"</think>\");\n const thoughtContent = tokensBuffer.substring(0, thinkEndIndex);\n const afterThink = tokensBuffer.substring(\n thinkEndIndex + \"</think>\".length\n );\n\n // Yield the reasoning content\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: {\n ...chunk.message.additional_kwargs,\n reasoning_content: thoughtContent,\n },\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: \"\",\n generationInfo: chunk.generationInfo,\n });\n yield reasoningChunk;\n\n // Reset buffer to what's after </think>\n tokensBuffer = afterThink || \"\";\n\n // Yield the rest as normal content if any\n if (tokensBuffer) {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: tokensBuffer,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: tokensBuffer,\n generationInfo: chunk.generationInfo,\n });\n yield contentChunk;\n tokensBuffer = \"\"; // consumed\n }\n } else if (isThinking) {\n // We are inside thinking block.\n // Check partial </think> match\n const possibleEndTag = \"</think>\";\n let splitIndex = -1;\n\n // Check if buffer ends with a prefix of </think> - Greedy check (longest first)\n for (let i = possibleEndTag.length - 1; i >= 1; i--) {\n if (tokensBuffer.endsWith(possibleEndTag.substring(0, i))) {\n splitIndex = tokensBuffer.length - i;\n break;\n }\n }\n\n if (splitIndex !== -1) {\n const safeToYield = tokensBuffer.substring(0, splitIndex);\n if (safeToYield) {\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: {\n ...chunk.message.additional_kwargs,\n reasoning_content: safeToYield,\n },\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: \"\",\n generationInfo: chunk.generationInfo,\n });\n yield reasoningChunk;\n }\n tokensBuffer = tokensBuffer.substring(splitIndex); // keep partial tag\n } else {\n // content is safe to yield as reasoning\n if (tokensBuffer) {\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: {\n ...chunk.message.additional_kwargs,\n reasoning_content: tokensBuffer,\n },\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: \"\",\n generationInfo: chunk.generationInfo,\n });\n yield reasoningChunk;\n tokensBuffer = \"\";\n }\n }\n } else {\n // NOT thinking.\n // Check partial start tag \"<think>\" - Greedy check (longest first)\n const possibleStartTag = \"<think>\";\n let splitIndex = -1;\n for (let i = possibleStartTag.length - 1; i >= 1; i--) {\n if (tokensBuffer.endsWith(possibleStartTag.substring(0, i))) {\n splitIndex = tokensBuffer.length - i;\n break;\n }\n }\n\n if (splitIndex !== -1) {\n // Yield safe content\n const safeToYield = tokensBuffer.substring(0, splitIndex);\n if (safeToYield) {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: safeToYield,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: safeToYield,\n generationInfo: chunk.generationInfo,\n });\n yield contentChunk;\n }\n tokensBuffer = tokensBuffer.substring(splitIndex); // keep partial tag\n } else {\n // Yield all\n if (tokensBuffer) {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: tokensBuffer,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: tokensBuffer,\n generationInfo: chunk.generationInfo,\n });\n yield contentChunk;\n tokensBuffer = \"\";\n }\n }\n }\n }\n\n // Flush remaining buffer at end of stream\n if (tokensBuffer) {\n // If we were thinking, it's unclosed thought.\n if (isThinking) {\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: { reasoning_content: tokensBuffer },\n }),\n text: \"\",\n });\n yield reasoningChunk;\n } else {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: tokensBuffer,\n }),\n text: tokensBuffer,\n });\n yield contentChunk;\n }\n }\n }\n\n protected override _convertCompletionsMessageToBaseMessage(\n message: OpenAIClient.ChatCompletionMessage,\n rawResponse: OpenAIClient.ChatCompletion\n ) {\n const langChainMessage = super._convertCompletionsMessageToBaseMessage(\n message,\n rawResponse\n );\n langChainMessage.additional_kwargs.reasoning_content =\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n (message as any).reasoning_content;\n // Override model_provider for DeepSeek-specific block translation\n langChainMessage.response_metadata = {\n ...langChainMessage.response_metadata,\n model_provider: \"deepseek\",\n };\n return langChainMessage;\n }\n\n /**\n * Return profiling information for the model.\n *\n * Provides information about the model's capabilities and constraints,\n * including token limits, multimodal support, and advanced features like\n * tool calling and structured output.\n *\n * @returns {ModelProfile} An object describing the model's capabilities and constraints\n *\n * @example\n * ```typescript\n * const model = new ChatDeepSeek({ model: \"deepseek-chat\" });\n * const profile = model.profile;\n * console.log(profile.maxInputTokens); // 128000\n * console.log(profile.imageInputs); // false\n * ```\n */\n get profile(): ModelProfile {\n return PROFILES[this.model] ?? {};\n }\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<false>\n ): Runnable<BaseLanguageModelInput, RunOutput>;\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<true>\n ): Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<boolean>\n ):\n | Runnable<BaseLanguageModelInput, RunOutput>\n | Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<boolean>\n ):\n | Runnable<BaseLanguageModelInput, RunOutput>\n | Runnable<\n BaseLanguageModelInput,\n { raw: BaseMessage; parsed: RunOutput }\n > {\n const ensuredConfig = { ...config };\n // Deepseek does not support json schema yet\n if (ensuredConfig?.method === undefined) {\n ensuredConfig.method = \"functionCalling\";\n }\n return super.withStructuredOutput<RunOutput>(outputSchema, ensuredConfig);\n }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAyZA,IAAa,eAAb,cAAkCA,yCAA+C;CAC/E,OAAO,UAAU;AACf,SAAO;CACR;CAED,WAAW;AACT,SAAO;CACR;CAED,IAAI,aAAoD;AACtD,SAAO,EACL,QAAQ,mBACT;CACF;CAED,kBAAkB;CAElB,eAAe;EAAC;EAAa;EAAe;CAAW;CAEvD,YAAYC,QAAqC;EAC/C,MAAM,SAAS,QAAQ,iEAAiC,mBAAmB;AAC3E,MAAI,CAAC,OACH,OAAM,IAAI,MACR,CAAC,qHAAqH,CAAC;EAI3H,MAAM;GACJ,GAAG;GACH;GACA,eAAe;IACb,SAAS;IACT,GAAG,QAAQ;GACZ;EACF,EAAC;CACH;CAED,AAAmB,2CAEjBC,OACAC,aACAC,aAOA;EACA,MAAM,eAAe,MAAM,2CACzB,OACA,aACA,YACD;EACD,aAAa,kBAAkB,oBAAoB,MAAM;EAEzD,aAAa,oBAAoB;GAC/B,GAAG,aAAa;GAChB,gBAAgB;EACjB;AACD,SAAO;CACR;CAED,OAAO,sBACLC,UACAC,SACAC,YACqC;EACrC,MAAM,SAAS,MAAM,sBAAsB,UAAU,SAAS,WAAW;EAGzE,IAAI,eAAe;EACnB,IAAI,aAAa;AAEjB,aAAW,MAAM,SAAS,QAAQ;AAChC,OAAI,QAAQ,QAAQ,QAClB;AAGF,OAAI,MAAM,QAAQ,kBAAkB,mBAAmB;IACrD,MAAM;AACN;GACD;GAED,MAAM,OAAO,MAAM;AACnB,OAAI,CAAC,MAAM;IACT,MAAM;AACN;GACD;GAGD,gBAAgB;AAGhB,OAAI,CAAC,cAAc,aAAa,SAAS,UAAU,EAAE;IACnD,aAAa;IACb,MAAM,aAAa,aAAa,QAAQ,UAAU;IAClD,MAAM,cAAc,aAAa,UAAU,GAAG,WAAW;IACzD,MAAM,aAAa,aAAa,UAC9B,aAAa,EACd;IAGD,eAAe,cAAc;AAE7B,QAAI,aAAa;KAEf,MAAM,WAAW,IAAIC,6CAAoB;MACvC,SAAS,IAAIC,yCAAe;OAC1B,SAAS;OACT,mBAAmB,MAAM,QAAQ;OACjC,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;IACP;GACF;AAGD,OAAI,cAAc,aAAa,SAAS,WAAW,EAAE;IACnD,aAAa;IACb,MAAM,gBAAgB,aAAa,QAAQ,WAAW;IACtD,MAAM,iBAAiB,aAAa,UAAU,GAAG,cAAc;IAC/D,MAAM,aAAa,aAAa,UAC9B,gBAAgB,EACjB;IAGD,MAAM,iBAAiB,IAAID,6CAAoB;KAC7C,SAAS,IAAIC,yCAAe;MAC1B,SAAS;MACT,mBAAmB;OACjB,GAAG,MAAM,QAAQ;OACjB,mBAAmB;MACpB;MACD,mBAAmB,MAAM,QAAQ;MACjC,YAAY,MAAM,QAAQ;MAC1B,kBAAkB,MAAM,QAAQ;MAChC,IAAI,MAAM,QAAQ;KACnB;KACD,MAAM;KACN,gBAAgB,MAAM;IACvB;IACD,MAAM;IAGN,eAAe,cAAc;AAG7B,QAAI,cAAc;KAChB,MAAM,eAAe,IAAID,6CAAoB;MAC3C,SAAS,IAAIC,yCAAe;OAC1B,SAAS;OACT,mBAAmB,MAAM,QAAQ;OACjC,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;KACN,eAAe;IAChB;GACF,WAAU,YAAY;IAGrB,MAAM,iBAAiB;IACvB,IAAI,aAAa;AAGjB,SAAK,IAAI,IAAI,GAA2B,KAAK,GAAG,IAC9C,KAAI,aAAa,SAAS,eAAe,UAAU,GAAG,EAAE,CAAC,EAAE;KACzD,aAAa,aAAa,SAAS;AACnC;IACD;AAGH,QAAI,eAAe,IAAI;KACrB,MAAM,cAAc,aAAa,UAAU,GAAG,WAAW;AACzD,SAAI,aAAa;MACf,MAAM,iBAAiB,IAAID,6CAAoB;OAC7C,SAAS,IAAIC,yCAAe;QAC1B,SAAS;QACT,mBAAmB;SACjB,GAAG,MAAM,QAAQ;SACjB,mBAAmB;QACpB;QACD,mBAAmB,MAAM,QAAQ;QACjC,YAAY,MAAM,QAAQ;QAC1B,kBAAkB,MAAM,QAAQ;QAChC,IAAI,MAAM,QAAQ;OACnB;OACD,MAAM;OACN,gBAAgB,MAAM;MACvB;MACD,MAAM;KACP;KACD,eAAe,aAAa,UAAU,WAAW;IAClD,WAEK,cAAc;KAChB,MAAM,iBAAiB,IAAID,6CAAoB;MAC7C,SAAS,IAAIC,yCAAe;OAC1B,SAAS;OACT,mBAAmB;QACjB,GAAG,MAAM,QAAQ;QACjB,mBAAmB;OACpB;OACD,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;KACN,eAAe;IAChB;GAEJ,OAAM;IAGL,MAAM,mBAAmB;IACzB,IAAI,aAAa;AACjB,SAAK,IAAI,IAAI,GAA6B,KAAK,GAAG,IAChD,KAAI,aAAa,SAAS,iBAAiB,UAAU,GAAG,EAAE,CAAC,EAAE;KAC3D,aAAa,aAAa,SAAS;AACnC;IACD;AAGH,QAAI,eAAe,IAAI;KAErB,MAAM,cAAc,aAAa,UAAU,GAAG,WAAW;AACzD,SAAI,aAAa;MACf,MAAM,eAAe,IAAID,6CAAoB;OAC3C,SAAS,IAAIC,yCAAe;QAC1B,SAAS;QACT,mBAAmB,MAAM,QAAQ;QACjC,mBAAmB,MAAM,QAAQ;QACjC,YAAY,MAAM,QAAQ;QAC1B,kBAAkB,MAAM,QAAQ;QAChC,IAAI,MAAM,QAAQ;OACnB;OACD,MAAM;OACN,gBAAgB,MAAM;MACvB;MACD,MAAM;KACP;KACD,eAAe,aAAa,UAAU,WAAW;IAClD,WAEK,cAAc;KAChB,MAAM,eAAe,IAAID,6CAAoB;MAC3C,SAAS,IAAIC,yCAAe;OAC1B,SAAS;OACT,mBAAmB,MAAM,QAAQ;OACjC,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;KACN,eAAe;IAChB;GAEJ;EACF;AAGD,MAAI,aAEF,KAAI,YAAY;GACd,MAAM,iBAAiB,IAAID,6CAAoB;IAC7C,SAAS,IAAIC,yCAAe;KAC1B,SAAS;KACT,mBAAmB,EAAE,mBAAmB,aAAc;IACvD;IACD,MAAM;GACP;GACD,MAAM;EACP,OAAM;GACL,MAAM,eAAe,IAAID,6CAAoB;IAC3C,SAAS,IAAIC,yCAAe,EAC1B,SAAS,aACV;IACD,MAAM;GACP;GACD,MAAM;EACP;CAEJ;CAED,AAAmB,wCACjBC,SACAC,aACA;EACA,MAAM,mBAAmB,MAAM,wCAC7B,SACA,YACD;EACD,iBAAiB,kBAAkB,oBAEhC,QAAgB;EAEnB,iBAAiB,oBAAoB;GACnC,GAAG,iBAAiB;GACpB,gBAAgB;EACjB;AACD,SAAO;CACR;;;;;;;;;;;;;;;;;;CAmBD,IAAI,UAAwB;AAC1B,SAAOC,yBAAS,KAAK,UAAU,CAAE;CAClC;CAqCD,qBAIEC,cAIAC,QAMI;EACJ,MAAM,gBAAgB,EAAE,GAAG,OAAQ;AAEnC,MAAI,eAAe,WAAW,QAC5B,cAAc,SAAS;AAEzB,SAAO,MAAM,qBAAgC,cAAc,cAAc;CAC1E;AACF"}
package/dist/chat_models.js CHANGED
@@ -397,6 +397,7 @@ var ChatDeepSeek = class extends ChatOpenAICompletions {
397
397
  let tokensBuffer = "";
398
398
  let isThinking = false;
399
399
  for await (const chunk of stream) {
400
+ if (options.signal?.aborted) return;
400
401
  if (chunk.message.additional_kwargs.reasoning_content) {
401
402
  yield chunk;
402
403
  continue;
package/dist/chat_models.js.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"file":"chat_models.js","names":["fields?: Partial<ChatDeepSeekInput>","delta: Record<string, any>","rawResponse: OpenAIClient.ChatCompletionChunk","defaultRole?:\n | \"function\"\n | \"user\"\n | \"system\"\n | \"developer\"\n | \"assistant\"\n | \"tool\"","messages: BaseMessage[]","options: this[\"ParsedCallOptions\"]","runManager?: CallbackManagerForLLMRun","message: OpenAIClient.ChatCompletionMessage","rawResponse: OpenAIClient.ChatCompletion","PROFILES","outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>","config?: StructuredOutputMethodOptions<boolean>"],"sources":["../src/chat_models.ts"],"sourcesContent":["import {\n BaseLanguageModelInput,\n StructuredOutputMethodOptions,\n} from \"@langchain/core/language_models/base\";\nimport { ModelProfile } from \"@langchain/core/language_models/profile\";\nimport { BaseMessage, AIMessageChunk } from \"@langchain/core/messages\";\nimport { Runnable } from \"@langchain/core/runnables\";\nimport { getEnvironmentVariable } from \"@langchain/core/utils/env\";\nimport { InteropZodType } from \"@langchain/core/utils/types\";\nimport {\n ChatOpenAICallOptions,\n ChatOpenAICompletions,\n ChatOpenAIFields,\n OpenAIClient,\n} from \"@langchain/openai\";\nimport { ChatGenerationChunk } from \"@langchain/core/outputs\";\nimport { CallbackManagerForLLMRun } from \"@langchain/core/callbacks/manager\";\nimport PROFILES from \"./profiles.js\";\n\nexport interface ChatDeepSeekCallOptions extends ChatOpenAICallOptions {\n headers?: Record<string, string>;\n}\n\nexport interface ChatDeepSeekInput extends ChatOpenAIFields {\n /**\n * The Deepseek API key to use for requests.\n * @default process.env.DEEPSEEK_API_KEY\n */\n apiKey?: string;\n /**\n * The name of the model to use.\n */\n model?: string;\n /**\n * Up to 4 sequences where the API will stop generating further tokens. The\n * returned text will not contain the stop sequence.\n * Alias for `stopSequences`\n */\n stop?: Array<string>;\n /**\n * Up to 4 sequences where the API will stop generating further tokens. The\n * returned text will not contain the stop sequence.\n */\n stopSequences?: Array<string>;\n /**\n * Whether or not to stream responses.\n */\n streaming?: boolean;\n /**\n * The temperature to use for sampling.\n */\n temperature?: number;\n /**\n * The maximum number of tokens that the model can process in a single response.