datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Nemotron-RL-Agentic-Conversational-Tool-Use-Pivot-v1
Dataset Description:
We created an RL dataset for conversational tool-use by utilizing existing expert tool-use trajectories. We pose each assistant step of the trajectory as a separate behavior cloning problem where the policy model is incentivized to match the tool call choices of the expert model. Each trajectory includes the use of tools for authentication, data lookup, servicing (i.e. booking reservations, changing them, getting discounts, etc), and more across 838 different… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-RL-Agentic-Conversational-Tool-Use-Pivot-v1.reason-tool-use-demo-1500
Dataset info
The dataset is a selection of reasoning toolcalls data from https://huggingface.co/datasets/interstellarninja/hermes_reasoning_tool_use, which contains data from Hermes-Tools、Glaive-FC、ToolAce、Nvidia-When2Call.
The format has been transformed to adapt llama-factory v1 training pipeline.
browser-tool-use-trajectories
Browser Tool Use Trajectories
Rights & intended use: legacy public research corpus / portfolio
artifact. Hosted frontier-model outputs are research-only inputs under
project policy (synthetic-factory#161):
intended_use: research_only, project_training_policy: blocked. Not
training data for any model-weight update. Machine-readable record:
rights.json.
Release status: The raw, uncurated payload is now published under
data/raw/. It is available for inspection and… See the full description on the dataset page: https://huggingface.co/datasets/rmems/browser-tool-use-trajectories.data-use-sft-tiered
Data-use SFT — tiered workflow (two task subsets)
Multitask SFT anchored exclusively on mentions the tiered extractor emits
(T1 evidential ∪ T2 declaration; see
rafmacalaba/data-use-mentions-tiered). Every row carries task
("provenance" | "usage_impact") and origin (prwp | fcv). Rows whose anchor
span was judged T3 (non-mention) or junk are dropped — audit trail in
manifest.jsonl (provenance) and manifest_usage.jsonl (usage/impact).
task = provenance (22,201 rows)… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/data-use-sft-tiered.data-use-mentions
Data-use mentions (NER / span extraction)
Data mentions extracted from World Bank Policy Research Working Papers and FCV documents, validated by a
context-only LLM judge, and formatted for span-extraction (GLiNER / GLiNER2) and
token-classification (LFM2.5-encoder) fine-tuning.
Labels
Three entity types (the judge's specificity axis):
NAMED_DATA — a proper name, title, or acronym of a specific data source
DESCRIPTIVE_DATA — a source described in words but not… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/data-use-mentions.data-use-mentions-tiered
Data-use mentions — tiered copy (T1∪T2-only supervision)
Derived from rafmacalaba/data-use-mentions (originals untouched).
Same windows, same text; spans judged tier3_nonmention or junk
(Luna verdicts; unjudged train spans via v3 tier classifier at
p_t3+p_junk >= 0.9) are UNTAGGED — text stays, so they
act as hard negatives for span-extraction training.
Total untagged: 28807 spans.
Labels: single DATA_MENTION class (kept spans = T1 evidential ∪
T2 declaration). Specificity… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/data-use-mentions-tiered.user_study-preference-personalized_0423_base_filtered
Filtered user study dataset
Source repo: ehejin/user_study-preference-personalized_0423_base
Each row is ONE item review (pre-rating, conversation, post-rating). Submission-level
fields (prolific_pid, demographics, background) are duplicated across rows that share
a submission.
The 25-50 rows here are the FIRST review for each unique pool index, selected the same
way the analysis plot uses — see scripts/plot_vote_shift_3way.py.
Total rows: 50
data-use-ner
Data-use-ner (human holdout)
GLiNER-format human-adjudicated holdout: 473 spans — annotator190 (190, origin=fcv_pads_east_africa) + jdc283 (283, origin=jdc_operational). Never trained on.
Source: rafmacalaba/datause-displacement-reviewed holdout (gliner_reviewed token spans + readable_reviewed passages, v2.4 labels) with v3 probe head_score (outputs/gliner_datause_v3_probe_human473.jsonl).
Columns
text (full passage = " ".join(tokenized_text); span char offsets… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/data-use-ner.UserBehavioralDivergence-simulated-conversationsEvoEval_tool_usepi-computer-use-sessions
Coding agent session traces for thomasmustier/pi-computer-use-sessions
This dataset contains redacted coding agent session traces collected while working on https://github.com/tmustier/pi-computer-use. The traces were exported with pi-share-hf from local pi workspaces and filtered to keep only sessions that passed deterministic redaction, secret scanning, visual review where applicable, and LLM review.
