ishagarg1103/counter-sft-01-dataset
Counter-SFT-01 A synthetic conversational dataset for supervised fine-tuning on a constrained counter-planning task. The model must move a counter from start to target using increments of 1, 2, or 3, with at most five increments. Required response format: <counter_plan>{"increments":[3,3,2],"final":12}</counter_plan> Splits Split Rows Start range Templates micro_train 32 0-10 A train 480 0-59 A, B, C validation 90 60-69 A, B, C test 90 70-79 A, B… See the full description on the dataset page: https://huggingface.co/datasets/ishagarg1103/counter-sft-01-dataset.
Counter-SFT-01
A synthetic conversational dataset for supervised fine-tuning on a constrained counter-planning task.
The model must move a counter from start to target using increments of 1, 2, or 3, with at most five increments.
Required response format:
<counterplan>{"increments":[3,3,2],"final":12}</counterplan>
Splits
Columns
id: unique example identifierstart: initial counter valuetarget: required final valuedifference: target minus starttemplate: prompt-template identifierreference_answer: deterministic correct answermessages: system, user, and assistant messages used for SFT
All reference answers passed format, action, budget, arithmetic, and final-value validation. No exact example occurs in multiple splits.
Dataset SHA-256:
9ce84be29d68eaa6c5d5dd9d9d50d3935dbd4c587f1bdf199559292476f9c583
Load
from datasets import load_dataset
dataset = load_dataset( "ishagarg1103/counter-sft-01-dataset" )
