rex099/Kimi-K2.6-Thinking-200x-Cleaned
Kimi K2.6 Thinking 200x — Cleaned Cleaned version of uniquealexx/Kimi-K2.6-Thinking-200x. Format — ShareGPT Each row has a single conversations field in ShareGPT format: { "conversations": [ {"from": "human", "value": "Find all positive integers n such that..."}, {"from": "gpt", "value": "<thinking> I need to find... </thinking> The answer is..."} ] } Usage with Unsloth from datasets import load_dataset dataset =… See the full description on the dataset page: https://huggingface.co/datasets/rex099/Kimi-K2.6-Thinking-200x-Cleaned.
019
Kimi K2.6 Thinking 200x — Cleaned
Cleaned version of uniquealexx/Kimi-K2.6-Thinking-200x.
Format — ShareGPT
Each row has a single conversations field in ShareGPT format:
{
"conversations": [
{"from": "human", "value": "Find all positive integers n such that..."},
{"from": "gpt", "value": "<thinking>
I need to find...
</thinking>
The answer is..."}
]
}Usage with Unsloth
from datasets import load_dataset
dataset = load_dataset("rex099/Kimi-K2.6-Thinking-200x-Cleaned", split="train")Changes from original
- Removed "The user asks/wants..." openings from all
<thinking>blocks — reasoning traces now read as standalone first-person reasoning - Fixed factual errors in reasoning traces and answers where the final answer contradicted the derivation
- Removed duplicate sentences where the same sentence appeared 3+ times consecutively
- Full reasoning trace length preserved (no trimming)