laion/nemotron-terminal-adapters_math

---
license: cc-by-4.0
task_categories:
- question-answering
language:
- en
tags:
- code
- terminal
- agent
- trace
- sft
configs:
- config_name: default
data_files:
- split: train
path: data.parquet
---

# nemotron-terminal-adapters_math

Per-source partition of [nvidia/Nemotron-Terminal-Corpus](https://huggingface.co/datasets/nvidia/Nemotron-Terminal-Corpus),
filtered to `source == "adapters_math"`. The `difficulty` column preserves the original
`easy` / `medium` / `mixed` split (`na` for the `dataset_adapters/*` files, which
did not carry a difficulty label).

Partitioning scheme:
- **adapters_{code,math,swe}** — rows from `dataset_adapters/{code,math,swe}.parquet`
- **{skill}** (e.g. `debugging`, `security`, …) — rows from
`synthetic_tasks/skill_based/{easy,medium,mixed}/{skill}/data_filtered.parquet`

## Columns

Same as the source dataset (`conversations`, `agent`, `model`, `model_provider`,
`date`, `task`, `episode`, `run_id`, `trial_name`, `enable_thinking`) plus:
- `source` — the partition key (`"adapters_math"` throughout this repo)
- `difficulty` — `easy` / `medium` / `mixed` / `na`
- `original_source` — only present in `adapters_code`; preserves the original
`source` column value (`OpenCodeReasoning` or `synthetic`) from the upstream file.

## Citation

```bibtex
@misc{pi2026dataengineeringscalingllm,
title={On Data Engineering for Scaling LLM Terminal Capabilities},
author={Renjie Pi and Grace Lam and Mohammad Shoeybi and Pooya Jannaty and Bryan Catanzaro and Wei Ping},
year={2026},
eprint={2602.21193},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2602.21193},
}
```

Original dataset license: CC-BY-4.0.

---
license: CC-BY-4.0
task_categories:
- 问答（question-answering）
language:
- 英语（en）
tags:
- 代码（code）
- 终端（terminal）
- AI智能体（Agent）
- 追踪（trace）
- 监督微调（SFT）
configs:
- config_name: default
data_files:
- split: 训练集（train）
path: data.parquet
---

# nemotron-terminal-adapters_math

本数据集为[nvidia/Nemotron-Terminal-Corpus](https://huggingface.co/datasets/nvidia/Nemotron-Terminal-Corpus)按源分区后的子集，仅筛选保留`source == "adapters_math"`的样本。其中`difficulty`字段保留了原始数据集的`easy`（简单）、`medium`（中等）、`mixed`（混合）划分标准；对于`dataset_adapters/*`系列文件，由于其未附带难度标签，故该字段值为`na`。

分区方案：
- **adapters_{code,math,swe}** — 取自`dataset_adapters/{code,math,swe}.parquet`的样本行
- **{skill}**（例如`debugging`（调试）、`security`（安全）等） — 取自`synthetic_tasks/skill_based/{easy,medium,mixed}/{skill}/data_filtered.parquet`的样本行

## 字段说明

字段与原始数据集一致，包含`conversations`、`agent`、`model`、`model_provider`、`date`、`task`、`episode`、`run_id`、`trial_name`、`enable_thinking`，额外新增字段如下：
- `source` — 分区键（本仓库中所有样本的该字段值均为`"adapters_math"`）
- `difficulty` — 难度标签，可选值为`easy` / `medium` / `mixed` / `na`
- `original_source` — 仅在`adapters_code`分区中存在，保留上游文件中原始的`source`字段值（`OpenCodeReasoning`或`synthetic`）

## 引用格式

bibtex
@misc{pi2026dataengineeringscalingllm,
title={面向扩展大语言模型（LLM）终端能力的数据工程研究},
author={Renjie Pi and Grace Lam and Mohammad Shoeybi and Pooya Jannaty and Bryan Catanzaro and Wei Ping},
year={2026},
eprint={2602.21193},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2602.21193},
}

原始数据集许可证：CC-BY-4.0.