---
language:
- en
- code
license: apache-2.0
task_categories:
- text-generation
tags:
- code
- instruction-tuning
- python
- chatml
size_categories:
- 1K<n<10K
configs:
- config_name: default
  data_files:
  - split: validation
    path: data/validation-*
dataset_info:
  features:
  - name: conversations
    list:
    - name: role
      dtype: string
    - name: content
      dtype: string
  - name: source
    dtype: string
  splits:
  - name: validation
    num_bytes: 1325626
    num_examples: 3000
  download_size: 75651
  dataset_size: 1325626
---

# AxiomCode-Instruct

A synthetic instruction-tuning dataset of Python coding tasks in ChatML conversation
format, designed to fine-tune a small code assistant model.

## Format

Each row contains a `conversations` list with a `user` turn (natural-language
instruction) and an `assistant` turn (raw Python solution), plus a `source` tag
indicating the category.

```json
{"conversations": [{"role": "user", "content": "<prompt>"}, {"role": "assistant", "content": "<code>"}], "source": "<tag>"}
```

## Categories

- string_utils
- math
- list_ops
- dict_collections
- algorithms
- file_io
- oop
- recursion

## Generation

Generated deterministically with seed 42 from templates, parametrization, and
randomization. No external data or API calls were used.