datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
TheBioCollection
TheBioCollection
TheBioCollection is a 52.6B-token pretraining-scale corpus for biology that transforms heterogeneous biological resources into LLM training-friendly data. It is built through a construction pipeline that collects resources across biological domains, refines them through deduplication, entity tagging and augmentation, enriches them with tool-computed biological properties, and render them as instruction-form data with programmatically verifiable answers. The… See the full description on the dataset page: https://huggingface.co/datasets/trillionlabs/TheBioCollection.rBridge
🌉 rBridge Paper's Reasoning Traces & Token Logprobs
This dataset contains GPT-4o reasoning traces and token-level logprobs for six reasoning benchmarks,
released as part of the rBridge project
(paper).
rBridge uses these traces as gold-label reasoning references. By computing a weighted negative log-likelihood
over these traces — where each token is weighted by the frontier model's confidence — small proxy models (≤1B)
can reliably predict the reasoning performance of much larger… See the full description on the dataset page: https://huggingface.co/datasets/trillionlabs/rBridge.TheBioCollection-Eval
TheBioCollection-Eval
TheBioCollection-Eval is a biological evaluation suite for assessing large language models (BioLMs) for biology across small molecules, proteins, genomic sequences, cells/pathways, and cross-domain reasoning. It is constructed by drawing subtasks from many scattered existing benchmarks (Mol-Instructions, MolLangBench, BioReason-Pro, PerturBench Replogle K562) and combining them with source-derived newly-constructed instruction datasets.
Evaluation… See the full description on the dataset page: https://huggingface.co/datasets/trillionlabs/TheBioCollection-Eval.SimScholar-SFT
S3 SFT Trajectories
Complete ReAct trajectories for scientific-literature search.
Code ·
S3 collection ·
Source corpus
This dataset contains 14,633 single- and two-hop tool-use trajectories. In each
trajectory, a policy searches and reads a fixed scientific corpus through nine
tools, then submits an answer with a correctness label. The messages column
uses OpenAI tool-calling chat format.
At a glance
Question type
Rows
Correct
Incorrect
Single-hop… See the full description on the dataset page: https://huggingface.co/datasets/trillionlabs/SimScholar-SFT.MWMBenchNemoSlides-DPO-mix-v1.0
Slide-DPO
Direct Preference Optimization dataset for training LLMs to generate slide
presentations in Slidev markdown format, derived from the
Slides-Align human
preference rankings over the
SlidesGen-Bench benchmark.
Each row is a preference pair: a brief plus an available image pool as the
prompt, and two Slidev-markdown responses (with <think> reasoning traces)
that were generated by differently-ranked AI slide-generation products for
the same brief.
Row schema… See the full description on the dataset page: https://huggingface.co/datasets/trillionlabs/NemoSlides-DPO-mix-v1.0.rbridge-mask
rBridge-Mask
Evaluation dataset for rBridge, a method for predicting LLM reasoning performance using small proxy models. Contains reasoning traces from frontier models with <span> tags marking key reasoning steps.
Overview
Each sample contains a question and a reasoning trace where important factual/reasoning content is tagged with <span>...</span>. rBridge computes the masked log-likelihood — only scoring tokens inside tagged regions — to predict downstream… See the full description on the dataset page: https://huggingface.co/datasets/trillionlabs/rbridge-mask.SimScholar-RL
S3 RL Dataset
Synthetic literature-search questions for agentic reinforcement learning.
Code ·
S3 collection ·
Source corpus
The S3 RL dataset contains 13,000 synthetic, single-hop questions for training
and analyzing scientific-literature search agents. Each question is grounded
in one paper from a fixed corpus of approximately 1.12 million
computer-science papers.
At a glance
Property
Value
Rows
13,000
Language
English
Task
Tool-assisted… See the full description on the dataset page: https://huggingface.co/datasets/trillionlabs/SimScholar-RL.NemoSlides-SFT-mix-v1.0
NemoSlides-SFT-mix-v1.0
NemoSlides-SFT-mix-v1.0 is a SFT training set designed to train large language models to use Slidev to generate slides.