metadata

license: cc-by-4.0
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
dataset_info:
  features:
    - name: system
      dtype: string
    - name: in
      dtype: string
    - name: out
      dtype: string
  splits:
    - name: train
      num_bytes: 3795873
      num_examples: 2384
    - name: validation
      num_bytes: 799795
      num_examples: 500
    - name: test
      num_bytes: 798213
      num_examples: 500
  download_size: 1055715
  dataset_size: 5393881
task_categories:
  - text-generation
language:
  - en
tags:
  - code
  - dsl
  - structured-output
  - tool-calling
pretty_name: DIA Intent Sequencer for simple calculator
size_categories:
  - 1K<n<10K

dia-intent-sequencer-calculator-dataset

This dataset is used to prototype models for the DIA DSL module. It serves as a demonstration and testbed to evaluate, within the context of the DIA DSL, the model's capability to handle nested, composable intent calls and to generalize across randomized function and parameter names.

It is provided for demonstration and experimentation purposes only.

It pairs symbolic math expressions using basic operators (combining integers with the operators +, -, /, *) and parentheses for grouping, with symbolic DSL function calls compatible with the module DIA intent sequencer format. Function and parameter names are randomized to help reduce memorization, encouraging structural reasoning. Numbers in the symbolic expressions are randomized integers between -99999 and 99999, sampled to cover and balance different number lengths (1-digit to 5-digit numbers).

The dataset was programmatically generated using calculator_synthetic_data_generator.py, which builds symbolic expressions and randomized tool mappings.

This dataset is part of a set of three DIA datasets, each demonstrating different capabilities of intent sequencing and tool-based reasoning:

📦 Format

The dataset uses a wide format with the following columns:

system — the system prompt that defines the model's task
in — a basic symbolic math expression (e.g., (2 + 3) * 4)
out — a symbolic DSL output composed of nested tool calls (e.g., multiply(a=add(a=2, b=3), b=4) — in practice, function and parameter names are randomized)

Each entry follows the structure:

system → in → out

📋 System Prompt Format

The system prompt is a structured instruction that tells the model to translate the user input into one or more atomic DIA DSL function calls, based on a list of available arithmetic functions: add, subtract, divide, and multiply.

Each example includes:

A list of available intents (i.e., functions), each with a randomized name
Randomized parameter names to reduce overfitting
Descriptions of function behavior and argument roles

Using randomized function and parameter names helps reduce the risk that the model memorizes specific function names like add, multiply, or divide, encouraging it to rely on semantic understanding of the prompt and function definitions.

💡 Example: Instead of hardcoding add(a=2, b=3), a model may need to generate something like csisz(o=2, u=3), where csisz is the randomized function name for add.

The system prompt always ends with the instruction below:

QUERY_FILL cannot be used as no information can be retrieved at runtime.

This instruction is included because, unlike scenarios involving real-world environments (e.g., a robotic arm querying an inventory of screws), the calculator demo contains no missing information to retrieve. Every input provides all operands explicitly, so there's nothing for the model to "fill in" via external queries. The model must rely entirely on the input structure without any lookup behavior.

🧠 DIA DSL Project

The DSL used in this dataset is defined in the DIA module.

For an example using a demo model built with this demo dataset, see calculator.py.

🚀 Using the Dataset

To load the dataset:

from fifo_tool_datasets.sdk.hf_dataset_adapters.dsl import DSLAdapter
dataset_dict = DSLAdapter().from_hub_to_dataset_dict("a6188466/dia-intent-sequencer-calculator-dataset")
print(dataset_dict["train"])
print(dataset_dict["validation"])
print(dataset_dict["test"])

🔧 Upload & Editing Tools

Uploaded via fifo-tool-datasets using the dsl adapter.

You can edit or extend the dataset using its .dat format and CLI tools.

⚠️ Disclaimer & Limitations

This dataset was programmatically generated for demonstration and experimentation purposes. It is intended solely for prototyping models for the DIA DSL module.

It focuses on four basic arithmetic operations and does not reflect real-world data distributions or edge cases. The dataset should be independently evaluated and extended to meet the specific requirements of any use case.

This dataset is provided as is, without any warranties, express or implied. The authors and contributors assume no responsibility for its accuracy, completeness, or suitability for any purpose.

🪪 License

This dataset is licensed under CC BY 4.0. See LICENSE for details.