Description #

Deep reasoning capability based on the DeepSeek-R1 model
Use the DeepSeek-R1 model with deep reasoning capability to analyze and understand text, and return structured output. Before answering a question, the model performs question analysis and decomposition, and generates a response based on this process.

Key features #

• Deep reasoning mechanism: Before answering a question, the model automatically decomposes the question and performs logical reasoning, generating a chain of thought (reasoning_content) to enhance the comprehensiveness and depth of the answer.
• Plain text input: This operator variant is designed specifically for text tasks and only supports input from the texts column.
• Flexible reasoning mode: Supports controlling the deep reasoning mode via the thinking_type parameter (enabled / disabled / auto), allowing flexible trade-offs between answer quality and performance.

Input and output specifications #

• Input format:
  • texts (text): string or list type, used as the user's question or prompt.
• Output format:
  • Default mode: Returns a struct type containing:
    • llm_result: The final answer generated by the model.
    • reasoning_content: The model's chain of thought or reasoning process.
  • Diagnostic and traceability mode:
    • When the environment variable LAS_LLM_FINISH_REASON_CHECK=true is set, the struct additionally includes a finish_reason field to determine the reason for output termination (such as stop, length).
    • When the environment variable LAS_LLM_BOTS_REFERENCES=true is set, the struct additionally includes a references field (JSON string) to carry external knowledge information referenced by the model.

Version compatibility notes #

This operator (ArkLLMThinkingVision) is used differently in Daft 0.6.5 (and earlier versions) compared to version 0.6.14 (and later versions).

• In Daft 0.6.14 and later versions, you can specify the images, texts, and videos fields as input, corresponding to the model's image, text, and video inputs, and simultaneous input is supported. The following is an example:

  df = df.with_column(
  
          "llm_result",
  
          las_udf(
  
              ArkLLMThinkingVision,
  
              construct_args={
  
                  "model": "deepseek-r1",
  
                  "system_text": "",
  
                  "inference_type": "online",
  
              },
  
          )(texts=col("query")),
  
      )
  

• In Daft 0.6.5 and earlier versions, only video or image input is supported, not both at the same time. The input type is specified via multimodal_type, which supports "video" and "image", "text". It also supports providing text information by specifying prompt. The following is an example:

df = df.with_column(

        "llm_result",

        las_udf(

            ArkLLMThinkingVision,

            construct_args={

                "model": "deepseek-r1",

                "multimodal_type": "text", 

                "inference_type": "online",

            },

        )(col("query")),

    )

Operator parameters #

Input #

Output #

By default, the returned field type is struct, which contains the following fields:

• llm_result: The final answer content output by the model.
• reasoning_content: The content of the model's chain of thought output (reasoning process).

When the relevant environment variables are set, the returned struct will additionally include the following fields:

• finish_reason (when LAS_LLM_FINISH_REASON_CHECK=true): The reason for the model output ending. Common values include stop (normal termination), length (reached maximum length), and content_filter (triggered content filtering policy).
• references (when LAS_LLM_BOTS_REFERENCES=true): Reference information in JSON string format, used to carry external knowledge referenced by the model.

Parameters #

If a parameter does not have a default value, it is required.

Examples #

The following code demonstrates how to use the ArkLLMThinkingVision.Deepseek-R1 operator for in-depth text reasoning.

from __future__ import annotations

import os

import daft
from daft import col
from daft.las.functions.ark_llm.ark_llm_thinking_vision import ArkLLMThinkingVision
from daft.las.functions.udf import las_udf

if os.getenv("DAFT_RUNNER", "native") == "ray":
    import logging

    import ray

    def configure_logging():
        logging.basicConfig(
            level=logging.INFO,
            format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
            datefmt="%Y-%m-%d %H:%M:%S.%s".format(),
        )
        logging.getLogger("tracing.span").setLevel(logging.WARNING)
        logging.getLogger("daft_io.stats").setLevel(logging.WARNING)
        logging.getLogger("DaftStatisticsManager").setLevel(logging.WARNING)
        logging.getLogger("DaftFlotillaScheduler").setLevel(logging.WARNING)
        logging.getLogger("DaftFlotillaDispatcher").setLevel(logging.WARNING)

    ray.init(dashboard_host="0.0.0.0", runtime_env={"worker_process_setup_hook": configure_logging})
    daft.set_runner_ray()

daft.set_execution_config(min_cpu_per_task=0)

if __name__ == "__main__":
    # Environment variable LAS_API_KEY must be configured: LAS_API_KEY can be obtained by creating it on the LAS service page
    queries = {"query": [""]}

    df = daft.from_pydict(queries)
    df = df.with_column(
        "llm_result",
        las_udf(
            ArkLLMThinkingVision,
            construct_args={
                "model": "deepseek-r1",
                "system_text": "",
                "inference_type": "online",
            },
        )(texts=col("query")),
    )

    df = df.with_column("reasoning_content", col("llm_result")["reasoning_content"])
    df = df.with_column("llm_result", col("llm_result")["llm_result"])
    df.show()

    # Output (the result of each large model inference may vary)
    # +------------------------------------+--------------------------------------------------+--------------------------------------------------------------------+
    # | query                              | llm_result                                       | reasoning_content                                                  |
    # | Utf8                               | Utf8                                             | Utf8                                                               |
    # +====================================+==================================================+====================================================================+
    # | Help me plan a 10-day trip to Xinjiang in May    | The following is a 10-day itinerary for Xinjiang in May, focusing mainly on Northern Xinjiang... | | Okay, the user asked me to help plan a 10-day trip to Xinjiang in May. First... |
    # +------------------------------------+--------------------------------------------------+--------------------------------------------------------------------+