Datasets:

em908
/

analyst-data

language: en
license: apache-2.0
task_categories:
  - text-generation
tags:
  - code
  - pandas
  - data-analysis
size_category: small
configs:
  - config_name: default
    data_files:
      - split: train
        path: train.jsonl
      - split: eval
        path: eval.jsonl

analyst-data

Training corpus for analyst-v1: a small pandas-codegen model.

613 train / 165 eval-pool chat-format pairs (eval pool = pairs with constructible test DataFrames, held out of training)
Sources: DS-1000 seed (real Kaggle-style data problems) + synthetic pairs generated from live Kaggle datasets (every pair execution-verified against the real DataFrame)

Format: {"system": ..., "user": "Schema:... Question:...", "assistant": ""}