by IFM
Pretrain-Behaviors is a behavior-focused text dataset from IFM covering reasoning, planning, data science, games and format rewriting. It is part of the K2 Horizon collection.
DatasetFree
Step 1: Install the datasets library.
Step 2: Pick a subset such as reasoning, planning or games.
Step 3: Load it with load_dataset("IFM/Pretrain-Behaviors", "reasoning", split="train", streaming=True).
Step 4: Print dataset.features to check the fields, since they differ by subset.
Step 5: Combine subsets with your own sampling weights and run your own risk checks and filtering.
Apache License 2.0.
reasoning, general, planning, data-science, games, other and format-rewrites.
No. Field names and nested structures can differ by subset, so inspect the features before building a pipeline.
Yes. Each subset has a train split in Parquet shards that supports streaming.