tamewild/PCSS-Qwen3-4B
Text Generation • 4B • Updated • 19
conversation listlengths 2 2 |
|---|
[
{
"content": "There are 5 houses in a row, numbered from 1 through 5 from left to right. Each house is home to one person and each resident has a unique attribute from the following characteristics:\n\nPreferred Furniture Style: Bohemian, Industrial, Modern, Rustic, Traditional\nNames: Frank, Charlie, Bob, Eve,... |
[
{
"content": "There are 5 houses in a row, numbered from 1 through 5 from left to right. Each house is home to one person and each resident has a unique attribute from the following characteristics:\n\nMode of Transportation to Work: Car, Scooter, Bus, Bike, Walk\nFavorite Cooking Method: Slow Cooking, Air Fryi... |
[
{
"content": "There are 5 houses in a row, numbered from 1 through 5 from left to right. Each house is home to one person and each resident has a unique attribute from the following characteristics:\n\nFavorite Store: Local Bookstore, Amazon, Online Craft Store, Specialty Coffee Shop, Target\nBirth Month: June,... |
[
{
"content": "There are 5 houses in a row, numbered from 1 through 5 from left to right. Each house is home to one person and each resident has a unique attribute from the following characteristics:\n\nFavorite Board Game: Codenames, Catan, Pandemic, 7 Wonders, Ticket to Ride\nNames: David, Frank, Charlie, Alic... |
[
{
"content": "There are 5 houses in a row, numbered from 1 through 5 from left to right. Each house is home to one person and each resident has a unique attribute from the following characteristics:\n\nFavorite Coffee: Espresso, Cold Brew, Latte, Mocha, Cappuccino\nFavorite Type of Art: Photography, Calligraphy... |
[
{
"content": "There are 5 houses in a row, numbered from 1 through 5 from left to right. Each house is home to one person and each resident has a unique attribute from the following characteristics:\n\nFavorite Smoothie: Green Smoothie, Tropical Twist, Mango Pineapple, Peanut Butter Banana, Berry Blast\nCompany... |
[
{
"content": "There are 5 houses in a row, numbered from 1 through 5 from left to right. Each house is home to one person and each resident has a unique attribute from the following characteristics:\n\nPet: Hamster, Rabbit, Fish, Cat, Dog\nFavorite Holiday: Thanksgiving, Christmas, Fourth of July, Easter, Hallo... |
[{"content":"There are 5 houses in a row, numbered from 1 through 5 from left to right. Each house i(...TRUNCATED) |
[{"content":"There are 5 houses in a row, numbered from 1 through 5 from left to right. Each house i(...TRUNCATED) |
[{"content":"There are 5 houses in a row, numbered from 1 through 5 from left to right. Each house i(...TRUNCATED) |
End of preview. Expand in Data Studio
Zebra 100
Overview
This synthetic micro-dataset contains 177 pure logic deduction traces (5x5 zebra puzzles), consisting of 100 training examples and 77 validation examples. It features strictly logic grid puzzles and includes zero mathematical data.
The 100 training examples were randomly sampled without replacement from the train split of tamewild/instruct5, while the 77 validation examples are kept exactly the same across both datasets.
It was created to minimize training time for our Qwen 3 4B Base experiments, and it also serves as an excellent sandbox for researchers looking to run rapid ablation studies or fast hyperparameter sweeps. Despite containing only 100 training examples, fine-tuning Qwen 3 4B Base on this subset was sufficient to elicit strong out-of-domain reasoning generalization.
For full training details and reproduction scripts, see the PCSS GitHub Repository.
Origin & Generation
As a direct subset of instruct5, this data shares the exact same generation methodology (generated using Qwen3-235B-A22B-Instruct-2507 with rejection sampling filtered by a correctness verifier).
For full details on the generation methodology, cost, and token usage, please refer to the instruct5 dataset card.
- Downloads last month
- 42
Models trained or fine-tuned on tamewild/zebra_100