Source git repo: https://github.com/tmustier/pi-computer-use
Data… See the full description on the dataset page: https://huggingface.co/datasets/thomasmustier/pi-computer-use-sessions.data-use-mentions-v2
data-use-mentions v2 (configs: gliner2_v2 | bio_v2 | gliner_v2)
Facets (row-aligned with rafmacalaba/data-use-mentions v1 formats, filtered to Luna-cleaned rows):
gliner2_v2 — input/output + output_meta overlay
bio_v2 — {tokens, ner_tags} verbatim from v1 bio, cleaned rows only
gliner_v2 — {tokenized_text, ner [start,end,LABEL]} verbatim from v1 gliner, cleaned rows only
Only rows whose spans were judged by the Luna relabel pipeline (annotation_guidelines.md v2.3,
dual-pass… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/data-use-mentions-v2.data-use-annotations
Data-use annotations
Public store of keep/drop rulings from the annotation review app (human_labeling/review.html).
Files
rulings/<annotator>.jsonl — one file per annotator, one JSON object per ruling: key (span UID), ruling (DATA_MENTION keep / NON_MENTION drop), queue (gold / sample), annotator (required, set in the UI), ts. Last write per (queue, key, annotator) wins.
from datasets import load_dataset
ds = load_dataset("rafmacalaba/data-use-annotations") #… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/data-use-annotations.amazon2023-user-interactions
Amazon Reviews 2023 — User Interactions (5-core, leave-one-out, sequential)
User–item interaction data for five Amazon Reviews 2023 categories, processed
into ready-to-use sequential / generative recommendation splits with the
de-facto standard recipe (5-core filtering → chronological ordering →
leave-one-out split).
Every record keeps the timestamp, and the splits are byte-for-byte
reproducible from the official Amazon Reviews 2023 release; the statistics also
match, exactly… See the full description on the dataset page: https://huggingface.co/datasets/yufan/amazon2023-user-interactions.user_responsetool-use-preference-pairs
Tool Use Preference Pairs
Rights & intended use: legacy public research corpus / portfolio
artifact. Hosted frontier-model outputs are research-only inputs under
project policy (synthetic-factory#161):
intended_use: research_only, project_training_policy: blocked. Not
training data for any model-weight update. Machine-readable record:
rights.json.
Release status: The raw, uncurated payload is now published under
data/raw/. It is available for inspection and reproducibility… See the full description on the dataset page: https://huggingface.co/datasets/rmems/tool-use-preference-pairs.us-earthquake
U.S. Earthquake Dataset
This dataset contains earthquake events in the U.S. from January 1, 2020, to December 31, 2023. It inclucdes 3,009 sequences with 29,521 events across 3 magnitude types. The original data can be accessed via USGS Earthquake Search. The detailed data preprocessing steps used to create this dataset can be found in the TPP-LLM paper and TPP-Embedding paper.
Update (2025-10-28): Added three timestamp fields (timestamp_event, timestamp_since_start… See the full description on the dataset page: https://huggingface.co/datasets/tppllm/us-earthquake.UserToolBench
UserToolBench
A User-Profile-Hidden Benchmark for Personalized Decision Making in Tool-Use LLMs
Can a tool-use LLM make the right decision for a particular user when the explicit user profile is hidden?
UserToolBench evaluates whether LLM agents can recover stable user preferences from interaction history, decide when clarification is necessary, and produce executable, user-aligned tool-call trajectories under incomplete information.
Overview… See the full description on the dataset page: https://huggingface.co/datasets/XuexiongYin/UserToolBench.verified-tool-use-dataset
Verified tool-use trajectories for LLM agents
This was a time-boxed experiment by an autonomous agent (Protogonos), now concluded. Nothing here is offered for sale or for hire, and no payment is accepted.
Multi-turn function-calling conversations for training and evaluating
tool-using agents — 48 trajectories across 16 domains, with every tool call
checked against its tool's JSON-Schema. The free sample in this repo is a real
slice of the full set: the viewer above renders it… See the full description on the dataset page: https://huggingface.co/datasets/protogonos/verified-tool-use-dataset.data-use-mentions-extended
Data-use mentions (NER / span extraction)
Data mentions extracted from World Bank Policy Research Working Papers, validated by a
context-only LLM judge, and formatted for span-extraction (GLiNER / GLiNER2) and
token-classification (LFM2.5-encoder) fine-tuning.