\n * This limits ensures computational efficiency and resource management.\n */\n maxTokens?: number;\n}\n\n/**\n * Deepseek chat model integration.\n *\n * The Deepseek API is compatible to the OpenAI API with some limitations.\n *\n * Setup:\n * Install `@langchain/deepseek` and set an environment variable named `DEEPSEEK_API_KEY`.\n *\n * ```bash\n * npm install @langchain/deepseek\n * export DEEPSEEK_API_KEY=\"your-api-key\"\n * ```\n *\n * ## [Constructor args](https://api.js.langchain.com/classes/_langchain_deepseek.ChatDeepSeek.html#constructor)\n *\n * ## [Runtime args](https://api.js.langchain.com/interfaces/_langchain_deepseek.ChatDeepSeekCallOptions.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 { ChatDeepSeek } from '@langchain/deepseek';\n *\n * const llm = new ChatDeepSeek({\n * model: \"deepseek-reasoner\",\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 * AIMessage {\n * \"content\": \"The French translation of \\\"I love programming\\\" is \\\"J'aime programmer\\\". In this sentence, \\\"J'aime\\\" is the first person singular conjugation of the French verb \\\"aimer\\\" which means \\\"to love\\\", and \\\"programmer\\\" is the French infinitive for \\\"to program\\\". I hope this helps! Let me know if you have any other questions.\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"tokenUsage\": {\n * \"completionTokens\": 82,\n * \"promptTokens\": 20,\n * \"totalTokens\": 102\n * },\n * \"finish_reason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"invalid_tool_calls\": []\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 * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"The\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" French\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" translation\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" of\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" \\\"\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"I\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" love\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * ...\n * AIMessageChunk {\n * \"content\": \".\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\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\": \"The French translation of \\\"I love programming\\\" is \\\"J'aime programmer\\\". In this sentence, \\\"J'aime\\\" is the first person singular conjugation of the French verb \\\"aimer\\\" which means \\\"to love\\\", and \\\"programmer\\\" is the French infinitive for \\\"to program\\\". I hope this helps! Let me know if you have any other questions.\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\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 llmForToolCalling = new ChatDeepSeek({\n * model: \"deepseek-chat\",\n * temperature: 0,\n * // other params...\n * });\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 = llmForToolCalling.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: 'GetWeather',\n * args: { location: 'Los Angeles, CA' },\n * type: 'tool_call',\n * id: 'call_cd34'\n * },\n * {\n * name: 'GetWeather',\n * args: { location: 'New York, NY' },\n * type: 'tool_call',\n * id: 'call_68rf'\n * },\n * {\n * name: 'GetPopulation',\n * args: { location: 'Los Angeles, CA' },\n * type: 'tool_call',\n * id: 'call_f81z'\n * },\n * {\n * name: 'GetPopulation',\n * args: { location: 'New York, NY' },\n * type: 'tool_call',\n * id: 'call_8byt'\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 = llmForToolCalling.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: \"Why don't cats play poker in the wild?\",\n * punchline: 'Because there are too many cheetahs.'\n * }\n * ```\n * </details>\n *\n * <br />\n */\nexport class ChatDeepSeek extends ChatOpenAICompletions<ChatDeepSeekCallOptions> {\n static lc_name() {\n return \"ChatDeepSeek\";\n }\n\n _llmType() {\n return \"deepseek\";\n }\n\n get lc_secrets(): { [key: string]: string } | undefined {\n return {\n apiKey: \"DEEPSEEK_API_KEY\",\n };\n }\n\n lc_serializable = true;\n\n lc_namespace = [\"langchain\", \"chat_models\", \"deepseek\"];\n\n constructor(fields?: Partial<ChatDeepSeekInput>) {\n const apiKey = fields?.apiKey || getEnvironmentVariable(\"DEEPSEEK_API_KEY\");\n if (!apiKey) {\n throw new Error(\n `Deepseek API key not found. Please set the DEEPSEEK_API_KEY environment variable or pass the key into \"apiKey\" field.`\n );\n }\n\n super({\n ...fields,\n apiKey,\n configuration: {\n baseURL: \"https://api.deepseek.com\",\n ...fields?.configuration,\n },\n });\n }\n\n protected override _convertCompletionsDeltaToBaseMessageChunk(\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n delta: Record<string, any>,\n rawResponse: OpenAIClient.ChatCompletionChunk,\n defaultRole?:\n | \"function\"\n | \"user\"\n | \"system\"\n | \"developer\"\n | \"assistant\"\n | \"tool\"\n ) {\n const messageChunk = super._convertCompletionsDeltaToBaseMessageChunk(\n delta,\n rawResponse,\n defaultRole\n );\n messageChunk.additional_kwargs.reasoning_content = delta.reasoning_content;\n // Override model_provider for DeepSeek-specific block translation\n messageChunk.response_metadata = {\n ...messageChunk.response_metadata,\n model_provider: \"deepseek\",\n };\n return messageChunk;\n }\n\n async *_streamResponseChunks(\n messages: BaseMessage[],\n options: this[\"ParsedCallOptions\"],\n runManager?: CallbackManagerForLLMRun\n ): AsyncGenerator<ChatGenerationChunk> {\n const stream = super._streamResponseChunks(messages, options, runManager);\n\n // State for parsing <think> tags\n let tokensBuffer = \"\";\n let isThinking = false;\n\n for await (const chunk of stream) {\n // If the model already provided reasoning_content natively, just yield it\n if (chunk.message.additional_kwargs.reasoning_content) {\n yield chunk;\n continue;\n }\n\n const text = chunk.text;\n if (!text) {\n yield chunk;\n continue;\n }\n\n // Append text to buffer to handle split tags\n tokensBuffer += text;\n\n // Check for <think> start tag\n if (!isThinking && tokensBuffer.includes(\"<think>\")) {\n isThinking = true;\n const thinkIndex = tokensBuffer.indexOf(\"<think>\");\n const beforeThink = tokensBuffer.substring(0, thinkIndex);\n