---
title: tlm-synth-ablation
canonical_url: "https://www.modelscope.cn/datasets/Shekswess/tlm-synth-ablation"
md_url: "https://www.modelscope.cn/datasets/Shekswess/tlm-synth-ablation.md"
repository: Shekswess/tlm-synth-ablation
last_updated: 2025-12-01
license: "Apache License 2.0"
storage_size: "238 MB"
downloads: 21
stars: 0
---

# tlm-synth-ablation

> tlm-synth-ablation - Shekswess 在 ModelScope 开源的数据集。Subset of PleIAs/SYNTH (first 113,000 rows, languages: en, exercises: math exercise, math mcq, mcq, memorization). Columns: text: user/assistant messages formatted with the chat template source: original…

Shekswess/tlm-synth-ablation 是 ModelScope 魔搭社区上的数据集，存储大小 238 MB，采用 Apache License 2.0 许可。

- **Repository**: Shekswess/tlm-synth-ablation
- **License**: Apache License 2.0
- **Storage size**: 238 MB
- **Downloads**: 21
- **Stars**: 0
- **Last updated**: 2025-12-01

Source: https://www.modelscope.cn/datasets/Shekswess/tlm-synth-ablation

---

## Dataset Description

Subset of PleIAs/SYNTH (first 113,000 rows, languages: en, exercises: math exercise, math mcq, mcq, memorization).
Columns:
- text: user/assistant messages formatted with the chat template
- source: original exercise name prefixed with PleIAs/SYNTH
- language: ISO language code
- num_tokens: token count of the formatted text using meta-llama/Llama-3.2-1B (no special tokens, capped at 2048 tokens)
Max rows requested: 113,000
Total tokens: 90,707,803
Ablation mode: capped at 10% of base rows (113,000 rows).
Language token totals:
  en: 90,707,803 tokens
Language share:
  en: 100.00% (113,000)
Exercise share:
  memorization: 94.00% (106,221)
  mcq: 2.70% (3,056)
  math exercise: 2.06% (2,323)
  math mcq: 1.24% (1,400)