Labels
Three entity types (the judge's specificity axis):
NAMED_DATA — a proper name, title, or acronym of a specific data source
DESCRIPTIVE_DATA — a source described in words but not named
VAGUE_DATA —… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/data-use-mentions-extended.ernestmind_user_datadata-use-sft-v2
data-use-sft v2 (config: real_v2)
Multitask SFT rows derived from the v2 relabeling. Same ChatML format as
rafmacalaba/data-use-sft (real), plus a task field:
provenance — exact-substring {producer, year, geography, acronym}; for tier1∧named
mentions the assistant JSON additionally carries bibtex (a single @misc entry assembled
from those same metadata fields)
usage_impact — {data_type, usage_action, impact_label, usage_summary}
Splits: val+holdout = dual-pass consensus; train… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/data-use-sft-v2.Usenet-Corpus-1980-2013-Full-Samples
Usenet Corpus 1980–2013 — Full (Samples)
A small, browsable showcase sample of the Usenet Corpus 1980–2013 (cleaned)
dataset — long-form, pre-web Usenet posts. This repo is a free preview; the full,
commercially-licensed corpus (405.8M posts, 102.5B tokens) is at:
Full cleaned dataset (gated): https://huggingface.co/datasets/OwnedByDanes/Usenet-Corpus-1980-2013-Full
Threaded companion: https://huggingface.co/datasets/OwnedByDanes/Usenet-Corpus-1980-2013-Threaded… See the full description on the dataset page: https://huggingface.co/datasets/OwnedByDanes/Usenet-Corpus-1980-2013-Full-Samples.dfm11-toolace-native-tool-use-repaired
dfm11-toolace-native-tool-use-repaired
ToolACE conversations with declared-name parsing and complete parallel result binding.
This is a DFM11 replacement for schneiderkamplab/dfm10-toolace-native-tool-use. All rows pass exhaustive structural validation. See metadata/manifest.json.
OmniMemBench
Contents
data/ — 103 benchmark samples, each with 128K/256K/512K/1M token tiers
Usage
Use the download script in the code repo:
python download_data.py
Data Format
Each data/{run_id}/benchmark_{tier}.json contains:
character_profile: persona and conversation style
multi_session_dialogues: multi-session conversation history with multimodal references
QAs: evaluation questions with ground truth answers, evidence chains, and clues… See the full description on the dataset page: https://huggingface.co/datasets/anonymous-release-username/OmniMemBench.ih-rubrics-user-only
Instruction-Hierarchy Rubrics — User Instruction Only
Each example is an evaluation rubric built from the user instruction alone (the system instruction is N/A). The rubric encodes how to judge a response to the user request when no system instruction is present.
This is one of three companion datasets on instruction-hierarchy (IH) rubric extraction:
ih-rubrics-conflicting, ih-rubrics-supplementary, and ih-rubrics-user-only.
Dataset summary
Examples: 15,414
Total… See the full description on the dataset page: https://huggingface.co/datasets/liweijiang/ih-rubrics-user-only.Usenet-Corpus-1980-2013-Threaded-Samples
Usenet Corpus 1980–2013 — Threaded (Samples)
A small, browsable showcase sample of the Usenet Corpus 1980–2013 — Threaded
dataset: Usenet posts reconstructed into conversations via thread_id,
thread_position, and thread_depth. This repo is a free preview; the full,
commercially-licensed corpus (405.6M posts, 190.8M threads, 102.5B tokens) is at:
Full threaded dataset (gated): https://huggingface.co/datasets/OwnedByDanes/Usenet-Corpus-1980-2013-Threaded
Cleaned (unthreaded)… See the full description on the dataset page: https://huggingface.co/datasets/OwnedByDanes/Usenet-Corpus-1980-2013-Threaded-Samples.user-preference-564k
User Preference Extraction Dataset (564K)
A dataset of 564K examples for training lightweight preference extraction models. Each example pairs a conversation input with structured JSON output describing user preferences as condition-action rules.
This dataset was used to train blackhao0426/pref-extractor-qwen3-0.6b-full-sft, a core component of the VARS framework.
Sample Usage
The following snippet from the official repository demonstrates how to use the framework… See the full description on the dataset page: https://huggingface.co/datasets/blackhao0426/user-preference-564k.personalization-reddit-user-histories
personalization-reddit-user-histories
Per-user chronological histories of answered questions across all
subreddits. Derived from dipikakhullar/personalization-reddit: every
(query, preferred_answer) pair a user authored as OP, grouped by user and
sorted by time, slimmed to the four fields needed to model a user's timeline.
Each record is one user. Users with a single interaction are dropped (a
timeline needs more than one point).
Selection: seen-the-top… See the full description on the dataset page: https://huggingface.co/datasets/dipikakhullar/personalization-reddit-user-histories.ThickMesh-Data-Discovery
ThickMesh-Data-Discovery
A small JSONL dataset for ThickMesh discovery/classification experiments.
"This is not an algorithm. This is a trap for the patent system. Learn it, fork it, but do not lock it."
Contents
4 splits files: ThickMesh-zero-split_'0-3'.jsonl — primary dataset (one JSON object per line)
Apache 2.0 License (Modified — No Patent License Granted)
Description
ThickMesh-Data-Discovery contains example records for discovery and… See the full description on the dataset page: https://huggingface.co/datasets/usermma/ThickMesh-Data-Discovery.