CoolFace
Datasetpublic

roshans89/mie-curriculum-sft-grade9

MIE Curriculum SFT Grade 9 This dataset contains answer-only supervised fine-tuning rows for Grade 9 curriculum tutoring. The data is provided as a single JSONL file: all_stamped.jsonl Dataset Summary Grade: 9 Language: English Rows: 42,771 Subjects: 19 KB-backed subunits covered: 509 Format: instruction/input/output JSONL Row Format Each training row contains: id grade subject unit_id subunit_id instruction input output Use these column… See the full description on the dataset page: https://huggingface.co/datasets/roshans89/mie-curriculum-sft-grade9.

sourceHugging Faceupdated 4mo agoView on Hugging Face
0likes13downloads
Dataset Card

MIE Curriculum SFT Grade 9

This dataset contains answer-only supervised fine-tuning rows for Grade 9 curriculum tutoring.

The data is provided as a single JSONL file:

text
all_stamped.jsonl

Dataset Summary

  • —Grade: 9
  • —Language: English
  • —Rows: 42,771
  • —Subjects: 19
  • —KB-backed subunits covered: 509
  • —Format: instruction/input/output JSONL

Row Format

Each training row contains:

text
id
grade
subject
unit_id
subunit_id
instruction
input
output

Use these column mappings:

text
Instruction column: instruction
Input column: input
Output column: output

The instruction value contains the grade instruction. The input value contains the student question. The output value is the assistant answer. The id, grade, subject, unit_id, and subunit_id values are audit and filtering columns.

Instructions start with this grade marker:

text
<GRADE_9> Answer using the Grade 9 curriculum.

Column Mapping

Map the columns manually:

text
instruction -> instruction
input -> input
output -> output

Use the input column as the prompt content and the output column as the supervised target. Ignore id, grade, subject, unit_id, and subunit_id during training.

Notes

  • —Assistant outputs contain final answer text only.
  • —No tool calls or routing traces are included.
  • —Source paths, bucket labels, and question type labels are not included in training rows.