const afterThink = tokensBuffer.substring(\n thinkIndex + \"<think>\".length\n );\n\n // We consumed up to <think>, so buffer becomes what's after\n tokensBuffer = afterThink || \"\"; // might be empty or part of thought\n\n if (beforeThink) {\n // Send the content before the tag\n const newChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: beforeThink,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: beforeThink,\n generationInfo: chunk.generationInfo,\n });\n yield newChunk;\n }\n }\n\n // Check for </think> end tag\n if (isThinking && tokensBuffer.includes(\"</think>\")) {\n isThinking = false;\n const thinkEndIndex = tokensBuffer.indexOf(\"</think>\");\n const thoughtContent = tokensBuffer.substring(0, thinkEndIndex);\n const afterThink = tokensBuffer.substring(\n thinkEndIndex + \"</think>\".length\n );\n\n // Yield the reasoning content\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: {\n ...chunk.message.additional_kwargs,\n reasoning_content: thoughtContent,\n },\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: \"\",\n generationInfo: chunk.generationInfo,\n });\n yield reasoningChunk;\n\n // Reset buffer to what's after </think>\n tokensBuffer = afterThink || \"\";\n\n // Yield the rest as normal content if any\n if (tokensBuffer) {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: tokensBuffer,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: tokensBuffer,\n generationInfo: chunk.generationInfo,\n });\n yield contentChunk;\n tokensBuffer = \"\"; // consumed\n }\n } else if (isThinking) {\n // We are inside thinking block.\n // Check partial </think> match\n const possibleEndTag = \"</think>\";\n let splitIndex = -1;\n\n // Check if buffer ends with a prefix of </think> - Greedy check (longest first)\n for (let i = possibleEndTag.length - 1; i >= 1; i--) {\n if (tokensBuffer.endsWith(possibleEndTag.substring(0, i))) {\n splitIndex = tokensBuffer.length - i;\n break;\n }\n }\n\n if (splitIndex !== -1) {\n const safeToYield = tokensBuffer.substring(0, splitIndex);\n if (safeToYield) {\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: {\n ...chunk.message.additional_kwargs,\n reasoning_content: safeToYield,\n },\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: \"\",\n generationInfo: chunk.generationInfo,\n });\n yield reasoningChunk;\n }\n tokensBuffer = tokensBuffer.substring(splitIndex); // keep partial tag\n } else {\n // content is safe to yield as reasoning\n if (tokensBuffer) {\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: {\n ...chunk.message.additional_kwargs,\n reasoning_content: tokensBuffer,\n },\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: \"\",\n generationInfo: chunk.generationInfo,\n });\n yield reasoningChunk;\n tokensBuffer = \"\";\n }\n }\n } else {\n // NOT thinking.\n // Check partial start tag \"<think>\" - Greedy check (longest first)\n const possibleStartTag = \"<think>\";\n let splitIndex = -1;\n for (let i = possibleStartTag.length - 1; i >= 1; i--) {\n if (tokensBuffer.endsWith(possibleStartTag.substring(0, i))) {\n splitIndex = tokensBuffer.length - i;\n break;\n }\n }\n\n if (splitIndex !== -1) {\n // Yield safe content\n const safeToYield = tokensBuffer.substring(0, splitIndex);\n if (safeToYield) {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: safeToYield,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: safeToYield,\n generationInfo: chunk.generationInfo,\n });\n yield contentChunk;\n }\n tokensBuffer = tokensBuffer.substring(splitIndex); // keep partial tag\n } else {\n // Yield all\n if (tokensBuffer) {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: tokensBuffer,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: tokensBuffer,\n generationInfo: chunk.generationInfo,\n });\n yield contentChunk;\n tokensBuffer = \"\";\n }\n }\n }\n }\n\n // Flush remaining buffer at end of stream\n if (tokensBuffer) {\n // If we were thinking, it's unclosed thought.\n if (isThinking) {\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: { reasoning_content: tokensBuffer },\n }),\n text: \"\",\n });\n yield reasoningChunk;\n } else {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: tokensBuffer,\n }),\n text: tokensBuffer,\n });\n yield contentChunk;\n }\n }\n }\n\n protected override _convertCompletionsMessageToBaseMessage(\n message: OpenAIClient.ChatCompletionMessage,\n rawResponse: OpenAIClient.ChatCompletion\n ) {\n const langChainMessage = super._convertCompletionsMessageToBaseMessage(\n message,\n rawResponse\n );\n langChainMessage.additional_kwargs.reasoning_content =\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n (message as any).reasoning_content;\n // Override model_provider for DeepSeek-specific block translation\n langChainMessage.response_metadata = {\n ...langChainMessage.response_metadata,\n model_provider: \"deepseek\",\n };\n return langChainMessage;\n }\n\n /**\n * Return profiling information for the model.\n *\n * Provides information about the model's capabilities and constraints,\n * including token limits, multimodal support, and advanced features like\n * tool calling and structured output.\n *\n * @returns {ModelProfile} An object describing the model's capabilities and constraints\n *\n * @example\n * ```typescript\n * const model = new ChatDeepSeek({ model: \"deepseek-chat\" });\n * const profile = model.profile;\n * console.log(profile.maxInputTokens); // 128000\n * console.log(profile.imageInputs); // false\n * ```\n */\n get profile(): ModelProfile {\n return PROFILES[this.model] ?? {};\n }\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<false>\n ): Runnable<BaseLanguageModelInput, RunOutput>;\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<true>\n ): Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<boolean>\n ):\n | Runnable<BaseLanguageModelInput, RunOutput>\n | Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<boolean>\n ):\n | Runnable<BaseLanguageModelInput, RunOutput>\n | Runnable<\n BaseLanguageModelInput,\n { raw: BaseMessage; parsed: RunOutput }\n > {\n const ensuredConfig = { ...config };\n // Deepseek does not support json schema yet\n if (ensuredConfig?.method === undefined) {\n ensuredConfig.method = \"functionCalling\";\n }\n return super.withStructuredOutput<RunOutput>(outputSchema, ensuredConfig);\n }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAyZA,IAAa,eAAb,cAAkC,sBAA+C;CAC/E,OAAO,UAAU;AACf,SAAO;CACR;CAED,WAAW;AACT,SAAO;CACR;CAED,IAAI,aAAoD;AACtD,SAAO,EACL,QAAQ,mBACT;CACF;CAED,kBAAkB;CAElB,eAAe;EAAC;EAAa;EAAe;CAAW;CAEvD,YAAYA,QAAqC;EAC/C,MAAM,SAAS,QAAQ,UAAU,uBAAuB,mBAAmB;AAC3E,MAAI,CAAC,OACH,OAAM,IAAI,MACR,CAAC,qHAAqH,CAAC;EAI3H,MAAM;GACJ,GAAG;GACH;GACA,eAAe;IACb,SAAS;IACT,GAAG,QAAQ;GACZ;EACF,EAAC;CACH;CAED,AAAmB,2CAEjBC,OACAC,aACAC,aAOA;EACA,MAAM,eAAe,MAAM,2CACzB,OACA,aACA,YACD;EACD,aAAa,kBAAkB,oBAAoB,MAAM;EAEzD,aAAa,oBAAoB;GAC/B,GAAG,aAAa;GAChB,gBAAgB;EACjB;AACD,SAAO;CACR;CAED,OAAO,sBACLC,UACAC,SACAC,YACqC;EACrC,MAAM,SAAS,MAAM,sBAAsB,UAAU,SAAS,WAAW;EAGzE,IAAI,eAAe;EACnB,IAAI,aAAa;AAEjB,aAAW,MAAM,SAAS,QAAQ;AAEhC,OAAI,MAAM,QAAQ,kBAAkB,mBAAmB;IACrD,MAAM;AACN;GACD;GAED,MAAM,OAAO,MAAM;AACnB,OAAI,CAAC,MAAM;IACT,MAAM;AACN;GACD;GAGD,gBAAgB;AAGhB,OAAI,CAAC,cAAc,aAAa,SAAS,UAAU,EAAE;IACnD,aAAa;IACb,MAAM,aAAa,aAAa,QAAQ,UAAU;IAClD,MAAM,cAAc,aAAa,UAAU,GAAG,WAAW;IACzD,MAAM,aAAa,aAAa,UAC9B,aAAa,EACd;IAGD,eAAe,cAAc;AAE7B,QAAI,aAAa;KAEf,MAAM,WAAW,IAAI,oBAAoB;MACvC,SAAS,IAAI,eAAe;OAC1B,SAAS;OACT,mBAAmB,MAAM,QAAQ;OACjC,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;IACP;GACF;AAGD,OAAI,cAAc,aAAa,SAAS,WAAW,EAAE;IACnD,aAAa;IACb,MAAM,gBAAgB,aAAa,QAAQ,WAAW;IACtD,MAAM,iBAAiB,aAAa,UAAU,GAAG,cAAc;IAC/D,MAAM,aAAa,aAAa,UAC9B,gBAAgB,EACjB;IAGD,MAAM,iBAAiB,IAAI,oBAAoB;KAC7C,SAAS,IAAI,eAAe;MAC1B,SAAS;MACT,mBAAmB;OACjB,GAAG,MAAM,QAAQ;OACjB,mBAAmB;MACpB;MACD,mBAAmB,MAAM,QAAQ;MACjC,YAAY,MAAM,QAAQ;MAC1B,kBAAkB,MAAM,QAAQ;MAChC,IAAI,MAAM,QAAQ;KACnB;KACD,MAAM;KACN,gBAAgB,MAAM;IACvB;IACD,MAAM;IAGN,eAAe,cAAc;AAG7B,QAAI,cAAc;KAChB,MAAM,eAAe,IAAI,oBAAoB;MAC3C,SAAS,IAAI,eAAe;OAC1B,SAAS;OACT,mBAAmB,MAAM,QAAQ;OACjC,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;KACN,eAAe;IAChB;GACF,WAAU,YAAY;IAGrB,MAAM,iBAAiB;IACvB,IAAI,aAAa;AAGjB,SAAK,IAAI,IAAI,GAA2B,KAAK,GAAG,IAC9C,KAAI,aAAa,SAAS,eAAe,UAAU,GAAG,EAAE,CAAC,EAAE;KACzD,aAAa,aAAa,SAAS;AACnC;IACD;AAGH,QAAI,eAAe,IAAI;KACrB,MAAM,cAAc,aAAa,UAAU,GAAG,WAAW;AACzD,SAAI,aAAa;MACf,MAAM,iBAAiB,IAAI,oBAAoB;OAC7C,SAAS,IAAI,eAAe;QAC1B,SAAS;QACT,mBAAmB;SACjB,GAAG,MAAM,QAAQ;SACjB,mBAAmB;QACpB;QACD,mBAAmB,MAAM,QAAQ;QACjC,YAAY,MAAM,QAAQ;QAC1B,kBAAkB,MAAM,QAAQ;QAChC,IAAI,MAAM,QAAQ;OACnB;OACD,MAAM;OACN,gBAAgB,MAAM;MACvB;MACD,MAAM;KACP;KACD,eAAe,aAAa,UAAU,WAAW;IAClD,WAEK,cAAc;KAChB,MAAM,iBAAiB,IAAI,oBAAoB;MAC7C,SAAS,IAAI,eAAe;OAC1B,SAAS;OACT,mBAAmB;QACjB,GAAG,MAAM,QAAQ;QACjB,mBAAmB;OACpB;OACD,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;KACN,eAAe;IAChB;GAEJ,OAAM;IAGL,MAAM,mBAAmB;IACzB,IAAI,aAAa;AACjB,SAAK,IAAI,IAAI,GAA6B,KAAK,GAAG,IAChD,KAAI,aAAa,SAAS,iBAAiB,UAAU,GAAG,EAAE,CAAC,EAAE;KAC3D,aAAa,aAAa,SAAS;AACnC;IACD;AAGH,QAAI,eAAe,IAAI;KAErB,MAAM,cAAc,aAAa,UAAU,GAAG,WAAW;AACzD,SAAI,aAAa;MACf,MAAM,eAAe,IAAI,oBAAoB;OAC3C,SAAS,IAAI,eAAe;QAC1B,SAAS;QACT,mBAAmB,MAAM,QAAQ;QACjC,mBAAmB,MAAM,QAAQ;QACjC,YAAY,MAAM,QAAQ;QAC1B,kBAAkB,MAAM,QAAQ;QAChC,IAAI,MAAM,QAAQ;OACnB;OACD,MAAM;OACN,gBAAgB,MAAM;MACvB;MACD,MAAM;KACP;KACD,eAAe,aAAa,UAAU,WAAW;IAClD,WAEK,cAAc;KAChB,MAAM,eAAe,IAAI,oBAAoB;MAC3C,SAAS,IAAI,eAAe;OAC1B,SAAS;OACT,mBAAmB,MAAM,QAAQ;OACjC,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;KACN,eAAe;IAChB;GAEJ;EACF;AAGD,MAAI,aAEF,KAAI,YAAY;GACd,MAAM,iBAAiB,IAAI,oBAAoB;IAC7C,SAAS,IAAI,eAAe;KAC1B,SAAS;KACT,mBAAmB,EAAE,mBAAmB,aAAc;IACvD;IACD,MAAM;GACP;GACD,MAAM;EACP,OAAM;GACL,MAAM,eAAe,IAAI,oBAAoB;IAC3C,SAAS,IAAI,eAAe,EAC1B,SAAS,aACV;IACD,MAAM;GACP;GACD,MAAM;EACP;CAEJ;CAED,AAAmB,wCACjBC,SACAC,aACA;EACA,MAAM,mBAAmB,MAAM,wCAC7B,SACA,YACD;EACD,iBAAiB,kBAAkB,oBAEhC,QAAgB;EAEnB,iBAAiB,oBAAoB;GACnC,GAAG,iBAAiB;GACpB,gBAAgB;EACjB;AACD,SAAO;CACR;;;;;;;;;;;;;;;;;;CAmBD,IAAI,UAAwB;AAC1B,SAAOC,iBAAS,KAAK,UAAU,CAAE;CAClC;CAqCD,qBAIEC,cAIAC,QAMI;EACJ,MAAM,gBAAgB,EAAE,GAAG,OAAQ;AAEnC,MAAI,eAAe,WAAW,QAC5B,cAAc,SAAS;AAEzB,SAAO,MAAM,qBAAgC,cAAc,cAAc;CAC1E;AACF"}
1
+ {"version":3,"file":"chat_models.js","names":["fields?: Partial<ChatDeepSeekInput>","delta: Record<string, any>","rawResponse: OpenAIClient.ChatCompletionChunk","defaultRole?:\n | \"function\"\n | \"user\"\n | \"system\"\n | \"developer\"\n | \"assistant\"\n | \"tool\"","messages: BaseMessage[]","options: this[\"ParsedCallOptions\"]","runManager?: CallbackManagerForLLMRun","message: OpenAIClient.ChatCompletionMessage","rawResponse: OpenAIClient.ChatCompletion","PROFILES","outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>","config?: StructuredOutputMethodOptions<boolean>"],"sources":["../src/chat_models.ts"],"sourcesContent":["import {\n BaseLanguageModelInput,\n StructuredOutputMethodOptions,\n} from \"@langchain/core/language_models/base\";\nimport { ModelProfile } from \"@langchain/core/language_models/profile\";\nimport { BaseMessage, AIMessageChunk } from \"@langchain/core/messages\";\nimport { Runnable } from \"@langchain/core/runnables\";\nimport { getEnvironmentVariable } from \"@langchain/core/utils/env\";\nimport { InteropZodType } from \"@langchain/core/utils/types\";\nimport {\n ChatOpenAICallOptions,\n ChatOpenAICompletions,\n ChatOpenAIFields,\n OpenAIClient,\n} from \"@langchain/openai\";\nimport { ChatGenerationChunk } from \"@langchain/core/outputs\";\nimport { CallbackManagerForLLMRun } from \"@langchain/core/callbacks/manager\";\nimport PROFILES from \"./profiles.js\";\n\nexport interface ChatDeepSeekCallOptions extends ChatOpenAICallOptions {\n headers?: Record<string, string>;\n}\n\nexport interface ChatDeepSeekInput extends ChatOpenAIFields {\n /**\n * The Deepseek API key to use for requests.\n * @default process.env.DEEPSEEK_API_KEY\n */\n apiKey?: string;\n /**\n * The name of the model to use.\n */\n model?: string;\n /**\n * Up to 4 sequences where the API will stop generating further tokens. The\n * returned text will not contain the stop sequence.\n * Alias for `stopSequences`\n */\n stop?: Array<string>;\n /**\n * Up to 4 sequences where the API will stop generating further tokens. The\n * returned text will not contain the stop sequence.\n */\n stopSequences?: Array<string>;\n /**\n * Whether or not to stream responses.