datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Athar-Embeddingsgemma-2b-dictionary-embeddings-all-layers
Gemma-2B Dictionary Embeddings - All Layers
This dataset contains pre-computed embeddings for 77,477 English words from WordNet using the Gemma-2B model across all 27 layers.
Dataset Structure
metadata.json: Contains dataset metadata (model info, dimensions, word count)
embeddings_layer_X.pkl: Pickle files containing embeddings for layer X (0-26)
Usage
import pickle
from huggingface_hub import hf_hub_download
# Download a specific layer
layer_0_path =… See the full description on the dataset page: https://huggingface.co/datasets/LeeHarrold/gemma-2b-dictionary-embeddings-all-layers.terminal-bench-4-qwen3-8b-embeddings-train20-seed42-successes
Terminal trajectory embedding task subset
20 selected tasks, 1,281 trajectories,
and 123,849 state/action pairs. Only trajectories with numeric reward == 1 are retained.
The saved vectors are exact selected rows of the existing embeddings; the encoder
was not rerun. train/metadata.json matches both tensor row orders.
source_row_indices.json records the original row indices; selection.json
records selection parameters, source checksums and output checksums.
selected_tasks.json… See the full description on the dataset page: https://huggingface.co/datasets/Hkang/terminal-bench-4-qwen3-8b-embeddings-train20-seed42-successes.airbnb_embeddings
Overview
This dataset consists of AirBnB listings with property descriptions, reviews, and other metadata.
It also contains text embeddings of the property descriptions as well as image embeddings of the listing image. The text embeddings were created using OpenAI's text-embedding-3-small model and the image embeddings using OpenAI's clip-vit-base-patch32 model available on Hugging Face.
The text embeddings have 1536 dimensions, while the image embeddings have 512 dimensions.… See the full description on the dataset page: https://huggingface.co/datasets/MongoDB/airbnb_embeddings.kilt-qwen-embeddingThe corpus refers to https://huggingface.co/datasets/corag/kilt-corpus
Rabbinic-Embedding-Leaderboard
Rabbinic Embedding Benchmark Leaderboard
This dataset stores the leaderboard results for the Rabbinic Hebrew/Aramaic Embedding Benchmark.
Structure
The leaderboard.json file contains an array of evaluation results:
[
{
"model_id": "model-org/model-name",
"model_name": "Model Display Name",
"mrr": 0.85,
"recall_at_1": 0.75,
"recall_at_5": 0.90,
"recall_at_10": 0.95,
"bitext_accuracy": 0.92,
"avg_true_pair_similarity": 0.85… See the full description on the dataset page: https://huggingface.co/datasets/Sefaria/Rabbinic-Embedding-Leaderboard.adp-v1-qwen3-8b-embeddings-32k
ADP Dataset V1 — Qwen3-8B state/action embeddings
All 1,270,306 trajectories and 2,592,191
typed ADP actions in the original V1 corpus (13 datasets / 18 subdirectories).
No sampling or duplication of alternate SFT formats; later corpus additions are
outside this V1 snapshot. Original source-specific licenses apply; no additional
rights to the underlying source data are granted.
Normalized, untruncated source and original file provenance:
Hkang/adp-v1-normalized-trajectories
at… See the full description on the dataset page: https://huggingface.co/datasets/Hkang/adp-v1-qwen3-8b-embeddings-32k.endless-terminals-qwen3-8b-embeddings-32k
Endless Terminals — Qwen3-8B state/action embeddings
All 44,128 released trajectories from
2,492 tasks, yielding 156,941 state/action pairs.
Includes o3 and Llama-3.2-3B-Instruct attempts, both successful and failed.
No new rollouts were generated. The upstream dataset card declares MIT; original
licenses remain applicable. The upstream generator code is Apache-2.0.
Normalized untruncated source with exact original file provenance:
Hkang/endless-terminals-normalized-trajectories… See the full description on the dataset page: https://huggingface.co/datasets/Hkang/endless-terminals-qwen3-8b-embeddings-32k.litecoder-terminal-sft-qwen3-8b-embeddings
LiteCoder Terminal SFT — Qwen3-8B state/action embeddings
Complete precomputation of 308,641 assistant steps from
11,255 trajectories in
Lite-Coder/LiteCoder-Terminal-SFT.
