Hello, despite being in the latter half of September, it is still quite hot in Japan. The photos feel mismatched, but I'm deliberately sticking to the autumn theme, hoping it will get cooler soon. However, it might stay hot for the rest of the month.

Now, about the fine-tuning of ChatGPT-3.5 that I introduced the other day, it's certainly a hot topic. I think there is a strong demand in companies to specialize its performance for specific tasks. For this reason, we conducted an experiment assuming cases where you would want to proceed even without data at hand by generating synthetic text and then fine-tuning it.

Experiment Details

Just like the previous experiment, we set a task to determine which financial product a given English-language complaint is about. They are complaints for the banking industry, so the task involves differentiating between six types of financial products such as mortgages and bank accounts. The data used for fine-tuning was minimal, with 100 samples for validation, just like last time. However, the training data is different this time. We generated customer complaint emails using GPT-4, and they are indistinguishable from real ones at a glance. GPT-4's performance is indeed impressive. We generated 15 similar customer complaints for training and then proceeded with fine-tuning.

2. Experiment Results

Since this was our first time using synthetic text, we were worried about the outcome, but we were able to confirm the effectiveness of fine-tuning as follows. Though the improvement isn't dramatic with just 15 samples, the accuracy for this task has improved compared to the base GPT-3.5, which had an accuracy of 0.5 to 0.55.

For more details on the experiment, please refer to this notebook.

3. Discussion

Fine-tuning with synthetic text was a method not even considered before, but with the arrival of GPT-4, it's becoming more realistic. There are several points to consider, such as the number of samples and how to write prompts, but the advantage of being able to start even without data is significant. Currently, GPT-4 is the only option for generation models, but it seems like new models like Gemini from Google will also be available next year. Technology is advancing rapidly, so we can expect a lot more in the future.

So, what did you think? We will continue to conduct various experiments and share our findings here. See you again soon!

Notice: ToshiStats Co., Ltd. and I do not accept any responsibility or liability for loss or damage occasioned to any person or property through using materials, instructions, methods, algorithms or ideas contained herein, or acting or refraining from acting as a result of such use. ToshiStats Co., Ltd. and I expressly disclaim all implied warranties, including merchantability or fitness for any particular purpose. There will be no duty on ToshiStats Co., Ltd. and me to correct any errors or defects in the codes and the software.

Since GPT4 , one of the most intelligent Large Language Model (LLM) in the world, was released on 14 March 2023, many people are surprised because it is very intelligent. This is great. But there is one problem for users. It is not free service. Users should pay the fee of GPT4, based on how much tokens they use on GPT4. Therefore when we continue to use GPT4 all day long, it must be very expensive. Of course we prefer more intelligence. But we should consider the cost of it. There is a trade off between them. What should we do to solve it? Last week, I found a good research paper called “Large Language Models as Tool Makers“(1). All charts below come from this awesome paper. The idea is simple and looks promising to tackle these problems. So let me explain more details.

1. Tool user and Tool maker

The basic idea is as follows. Let us have two LLMs, one is called “Tool maker” and another is called “Tool user”. When a new task comes to us, Tool maker create “tools” for this task. Once “tools” are ready, they are passed to Tool user for inference. These tools are reusable to solve similar tasks in the future. So GPT4 can be used only as Tool maker as it should be more intelligent. Light weights models, such as GPT3.5, can be used as a Tool user. Then we can reduce the cost of computation for inference. It sounds great! The chart below explains how it works.

2. How can we create tools for our task?

As we want to keep the accuracy of the results, Tool maker should create better tools. There are three steps to do that.

• Tool Proposing: The tool maker generates a Python function to solve the given task. If the proposed tool makes errors, the tool maker makes another tool.

• Tool Verification: The tool maker generates unit tests using validation samples and subsequently executes these tests. 3 validation samples are prepared here. If the tool fails any of these tests, the tool makes an attempt to fix the issues. The paper explains it as follows “This stage fulfills two key roles: 1) it provides examples that demonstrate how to convert natural language questions into function calls, and 2) it verifies the tool’s reliability, enabling the entire process to be fully automated.”

• Tool Wrapping: If the execution or verification passes the preset threshold, tool maker prepares the wrapped tool for tool user. This step involves wrapping up the function code and providing demonstrations of how to convert a task into a function call. This final product is then ready for use by the tool user.

The chart below shows us how it works.

Once the tool is ready, the tool is passed to the tool-user. The tool user should solve various instances of the task by using tools made by the Tool maker. The prompt for this stage is the wrapped tool which contains the function for solving the task and demonstrations of how to convert a task query into a function call. With the demonstrations, tool user can then generate the required function call in an in-context learning fashion. The function calls are then executed to solve the task. The chart below shows us how the processes from Tool maker to Tool user are going.

3. Can we confirm if tools that fit our tasks are available or not?

Here, we use a third LLM called “the dispatcher”. Because the dispatcher maintains a record of existing tools produced by the tool maker, it can confirm if tools that fit our task are available at the time when a task is received, If no appropriate tool is found, the dispatcher identifies the instance as a new task and solves the instance with a powerful model, such as GPT4. The dispatcher’s workflow is shown here.

That is it! This is a major part of “Large Language Models as Tool Makers” or “LATM” in short. By LATM, we might reduce computational cost for heavy models, such as GPT4. It is amazing! Hope you enjoy the article today. I will update new technologies around LLM in the near future. Stay tuned!

1) “Large Language Models as Tool Makers” Tianle Cai, Xuezhi Wang, Tengyu Ma, Xinyun Chen, Denny Zhou, 26 May 2023, https://arxiv.org/abs/2305.17126

Fine-tuning GPT-3.5 with synthetic text generated by GPT-4. The accuracy has improved! In the future, we might not even need training text???

"Large Language Models as Tool Makers" is a good idea to enhance the accuracy of the model while keeping the cost of computation low!