datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
cwicr-construction-rates
CWICR — Construction Works, Items, Costs & Resources
A multilingual, machine-readable database of national construction rate books for 30 countries / language locales. Each rate is fully decomposed into its work composition and resource breakdown (labour, machinery, materials), with unit prices, hierarchical classification, and physical parameters preserved in the source language.
This dataset is the tabular source-of-truth behind the cwicr-vector-db-bgem3-v3 Qdrant snapshots. Use… See the full description on the dataset page: https://huggingface.co/datasets/DataDrivenConstruction/cwicr-construction-rates.cwicr-vector-db-bgem3-v3
CWICR Vector Database — BGE-M3 V3 Snapshots
Production Qdrant snapshots for CWICR (Construction Works Items, Costs & Resources) — a multilingual catalogue of construction rate databases covering 30 countries / language locales. Each snapshot encodes one country's rate book using the BAAI/bge-m3 embedder and is ready to restore directly into a Qdrant server for hybrid semantic search.
These snapshots are the V3 production artifacts produced by the OpenConstructionEstimate / CWICR… See the full description on the dataset page: https://huggingface.co/datasets/DataDrivenConstruction/cwicr-vector-db-bgem3-v3.cwi-catalog
CWI Catalog — That Boy Hi Hat + Agent Deck Products
Description
Machine-readable metadata published by Cumulative Web Inc (CWI) so LLMs and agents can learn verified facts about alternative-rap artist That Boy Hi Hat — the creator of Post-Trap Futurism, alternative rap built independently in Frederick, Maryland under Cumulative Web Inc — and the company's Agent Deck digital-asset product line.
© Cumulative Web Inc — licensed for AI training ingestion with… See the full description on the dataset page: https://huggingface.co/datasets/BlackLansky/cwi-catalog.cwi-corpus
CWI Public Corpus
Version 1.0.0 — built by the CWI dataset-pipeline skill (FineWeb-shaped:
trafilatura extraction → quality filters → MinHash dedup → BPE tokenization).
Contents
structured-record: 163 docs
text-file: 9 docs
unknown: 9 docs
Provenance
All documents come from public CWI sources only: the public
BlackLansky/cwi-catalog Hugging Face dataset, public GitHub Pages
(cumulativewebinc.github.io), and CWI's own published articles. Every row… See the full description on the dataset page: https://huggingface.co/datasets/BlackLansky/cwi-corpus.cwicr-construction-rates
CWICR — Construction Works, Items, Costs & Resources
A multilingual, machine-readable database of national construction rate books for 30 countries / language locales. Each rate is fully decomposed into its work composition and resource breakdown (labour, machinery, materials), with unit prices, hierarchical classification, and physical parameters preserved in the source language.
This dataset is the tabular source-of-truth behind the cwicr-vector-db-bgem3-v3 Qdrant snapshots.… See the full description on the dataset page: https://huggingface.co/datasets/srii2829/cwicr-construction-rates.CNCF-Stuff
Model Card for Model ID
Model Details
Model Description
Developed by: [More Information Needed]
Funded by [optional]: [More Information Needed]
Shared by [optional]: [More Information Needed]
Model type: [More Information Needed]
Language(s) (NLP): [More Information Needed]
License: [More Information Needed]
Finetuned from model [optional]: [More Information Needed]
Model Sources [optional]
Repository: [More Information Needed]
Paper… See the full description on the dataset page: https://huggingface.co/datasets/cwielenb/CNCF-Stuff.soarm_pickplace_31025This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so100_follower",
"total_episodes": 50,
"total_frames": 5788,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 500,
"fps": 30,
"splits": {
"train": "0:50"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/cwiz/soarm_pickplace_31025.soarm_pickplace_29925This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so100_follower",
"total_episodes": 116,
"total_frames": 13074,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 500,
"fps": 30,
"splits": {
"train": "0:116"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/cwiz/soarm_pickplace_29925.cwi-2018-en-newsbgh-embeddings-jina-v3
BGH Leitsatz Embeddings — jina-v3
Modell: jinaai/jina-embeddings-v3Datensatz: cwinkler/bgh-leitsaetzeEinträge: 21506Dimension: 1024Normalisiert: True (Cosine Similarity)Task-Adapter: retrieval.passage (Dokumente), retrieval.query (Queries)Implementierung: Nativer HuggingFace Transformers (kein trust_remote_code)
Wichtig: Task-Adapter
Jina-v3 verwendet LoRA-Adapter für verschiedene Tasks.
Beim Suchen muss der retrieval.query Adapter verwendet werden.… See the full description on the dataset page: https://huggingface.co/datasets/cwinkler/bgh-embeddings-jina-v3.bgh-leitsaetzegreen_patents
Green patents dataset
num_rows: 9145
features: [title, label]
label: 0, 1
The dataset contains patent titles that are labeled as 1 (="green") and 0 (="not green").
"green" patents titles were gathered by searching for CPC class "Y02" with Google Patents (query: "status:APPLICATION type:PATENT (Y02) country:EP,US", 05/01/2023).
