Skip to content

README.md

Latest commit

 

History

History
110 lines (92 loc) · 2.99 KB

README.md

File metadata and controls

110 lines (92 loc) · 2.99 KB

Guideline

The processed data will be saved in the data/train and data/test directories. The naming convention is <domain>__<dataset_name>_<dataset_size>, where <domain> is one of math, codegen, logic, simulation, table for now.

Prepare the training data from scratch

Math

DAPO+OR1

python data_preprocess/math/dapo_or1_merge_dedup_apr30.py

(Please see the comment if want to apply llm as judge)

Note:

Original: OR1 (105055) + DAPO (17917)

After dedup: 117192

After merging and removing instance with too long answer >100 char (script): 116632

DeepScaler

python data_preprocess/math/deepscaler_preview.py --train-sample-size <train_sample_size>

Code

leetcode2k

python data_preprocess/codegen/leetcode2k.py --train-sample-size <train_sample_size>

taco

python data_preprocess/codegen/taco.py --train-sample-size <train_sample_size>

primeintellect

python data_preprocess/codegen/primeintellect.py --train-sample-size <train_sample_size>

humaneval

python data_preprocess/codegen/humaneval.py

mbpp

python data_preprocess/codegen/mbpp.py

livecodebench

python data_preprocess/codegen/livecodebench.py

Logic

zebra_puzzle_dataset

python data_preprocess/logic/zebrapuzzle_gen/puzzle_generator.py --output_dir data/raw --num_puzzles <num_puzzles> --num_processes <num_processes>
cd ..
python data_preprocess/logic/process_zebrapuzzle_dataset.py

graph_logical_dataset

uv pip install pybind11
uv pip install Faker==37.1.0
cd data_preprocess/logic/graph_dataset_gen/
python logic.py --num_samples <num_samples>
cd ../../..  # return to Reasoning360
python data_preprocess/logic/process_graph_dataset.py

ordering_puzzle_dataset

uv pip install Faker==37.1.0
python data_preprocess/logic/puzzle_gen.py --test True --num_puzzles <num_puzzles>
python data_preprocess/logic/process_puzzles_dataset.py

ARC-AGI

python data_preprocess/logic/arcagi.py --name arcagi1
python data_preprocess/logic/arcagi.py --name arcagi2
python data_preprocess/logic/barc.py --train-sample-size <train_sample_size> --test-sample-size <test_sample_size>

Simulation

python data_preprocess/simulation/codeio.py --train-sample-size <train_sample_size> --test-sample-size <test_sample_size>

Table

uv pip install gdown
python data_preprocess/table/multihier.py

STEM

python data_preprocess/stem/webinstruct.py

Congratulations you get all raw data preprocessed into .parquet. You can check the model_filtering/ directory to perform difficulty-level filtering.

Add a new dataset

  1. Add a new script in data_preprocess/<domain>/<dataset_name>.py
  2. Add a new entry in tests/data_process/test_data_preprocess.py.
  3. Run pytest tests/data_process to check the functionality of the data preprocessing scripts.