Datasets:
license: apache-2.0
task_categories:
- text-generation
- question-answering
language:
- en
pretty_name: Fable 5.1's Style
tags:
- fable 5.1
- coding
- synthetic
- thinking
- think
- reason
- reasoning
- distill
- distillation
- agent
- agentic
- SFT
- CoT
- code
- programming
- thought
- thoughts
- SWE
- tool-use
- mythos 5.1
- mythos
- fable class
- mythos class
- fable 5
- mythos 5
size_categories:
- 100K<n<1M
configs:
- config_name: default
data_files:
- split: train
path: train.jsonl
Dataset Description
This dataset contains 10,000 agentic coding and reasoning multi-turn traces took from the Step 3.5 Flash SFT Code dataset.
It was remade to sound very similar to the Fable 5.1 model on max reasoning effort in Fable-5.1-Max-Reasoning-Filtered-10000x.
It holds over 2,000,000,000 tokens of step-by-step chain-of-thought programming across multiple complex coding domains.
It has also been deduplicated and filtered to remove no-reasoning and lower-quality traces, keeping only high-quality slow reasoning traces.
Dataset Statistics
| Metric | Value |
|---|---|
| Total Examples | 473,635 Traces |
| Total Token Count | ~2,000,000,000 Tokens |
| Total Dataset Size | 6.8 GB |
| Average Trace Size | 14.3 KB |
| Average Token Count | ~4,500 Tokens |
Dataset Contents & Coverage
The dataset includes step-by-step problem-solving for complex coding tasks, including:
Algorithm design, implementation, and performance optimization.
Advanced debugging and error-handling.
Multi-step logic design and compliance with complex prompt constraints.
Uses
Distilling Fable 5.1-style agentic coding and reasoning down to smaller LLMs.
Improve general coding and reasoning quality.
Teaching models to generate clear chain-of-thought steps and tool-use before outputting their final answer.