\n */\n streaming?: boolean;\n /**\n * The temperature to use for sampling.\n */\n temperature?: number;\n /**\n * The maximum number of tokens that the model can process in a single response.\n * This limits ensures computational efficiency and resource management.\n */\n maxTokens?: number;\n}\n\n/**\n * Deepseek chat model integration.\n *\n * The Deepseek API is compatible to the OpenAI API with some limitations.\n *\n * Setup:\n * Install `@langchain/deepseek` and set an environment variable named `DEEPSEEK_API_KEY`.\n *\n * ```bash\n * npm install @langchain/deepseek\n * export DEEPSEEK_API_KEY=\"your-api-key\"\n * ```\n *\n * ## [Constructor args](https://api.js.langchain.com/classes/_langchain_deepseek.ChatDeepSeek.html#constructor)\n *\n * ## [Runtime args](https://api.js.langchain.com/interfaces/_langchain_deepseek.ChatDeepSeekCallOptions.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 { ChatDeepSeek } from '@langchain/deepseek';\n *\n * const llm = new ChatDeepSeek({\n * model: \"deepseek-reasoner\",\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 * AIMessage {\n * \"content\": \"The French translation of \\\"I love programming\\\" is \\\"J'aime programmer\\\". In this sentence, \\\"J'aime\\\" is the first person singular conjugation of the French verb \\\"aimer\\\" which means \\\"to love\\\", and \\\"programmer\\\" is the French infinitive for \\\"to program\\\". I hope this helps! Let me know if you have any other questions.\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"tokenUsage\": {\n * \"completionTokens\": 82,\n * \"promptTokens\": 20,\n * \"totalTokens\": 102\n * },\n * \"finish_reason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"invalid_tool_calls\": []\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 * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"The\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" French\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" translation\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" of\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" \\\"\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"I\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \" love\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * ...\n * AIMessageChunk {\n * \"content\": \".\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": null\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\n * }\n * AIMessageChunk {\n * \"content\": \"\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\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\": \"The French translation of \\\"I love programming\\\" is \\\"J'aime programmer\\\". In this sentence, \\\"J'aime\\\" is the first person singular conjugation of the French verb \\\"aimer\\\" which means \\\"to love\\\", and \\\"programmer\\\" is the French infinitive for \\\"to program\\\". I hope this helps! Let me know if you have any other questions.\",\n * \"additional_kwargs\": {\n * \"reasoning_content\": \"...\",\n * },\n * \"response_metadata\": {\n * \"finishReason\": \"stop\"\n * },\n * \"tool_calls\": [],\n * \"tool_call_chunks\": [],\n * \"invalid_tool_calls\": []\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 llmForToolCalling = new ChatDeepSeek({\n * model: \"deepseek-chat\",\n * temperature: 0,\n * // other params...\n * });\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 = llmForToolCalling.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: 'GetWeather',\n * args: { location: 'Los Angeles, CA' },\n * type: 'tool_call',\n * id: 'call_cd34'\n * },\n * {\n * name: 'GetWeather',\n * args: { location: 'New York, NY' },\n * type: 'tool_call',\n * id: 'call_68rf'\n * },\n * {\n * name: 'GetPopulation',\n * args: { location: 'Los Angeles, CA' },\n * type: 'tool_call',\n * id: 'call_f81z'\n * },\n * {\n * name: 'GetPopulation',\n * args: { location: 'New York, NY' },\n * type: 'tool_call',\n * id: 'call_8byt'\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 = llmForToolCalling.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: \"Why don't cats play poker in the wild?\",\n * punchline: 'Because there are too many cheetahs.'\n * }\n * ```\n * </details>\n *\n * <br />\n */\nexport class ChatDeepSeek extends ChatOpenAICompletions<ChatDeepSeekCallOptions> {\n static lc_name() {\n return \"ChatDeepSeek\";\n }\n\n _llmType() {\n return \"deepseek\";\n }\n\n get lc_secrets(): { [key: string]: string } | undefined {\n return {\n apiKey: \"DEEPSEEK_API_KEY\",\n };\n }\n\n lc_serializable = true;\n\n lc_namespace = [\"langchain\", \"chat_models\", \"deepseek\"];\n\n constructor(fields?: Partial<ChatDeepSeekInput>) {\n const apiKey = fields?.apiKey || getEnvironmentVariable(\"DEEPSEEK_API_KEY\");\n if (!apiKey) {\n throw new Error(\n `Deepseek API key not found. Please set the DEEPSEEK_API_KEY environment variable or pass the key into \"apiKey\" field.`\n );\n }\n\n super({\n ...fields,\n apiKey,\n configuration: {\n baseURL: \"https://api.deepseek.com\",\n ...fields?.configuration,\n },\n });\n }\n\n protected override _convertCompletionsDeltaToBaseMessageChunk(\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n delta: Record<string, any>,\n rawResponse: OpenAIClient.ChatCompletionChunk,\n defaultRole?:\n | \"function\"\n | \"user\"\n | \"system\"\n | \"developer\"\n | \"assistant\"\n | \"tool\"\n ) {\n const messageChunk = super._convertCompletionsDeltaToBaseMessageChunk(\n delta,\n rawResponse,\n defaultRole\n );\n messageChunk.additional_kwargs.reasoning_content = delta.reasoning_content;\n // Override model_provider for DeepSeek-specific block translation\n messageChunk.response_metadata = {\n ...messageChunk.response_metadata,\n model_provider: \"deepseek\",\n };\n return messageChunk;\n }\n\n async *_streamResponseChunks(\n messages: BaseMessage[],\n options: this[\"ParsedCallOptions\"],\n runManager?: CallbackManagerForLLMRun\n ): AsyncGenerator<ChatGenerationChunk> {\n const stream = super._streamResponseChunks(messages, options, runManager);\n\n // State for parsing <think> tags\n let tokensBuffer = \"\";\n let isThinking = false;\n\n for await (const chunk of stream) {\n if (options.signal?.aborted) {\n return;\n }\n // If the model already provided reasoning_content natively, just yield it\n if (chunk.message.additional_kwargs.reasoning_content) {\n yield chunk;\n continue;\n }\n\n const text = chunk.text;\n if (!text) {\n yield chunk;\n continue;\n }\n\n // Append text to buffer to handle split tags\n tokensBuffer += text;\n\n // Check for <think> start tag\n if (!isThinking && tokensBuffer.includes(\"<think>\")) {\n isThinking = true;\n const thinkIndex = tokensBuffer.indexOf(\"<think>\");\n const beforeThink = tokensBuffer.substring(0, thinkIndex);\n const afterThink = tokensBuffer.substring(\n thinkIndex + \"<think>\".length\n );\n\n // We consumed up to <think>, so buffer becomes what's after\n tokensBuffer = afterThink || \"\"; // might be empty or part of thought\n\n if (beforeThink) {\n // Send the content before the tag\n const newChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: beforeThink,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: beforeThink,\n generationInfo: chunk.generationInfo,\n });\n yield newChunk;\n }\n }\n\n // Check for </think> end tag\n if (isThinking && tokensBuffer.includes(\"</think>\")) {\n isThinking = false;\n const thinkEndIndex = tokensBuffer.indexOf(\"</think>\");\n const thoughtContent = tokensBuffer.substring(0, thinkEndIndex);\n const afterThink = tokensBuffer.substring(\n thinkEndIndex + \"</think>\".length\n );\n\n // Yield the reasoning content\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: {\n ...chunk.message.additional_kwargs,\n reasoning_content: thoughtContent,\n },\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: \"\",\n generationInfo: chunk.generationInfo,\n });\n yield reasoningChunk;\n\n // Reset buffer to what's after </think>\n tokensBuffer = afterThink || \"\";\n\n // Yield the rest as normal content if any\n if (tokensBuffer) {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: tokensBuffer,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: tokensBuffer,\n generationInfo: chunk.generationInfo,\n });\n yield contentChunk;\n tokensBuffer = \"\"; // consumed\n }\n } else if (isThinking) {\n // We are inside thinking block.\n // Check partial </think> match\n const possibleEndTag = \"</think>\";\n let splitIndex = -1;\n\n // Check if buffer ends with a prefix of </think> - Greedy check (longest first)\n for (let i = possibleEndTag.length - 1; i >= 1; i--) {\n if (tokensBuffer.endsWith(possibleEndTag.substring(0, i))) {\n splitIndex = tokensBuffer.length - i;\n break;\n }\n }\n\n if (splitIndex !== -1) {\n const safeToYield = tokensBuffer.substring(0, splitIndex);\n if (safeToYield) {\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: {\n ...chunk.message.additional_kwargs,\n reasoning_content: safeToYield,\n },\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: \"\",\n generationInfo: chunk.generationInfo,\n });\n yield reasoningChunk;\n }\n tokensBuffer = tokensBuffer.substring(splitIndex); // keep partial tag\n } else {\n // content is safe to yield as reasoning\n if (tokensBuffer) {\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: {\n ...chunk.message.additional_kwargs,\n reasoning_content: tokensBuffer,\n },\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: \"\",\n generationInfo: chunk.generationInfo,\n });\n yield reasoningChunk;\n tokensBuffer = \"\";\n }\n }\n } else {\n // NOT thinking.\n // Check partial start tag \"<think>\" - Greedy check (longest first)\n const possibleStartTag = \"<think>\";\n let splitIndex = -1;\n for (let i = possibleStartTag.length - 1; i >= 1; i--) {\n if (tokensBuffer.endsWith(possibleStartTag.substring(0, i))) {\n splitIndex = tokensBuffer.length - i;\n break;\n }\n }\n\n if (splitIndex !== -1) {\n // Yield safe content\n const safeToYield = tokensBuffer.substring(0, splitIndex);\n if (safeToYield) {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: safeToYield,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: safeToYield,\n generationInfo: chunk.generationInfo,\n });\n yield contentChunk;\n }\n tokensBuffer = tokensBuffer.substring(splitIndex); // keep partial tag\n } else {\n // Yield all\n if (tokensBuffer) {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: tokensBuffer,\n additional_kwargs: chunk.message.additional_kwargs,\n response_metadata: chunk.message.response_metadata,\n tool_calls: chunk.message.tool_calls,\n tool_call_chunks: chunk.message.tool_call_chunks,\n id: chunk.message.id,\n }),\n text: tokensBuffer,\n generationInfo: chunk.generationInfo,\n });\n yield contentChunk;\n tokensBuffer = \"\";\n }\n }\n }\n }\n\n // Flush remaining buffer at end of stream\n if (tokensBuffer) {\n // If we were thinking, it's unclosed thought.\n if (isThinking) {\n const reasoningChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: \"\",\n additional_kwargs: { reasoning_content: tokensBuffer },\n }),\n text: \"\",\n });\n yield reasoningChunk;\n } else {\n const contentChunk = new ChatGenerationChunk({\n message: new AIMessageChunk({\n content: tokensBuffer,\n }),\n text: tokensBuffer,\n });\n yield contentChunk;\n }\n }\n }\n\n protected override _convertCompletionsMessageToBaseMessage(\n message: OpenAIClient.ChatCompletionMessage,\n rawResponse: OpenAIClient.ChatCompletion\n ) {\n const langChainMessage = super._convertCompletionsMessageToBaseMessage(\n message,\n rawResponse\n );\n langChainMessage.additional_kwargs.reasoning_content =\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n (message as any).reasoning_content;\n // Override model_provider for DeepSeek-specific block translation\n langChainMessage.response_metadata = {\n ...langChainMessage.response_metadata,\n model_provider: \"deepseek\",\n };\n return langChainMessage;\n }\n\n /**\n * Return profiling information for the model.\n *\n * Provides information about the model's capabilities and constraints,\n * including token limits, multimodal support, and advanced features like\n * tool calling and structured output.