I ran a fast experiment investigating how DeepSeek-R1 performs on agentic tasks, in spite of not supporting tool use natively, and I was quite amazed by preliminary outcomes. This experiment runs DeepSeek-R1 in a single-agent setup, where the model not only prepares the actions however likewise creates the actions as executable Python code. On a subset1 of the GAIA validation split, DeepSeek-R1 outperforms Claude 3.5 Sonnet by 12.5% outright, from 53.1% to 65.6% proper, and other designs by an even larger margin:

The experiment followed model usage standards from the DeepSeek-R1 paper and the design card: Don't utilize few-shot examples, prevent including a system prompt, and set the temperature to 0.5 - 0.7 (0.6 was utilized). You can find additional evaluation details here.

Approach

DeepSeek-R1's strong coding capabilities allow it to serve as an agent without being clearly trained for tool usage. By permitting the design to generate actions as Python code, it can flexibly engage with environments through code execution.

Tools are implemented as Python code that is included straight in the timely. This can be a simple function definition or a module of a bigger bundle - any valid Python code. The model then creates code actions that call these tools.

Results from executing these actions feed back to the design as follow-up messages, driving the next steps up until a last answer is reached. The agent framework is a simple iterative coding loop that mediates the conversation between the design and king-wifi.win its environment.

Conversations

DeepSeek-R1 is utilized as chat model in my experiment, where the model autonomously pulls extra context from its environment by utilizing tools e.g. by using a search engine or bring data from . This drives the conversation with the environment that continues till a final response is reached.

In contrast, o1 designs are known to carry out improperly when utilized as chat models i.e. they do not attempt to pull context throughout a conversation. According to the linked article, o1 models perform best when they have the complete context available, with clear guidelines on what to do with it.

Initially, I likewise attempted a full context in a single timely approach at each step (with results from previous steps consisted of), however this led to substantially lower scores on the GAIA subset. Switching to the conversational technique explained above, I was able to reach the reported 65.6% efficiency.

This raises an intriguing question about the claim that o1 isn't a chat design - possibly this observation was more pertinent to older o1 designs that lacked tool use abilities? After all, isn't tool use support an essential mechanism for enabling models to pull additional context from their environment? This conversational method certainly appears efficient for DeepSeek-R1, though I still need to conduct similar explores o1 designs.

Generalization

Although DeepSeek-R1 was mainly trained with RL on mathematics and coding tasks, it is amazing that generalization to agentic tasks with tool use through code actions works so well. This capability to generalize to agentic tasks reminds of current research study by DeepMind that shows that RL generalizes whereas SFT memorizes, although generalization to tool use wasn't examined because work.

Despite its ability to generalize to tool usage, DeepSeek-R1 typically produces long reasoning traces at each action, compared to other designs in my experiments, limiting the usefulness of this design in a single-agent setup. Even simpler jobs in some cases take a long period of time to complete. Further RL on agentic tool use, be it by means of code actions or not, might be one choice to enhance performance.

Underthinking

I also observed the underthinking phenomon with DeepSeek-R1. This is when a thinking model regularly switches between various reasoning thoughts without adequately exploring promising paths to reach an appropriate service. This was a significant factor for overly long thinking traces produced by DeepSeek-R1. This can be seen in the tape-recorded traces that are available for download.

Future experiments

Another common application of reasoning designs is to utilize them for preparing just, while using other designs for generating code actions. This could be a potential new function of freeact, if this separation of functions shows useful for more complex jobs.

I'm likewise curious about how thinking models that currently support tool usage (like o1, o3, ...) carry out in a single-agent setup, with and without generating code actions. Recent advancements like OpenAI's Deep Research or Hugging Face's open-source Deep Research, larsaluarna.se which also uses code actions, look intriguing.