datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
tool-call-efficiency
tool-call-efficiency
Made with the whileai SDK · Collections: Efficiency, Start here: foundational post-training datasets
Teach an agent to make every tool call count.
An agent that calls a tool twice with the same arguments, looks up what
the user just told it, or keeps calling after the task is done is slow,
expensive, and harder to trust. Ask a base Qwen3-4B to work through
1,133 tool-using tasks across six agents and it does this a lot:
only 52% of its 6,681 rollouts finish… See the full description on the dataset page: https://huggingface.co/datasets/while-ai/tool-call-efficiency.tac-closing-efficiency-sft
TAC closing-efficiency slice
500 synthetic multi-turn tool-use trajectories that teach an agent to close bookings decisively — the welfare-neutral capability piece of the tool-use SFT mix used to train somaxsoma/qwen2.5-7b-tac-recovery-sft.
What it teaches
Built to fix the dominant failure mode observed on the TAC benchmark — the model reformulating search keywords in a loop and never closing a booking. Three patterns:
settle/browse (200): after failed keyword… See the full description on the dataset page: https://huggingface.co/datasets/somaxsoma/tac-closing-efficiency-sft.reasoning_efficiency
Reasoning Efficiency Evaluation Artifact
Anonymous review dataset accompanying the NeurIPS 2026 Evaluations & Datasets submission
“Diagnosing Reasoning Efficiency with Trace-Optional Evaluation”.
The artifact contains benchmark instances, raw visible model outputs, token/count metadata,
correctness and truncation flags, native workload metadata, derived model-level metrics,
and decomposition tables used by the paper.
Files
instances/*.jsonl.gz: benchmark prompts, gold… See the full description on the dataset page: https://huggingface.co/datasets/reasoning-efficiency-authors/reasoning_efficiency.