Source revision: 6acdbbdb29979e4b8ea717b12accc8214606d087. Source SHA-256: 50f58e33b4caaac73e6e12afe1a90206d18e4f47080da8b0b7d9a704a0c7c43f.
Encoder: Qwen/Qwen3-8B at b968826d9c46dd6066d109eabc6255188de91218.
Encoding contract
Each state contains only messages BEFORE its assistant action… See the full description on the dataset page: https://huggingface.co/datasets/Hkang/litecoder-terminal-sft-qwen3-8b-embeddings.d2f-turn-embeddings-soda
Turn embeddings for Soda (Dialog2Flow encoder)
One 768-d float16 vector per utterance of allenai/soda, computed with the frozen
encoder sergioburdisso/dialog2flow-joint-bert-base
(SentenceTransformer recipe, convert_to_numpy, no normalization). The encoder
truncates inputs at 64 tokens. If you use these embeddings, please cite the
Dialog2Flow paper (Burdisso et al., EMNLP 2024) and the source corpus.
Files
soda_e_t.f16.npy — numpy array (n_turns, 768), float16;… See the full description on the dataset page: https://huggingface.co/datasets/jumafernandez/d2f-turn-embeddings-soda.speaker-embeddingsasr-ser-quechua-collao-embeddings
ASR-SER embeddings for Quechua Collao
This repository contains embeddings only. It does not contain raw audio.
These embeddings were extracted for ASR-to-SER transfer experiments on the Quechua Collao emotional speech corpus. The release is intended for reproducible feature-extraction experiments and downstream analysis.
Dataset contents
One PyTorch tensor per utterance stored as an embedding file under embeddings/
A sanitized metadata table describing utterance… See the full description on the dataset page: https://huggingface.co/datasets/QuechuaBase/asr-ser-quechua-collao-embeddings.Embeddingsprimevul-codebert-embeddings
PrimeVul Embeddings for PU Learning
Pre-extracted [CLS] token embeddings from two code models for all functions in the PrimeVul v0.1 vulnerability detection dataset, plus the raw PrimeVul v0.1 JSONL source files.
CodeBERT Embeddings (root .npz files)
Each .npz file contains frozen CodeBERT embeddings (768-dimensional vectors) for C/C++ functions, along with their labels and CWE type annotations. These were extracted once using a frozen CodeBERT model and are used for… See the full description on the dataset page: https://huggingface.co/datasets/db-d2/primevul-codebert-embeddings.d2f-turn-embeddings-wildchat
Turn embeddings for Wildchat (Dialog2Flow encoder)
One 768-d float16 vector per utterance of allenai/WildChat-1M, computed with the frozen
encoder sergioburdisso/dialog2flow-joint-bert-base
(SentenceTransformer recipe, convert_to_numpy, no normalization). The encoder
truncates inputs at 64 tokens. If you use these embeddings, please cite the
Dialog2Flow paper (Burdisso et al., EMNLP 2024) and the source corpus.
Files
wildchat_e_t.f16.npy — numpy array (n_turns… See the full description on the dataset page: https://huggingface.co/datasets/jumafernandez/d2f-turn-embeddings-wildchat.d2f-turn-embeddings-taskmaster
Turn embeddings for Taskmaster (Dialog2Flow encoder)
One 768-d float16 vector per utterance of Taskmaster-1/2/3 (Google Research), computed with the frozen
encoder sergioburdisso/dialog2flow-joint-bert-base
(SentenceTransformer recipe, convert_to_numpy, no normalization). The encoder
truncates inputs at 64 tokens. If you use these embeddings, please cite the
Dialog2Flow paper (Burdisso et al., EMNLP 2024) and the source corpus.
Files
taskmaster_e_t.f16.npy —… See the full description on the dataset page: https://huggingface.co/datasets/jumafernandez/d2f-turn-embeddings-taskmaster.paragru-2-4-8m-embeddingsd2f-turn-embeddings-personachat
Turn embeddings for Personachat (Dialog2Flow encoder)
One 768-d float16 vector per utterance of Persona-Chat (Zhang et al., 2018), released within ParlAI, computed with the frozen
encoder sergioburdisso/dialog2flow-joint-bert-base
(SentenceTransformer recipe, convert_to_numpy, no normalization). The encoder
truncates inputs at 64 tokens. If you use these embeddings, please cite the
Dialog2Flow paper (Burdisso et al., EMNLP 2024) and the source corpus.