\n *\n * @returns {ModelProfile} An object describing the model's capabilities and constraints\n *\n * @example\n * ```typescript\n * const model = new ChatDeepSeek({ model: \"deepseek-chat\" });\n * const profile = model.profile;\n * console.log(profile.maxInputTokens); // 128000\n * console.log(profile.imageInputs); // false\n * ```\n */\n get profile(): ModelProfile {\n return PROFILES[this.model] ?? {};\n }\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<false>\n ): Runnable<BaseLanguageModelInput, RunOutput>;\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<true>\n ): Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<boolean>\n ):\n | Runnable<BaseLanguageModelInput, RunOutput>\n | Runnable<BaseLanguageModelInput, { raw: BaseMessage; parsed: RunOutput }>;\n\n withStructuredOutput<\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n RunOutput extends Record<string, any> = Record<string, any>,\n >(\n outputSchema:\n | InteropZodType<RunOutput>\n // eslint-disable-next-line @typescript-eslint/no-explicit-any\n | Record<string, any>,\n config?: StructuredOutputMethodOptions<boolean>\n ):\n | Runnable<BaseLanguageModelInput, RunOutput>\n | Runnable<\n BaseLanguageModelInput,\n { raw: BaseMessage; parsed: RunOutput }\n > {\n const ensuredConfig = { ...config };\n // Deepseek does not support json schema yet\n if (ensuredConfig?.method === undefined) {\n ensuredConfig.method = \"functionCalling\";\n }\n return super.withStructuredOutput<RunOutput>(outputSchema, ensuredConfig);\n }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAyZA,IAAa,eAAb,cAAkC,sBAA+C;CAC/E,OAAO,UAAU;AACf,SAAO;CACR;CAED,WAAW;AACT,SAAO;CACR;CAED,IAAI,aAAoD;AACtD,SAAO,EACL,QAAQ,mBACT;CACF;CAED,kBAAkB;CAElB,eAAe;EAAC;EAAa;EAAe;CAAW;CAEvD,YAAYA,QAAqC;EAC/C,MAAM,SAAS,QAAQ,UAAU,uBAAuB,mBAAmB;AAC3E,MAAI,CAAC,OACH,OAAM,IAAI,MACR,CAAC,qHAAqH,CAAC;EAI3H,MAAM;GACJ,GAAG;GACH;GACA,eAAe;IACb,SAAS;IACT,GAAG,QAAQ;GACZ;EACF,EAAC;CACH;CAED,AAAmB,2CAEjBC,OACAC,aACAC,aAOA;EACA,MAAM,eAAe,MAAM,2CACzB,OACA,aACA,YACD;EACD,aAAa,kBAAkB,oBAAoB,MAAM;EAEzD,aAAa,oBAAoB;GAC/B,GAAG,aAAa;GAChB,gBAAgB;EACjB;AACD,SAAO;CACR;CAED,OAAO,sBACLC,UACAC,SACAC,YACqC;EACrC,MAAM,SAAS,MAAM,sBAAsB,UAAU,SAAS,WAAW;EAGzE,IAAI,eAAe;EACnB,IAAI,aAAa;AAEjB,aAAW,MAAM,SAAS,QAAQ;AAChC,OAAI,QAAQ,QAAQ,QAClB;AAGF,OAAI,MAAM,QAAQ,kBAAkB,mBAAmB;IACrD,MAAM;AACN;GACD;GAED,MAAM,OAAO,MAAM;AACnB,OAAI,CAAC,MAAM;IACT,MAAM;AACN;GACD;GAGD,gBAAgB;AAGhB,OAAI,CAAC,cAAc,aAAa,SAAS,UAAU,EAAE;IACnD,aAAa;IACb,MAAM,aAAa,aAAa,QAAQ,UAAU;IAClD,MAAM,cAAc,aAAa,UAAU,GAAG,WAAW;IACzD,MAAM,aAAa,aAAa,UAC9B,aAAa,EACd;IAGD,eAAe,cAAc;AAE7B,QAAI,aAAa;KAEf,MAAM,WAAW,IAAI,oBAAoB;MACvC,SAAS,IAAI,eAAe;OAC1B,SAAS;OACT,mBAAmB,MAAM,QAAQ;OACjC,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;IACP;GACF;AAGD,OAAI,cAAc,aAAa,SAAS,WAAW,EAAE;IACnD,aAAa;IACb,MAAM,gBAAgB,aAAa,QAAQ,WAAW;IACtD,MAAM,iBAAiB,aAAa,UAAU,GAAG,cAAc;IAC/D,MAAM,aAAa,aAAa,UAC9B,gBAAgB,EACjB;IAGD,MAAM,iBAAiB,IAAI,oBAAoB;KAC7C,SAAS,IAAI,eAAe;MAC1B,SAAS;MACT,mBAAmB;OACjB,GAAG,MAAM,QAAQ;OACjB,mBAAmB;MACpB;MACD,mBAAmB,MAAM,QAAQ;MACjC,YAAY,MAAM,QAAQ;MAC1B,kBAAkB,MAAM,QAAQ;MAChC,IAAI,MAAM,QAAQ;KACnB;KACD,MAAM;KACN,gBAAgB,MAAM;IACvB;IACD,MAAM;IAGN,eAAe,cAAc;AAG7B,QAAI,cAAc;KAChB,MAAM,eAAe,IAAI,oBAAoB;MAC3C,SAAS,IAAI,eAAe;OAC1B,SAAS;OACT,mBAAmB,MAAM,QAAQ;OACjC,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;KACN,eAAe;IAChB;GACF,WAAU,YAAY;IAGrB,MAAM,iBAAiB;IACvB,IAAI,aAAa;AAGjB,SAAK,IAAI,IAAI,GAA2B,KAAK,GAAG,IAC9C,KAAI,aAAa,SAAS,eAAe,UAAU,GAAG,EAAE,CAAC,EAAE;KACzD,aAAa,aAAa,SAAS;AACnC;IACD;AAGH,QAAI,eAAe,IAAI;KACrB,MAAM,cAAc,aAAa,UAAU,GAAG,WAAW;AACzD,SAAI,aAAa;MACf,MAAM,iBAAiB,IAAI,oBAAoB;OAC7C,SAAS,IAAI,eAAe;QAC1B,SAAS;QACT,mBAAmB;SACjB,GAAG,MAAM,QAAQ;SACjB,mBAAmB;QACpB;QACD,mBAAmB,MAAM,QAAQ;QACjC,YAAY,MAAM,QAAQ;QAC1B,kBAAkB,MAAM,QAAQ;QAChC,IAAI,MAAM,QAAQ;OACnB;OACD,MAAM;OACN,gBAAgB,MAAM;MACvB;MACD,MAAM;KACP;KACD,eAAe,aAAa,UAAU,WAAW;IAClD,WAEK,cAAc;KAChB,MAAM,iBAAiB,IAAI,oBAAoB;MAC7C,SAAS,IAAI,eAAe;OAC1B,SAAS;OACT,mBAAmB;QACjB,GAAG,MAAM,QAAQ;QACjB,mBAAmB;OACpB;OACD,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;KACN,eAAe;IAChB;GAEJ,OAAM;IAGL,MAAM,mBAAmB;IACzB,IAAI,aAAa;AACjB,SAAK,IAAI,IAAI,GAA6B,KAAK,GAAG,IAChD,KAAI,aAAa,SAAS,iBAAiB,UAAU,GAAG,EAAE,CAAC,EAAE;KAC3D,aAAa,aAAa,SAAS;AACnC;IACD;AAGH,QAAI,eAAe,IAAI;KAErB,MAAM,cAAc,aAAa,UAAU,GAAG,WAAW;AACzD,SAAI,aAAa;MACf,MAAM,eAAe,IAAI,oBAAoB;OAC3C,SAAS,IAAI,eAAe;QAC1B,SAAS;QACT,mBAAmB,MAAM,QAAQ;QACjC,mBAAmB,MAAM,QAAQ;QACjC,YAAY,MAAM,QAAQ;QAC1B,kBAAkB,MAAM,QAAQ;QAChC,IAAI,MAAM,QAAQ;OACnB;OACD,MAAM;OACN,gBAAgB,MAAM;MACvB;MACD,MAAM;KACP;KACD,eAAe,aAAa,UAAU,WAAW;IAClD,WAEK,cAAc;KAChB,MAAM,eAAe,IAAI,oBAAoB;MAC3C,SAAS,IAAI,eAAe;OAC1B,SAAS;OACT,mBAAmB,MAAM,QAAQ;OACjC,mBAAmB,MAAM,QAAQ;OACjC,YAAY,MAAM,QAAQ;OAC1B,kBAAkB,MAAM,QAAQ;OAChC,IAAI,MAAM,QAAQ;MACnB;MACD,MAAM;MACN,gBAAgB,MAAM;KACvB;KACD,MAAM;KACN,eAAe;IAChB;GAEJ;EACF;AAGD,MAAI,aAEF,KAAI,YAAY;GACd,MAAM,iBAAiB,IAAI,oBAAoB;IAC7C,SAAS,IAAI,eAAe;KAC1B,SAAS;KACT,mBAAmB,EAAE,mBAAmB,aAAc;IACvD;IACD,MAAM;GACP;GACD,MAAM;EACP,OAAM;GACL,MAAM,eAAe,IAAI,oBAAoB;IAC3C,SAAS,IAAI,eAAe,EAC1B,SAAS,aACV;IACD,MAAM;GACP;GACD,MAAM;EACP;CAEJ;CAED,AAAmB,wCACjBC,SACAC,aACA;EACA,MAAM,mBAAmB,MAAM,wCAC7B,SACA,YACD;EACD,iBAAiB,kBAAkB,oBAEhC,QAAgB;EAEnB,iBAAiB,oBAAoB;GACnC,GAAG,iBAAiB;GACpB,gBAAgB;EACjB;AACD,SAAO;CACR;;;;;;;;;;;;;;;;;;CAmBD,IAAI,UAAwB;AAC1B,SAAOC,iBAAS,KAAK,UAAU,CAAE;CAClC;CAqCD,qBAIEC,cAIAC,QAMI;EACJ,MAAM,gBAAgB,EAAE,GAAG,OAAQ;AAEnC,MAAI,eAAe,WAAW,QAC5B,cAAc,SAAS;AAEzB,SAAO,MAAM,qBAAgC,cAAc,cAAc;CAC1E;AACF"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@langchain/deepseek",
3
- "version": "1.0.7",
3
+ "version": "1.0.8",
4
4
  "description": "Deepseek integration for LangChain.js",
5
5
  "type": "module",
6
6
  "author": "LangChain",
@@ -14,7 +14,7 @@
14
14
  },
15
15
  "homepage": "https://github.com/langchain-ai/langchainjs/tree/main/libs/langchain-deepseek",
16
16
  "dependencies": {
17
- "@langchain/openai": "1.2.3"
17
+ "@langchain/openai": "1.2.4"
18
18
  },
19
19
  "peerDependencies": {
20
20
  "@langchain/core": "^1.0.0"
@@ -28,9 +28,9 @@
28
28
  "prettier": "^3.5.0",
29
29
  "typescript": "~5.8.3",
30
30
  "vitest": "^3.2.4",
31
+ "@langchain/core": "1.1.18",
31
32
  "@langchain/eslint": "0.1.1",
32
- "@langchain/core": "1.1.17",
33
- "@langchain/standard-tests": "0.0.20",
33
+ "@langchain/standard-tests": "0.0.21",
34
34
  "@langchain/tsconfig": "0.0.1"
35
35
  },
36
36
  "publishConfig": {