---
tags:
- Gpt
- Gpt5.5
- Gpt5.6
- Opus
- Fable
- Mythos
- Gemini
- Think
- ThinkTraces
- FrontierTraces
- Qwen
- Kimi
- Grok
---
# Frontier_Traces_Pre_Train

**Massive merged & deduplicated frontier model distillation traces** for pre-training from scratch.

## Dataset Details
- **Size**: 316,299 unique examples
- **Format**: Single `text` column (all conversations normalized)
- **Deduplicated**: Yes (MD5 hash on full text)
- **Sources**: 
  - GPT-5.x series
  - Claude Opus/Mythos
  - Grok 4.x
  - Gemini 3.x / Pro
  - Qwen
  - DeepMind Alpha
  - Fable traces
  - And many more high-quality distilled reasoning & coding traces

## Usage
```python
from datasets import load_dataset

dataset = load_dataset("yourusername/Frontier_Traces_Pre_Train", split="train")
print(dataset[0])