Files… See the full description on the dataset page: https://huggingface.co/datasets/jumafernandez/d2f-turn-embeddings-personachat.Gene-Embedding-Hub-contributions-stagingd2f-turn-embeddings-ultrachat
Turn embeddings for Ultrachat (Dialog2Flow encoder)
One 768-d float16 vector per utterance of openbmb/UltraChat, computed with the frozen
encoder sergioburdisso/dialog2flow-joint-bert-base
(SentenceTransformer recipe, convert_to_numpy, no normalization). The encoder
truncates inputs at 64 tokens. If you use these embeddings, please cite the
Dialog2Flow paper (Burdisso et al., EMNLP 2024) and the source corpus.
Files
ultrachat_e_t.f16.npy — numpy array (n_turns… See the full description on the dataset page: https://huggingface.co/datasets/jumafernandez/d2f-turn-embeddings-ultrachat.decidim-barcelona-proposals-embeddings-768d
Decidim Barcelona Proposal Topics 2016-2024
📊 Exploring the top 20 emerging topics from 31,775 citizen proposals in decidim.barcelona, with topic modelling (BERTopic) and deicdim-based open data.
31,775 proposal descriptions from decidim.barcelona (2016-2024), iterating through various parameters and data cleaning techniques, to extract 20 clearly recurrent topics emerging across 270 participatory processes.
Sentence embeddings generated using the HuggingFace sentence-transformers… See the full description on the dataset page: https://huggingface.co/datasets/darredondort/decidim-barcelona-proposals-embeddings-768d.sec-embeddings-sp500
SEC S&P 500 10q Embeddings Dataset
Overview
This dataset contains vector embeddings of SEC 10-Q filings for S&P 500 companies
Dataset Details
Total chunks: 36,927
Embedding model: BAAI/bge-large-en-v1.5
Embedding dimension: 1024
Processing date: 2025-07-15T19:26:08.380350
Quality Metrics
Average chunk length: 3182 characters
Financial relevance score: 0.146
arxiv-cs2021-embeddings-bge-m3arxiv-embeddingsanssi-bge-m3-embeddings
ANSSI cybersecurity guides — BGE-M3 embeddings
Semantic embeddings of 11 ANSSI (French national cybersecurity agency)
guides, produced by the open-source project
llm-verification-harness.
Project positioning. Transpose aerospace/defense IVVQ
(Integration, Verification, Validation, Qualification) practices to
non-deterministic RAG/LLM systems. The project's signature deliverable
is a Verification Control Document auto-generated per run (Brique 7).
This dataset is an intermediate… See the full description on the dataset page: https://huggingface.co/datasets/adriencr81/anssi-bge-m3-embeddings.quran_embeddings
Quran Embeddings Dataset
This repository contains vector embeddings for the Holy Quran, generated using OpenAI's embedding model. These embeddings can be used for semantic search, question answering, and other natural language processing tasks related to Quranic text.
Dataset Information
The dataset consists of a single JSON file:
quran_embeddings.json: Contains embeddings for each verse (ayah) of the Quran with associated metadata
Metadata Structure
Each… See the full description on the dataset page: https://huggingface.co/datasets/promehedi/quran_embeddings.airbnb_embeddings
Overview
This dataset consists of AirBnB listings with property descriptions, reviews, and other metadata.
It also contains text embeddings of the property descriptions as well as image embeddings of the listing image. The text embeddings were created using OpenAI's text-embedding-3-small model and the image embeddings using OpenAI's clip-vit-base-patch32 model available on Hugging Face.
The text embeddings have 1536 dimensions, while the image embeddings have 512 dimensions.… See the full description on the dataset page: https://huggingface.co/datasets/Maki-99/airbnb_embeddings.ESCO-FULL-SKILL-EMBEDDING-3072triplet-loss-for-embedding-affiliations-sample-1
Affiliation Triplet Curriculum Dataset
This dataset is designed for training an embedding model using triplet loss. It contains triplets of affiliation strings (anchor, positive, negative) structured to teach a model to recognize when two strings refer to the same institution.
The dataset is sorted from easiest to hardest to facilitate curriculum learning, allowing the model to learn from simple examples before progressing to more challenging ones.
Dataset Details… See the full description on the dataset page: https://huggingface.co/datasets/cometadata/triplet-loss-for-embedding-affiliations-sample-1.propertylm-uk-qdrant