datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
coding-variant
GenBench CoCG QA Dataset
Multi-hop genetic reasoning QA items generated from GenBench's knowledge
graph (Ensembl, ClinVar, VEP, BioGRID, STRING, Reactome, UniProt, GO,
SIGNOR, OmniPath, KEGG, DisGeNET, OpenTargets, PubTator3, GTEx, and more),
built for CoCG (Co-Evolving Confidence Graph) agent training.
2513 items across 11 task types.
Task types
task_type
count
coding_variant
53
conservation_reasoning
246
counterfactual
246
disease_reasoning
246… See the full description on the dataset page: https://huggingface.co/datasets/genbench-iitp/coding-variant.single-turn-eval-stage1_proof_pr_delta_variants-n32
Single-turn eval — violetxi/stage1_proof_pr_delta_variants
Generated by teaching/inference/single_turn_eval_vllm.py. One row per problem; samples is the list of model responses, scores is per-sample correctness, and mean/best/worst are the aggregates used by mean@N / best@N / worst@N.
Eval results (n_samples_per_example = 32)
Overall
metric
value
n_examples
1006
mean@32
0.1922
best@32
0.4175
worst@32
0.0457
pass_rate
0.4175… See the full description on the dataset page: https://huggingface.co/datasets/violetxi/single-turn-eval-stage1_proof_pr_delta_variants-n32.adaptive_rag_hotpotqa
Adaptive RAG HotpotQA Dataset
This dataset is a processed version of HotpotQA designed for training Adaptive Retrieval-Augmented Generation (RAG) systems.
Features
input: The input text for the model
output: The target output text
retrieval_label: Whether retrieval is needed (0/1)
hop: The reasoning hop number (1 or 2)
type: The type of example (multi_hop_qa, single_hop_qa, multi_hop_gating, etc.)
metadata: Additional information about the example including:
answer:… See the full description on the dataset page: https://huggingface.co/datasets/varun500/adaptive_rag_hotpotqa.