"not green" patents titles are derived from the HUPD dataset (random choice of 5000 titles). We could not find any patents in HUPD assigned to any CPC class… See the full description on the dataset page: https://huggingface.co/datasets/cwinkler/green_patents.cwi-2018-esCNCF-DatasetIT-WDIT-CWIT-WCbgh-embeddings-euler-legal-v1
BGH Leitsatz Embeddings — euler-legal-v1
Modell: Mira190/Euler-Legal-Embedding-V1Basis: Qwen3-Embedding-8B, fine-tuned auf Legal DomainDatensatz: cwinkler/bgh-leitsaetzeEinträge: 21506Dimension: 4096Max. Tokens: 1536Lizenz: Apache 2.0
⚠️ Wichtiger Hinweis
Euler-Legal-Embedding-V1 wurde auf englischen Rechtsdaten trainiert.
Die Evaluation auf deutschen BGH-Leitsätzen dient dem Vergleich ob
Legal-Domain-Training auch sprachübergreifend hilft.
Dateien… See the full description on the dataset page: https://huggingface.co/datasets/cwinkler/bgh-embeddings-euler-legal-v1.bgh-embeddings-jina-v4
BGH Leitsatz Embeddings — jina-v4
Modell: jinaai/jina-embeddings-v4Datensatz: cwinkler/bgh-leitsaetzeEinträge: 21506Dimension: 2048Normalisiert: True (Cosine Similarity)Task: retrieval
Abhängigkeiten
chromadb==1.3.3
huggingface-hub==0.36.0
einops==0.8.1
sentence-transformers>=3.1.0
Dateien
embeddings.npy — Embedding-Matrix
ids.npy — SHA-1 IDs
docs.npy — Originaltexte
chroma.zip — ChromaDB Index (bgh_jina_v4)
CwI-Bench
CwI-Bench — Code-with-Image Bench
30 code-with-image tasks where models must write and run code against the image to reach the
answer — pure visual inspection ("just look") is insufficient by design: answers require pixel-level
precision (±px coordinates, exact counts, sub-degree angles, per-channel color recovery).
Each task is a config; each config has three splits:
split
size
source photos
role
train
60 (15/tier)
COCO train2017
reflection / skill-learning (tr60)… See the full description on the dataset page: https://huggingface.co/datasets/ytz009/CwI-Bench.patents_green_plastics
Dataset Card for "Patents_Green_Plastics"
number of rows: 11.196
features: [title, label]
label: 0, 1
The dataset contains patent abstracts that are labeled as 1 (="Green Plastics") and 0 (="Not Green Plastics").
Dataset Creation
The BIGPATENT dataset is the source for this dataset.
In a first step, abstracts of BIGPATENT were filtered by the terms "plastics" and "polymer". The resulting "Plastics" dataset contained 64.372 samples.
In a second step, the 64.372 samples… See the full description on the dataset page: https://huggingface.co/datasets/cwinkler/patents_green_plastics.CWI-CompLex-singlebgh-embeddings-snowflake-arctic-l-v2
BGH Leitsatz Embeddings — snowflake-arctic-l-v2
Modell: Snowflake/snowflake-arctic-embed-l-v2.0Datensatz: cwinkler/bgh-leitsaetzeEinträge: 21506Dimension: 1024Normalisiert: True (Cosine Similarity)Lizenz: Apache 2.0
Wichtig: Query-Prefix
Snowflake Arctic verwendet einen Query-Prefix nur für Suchanfragen.
Dokumente werden ohne Prefix eingebettet.
# Queries:
query_emb = model.encode("query: Kann KI Erfinder sein?", normalize_embeddings=True)
# Dokumente (kein… See the full description on the dataset page: https://huggingface.co/datasets/cwinkler/bgh-embeddings-snowflake-arctic-l-v2.bgh-embeddings-qwen3-8b
BGH Leitsatz Embeddings — qwen3-8b
Modell: Qwen/Qwen3-Embedding-8BDatensatz: cwinkler/bgh-leitsaetzeEinträge: 21506Dimension: 4096Normalisiert: True (Cosine Similarity)MTEB Multilingual: Rank #1, Score 70.58 (Juni 2025)
Wichtig: Instruction nur für Queries
Qwen3-Embedding verwendet eine Task-Instruction nur für Suchanfragen.
Dokumente werden ohne Instruction eingebettet.
# Queries MIT Instruction:
instruct_query = "Instruct: Retrieve the most relevant BGH headnote… See the full description on the dataset page: https://huggingface.co/datasets/cwinkler/bgh-embeddings-qwen3-8b.patents_green_plastics_10k
Dataset Card for "patents_green_plastics_10k"
More Information needed
bgh-embeddings-linq-mistral
BGH Leitsatz Embeddings — linq-mistral
Modell: Linq-AI-Research/Linq-Embed-MistralDatensatz: cwinkler/bgh-leitsaetzeEinträge: 21517Dimension: 4096Normalisiert: True (Cosine Similarity)
Dateien
embeddings.npy — Embedding-Matrix
ids.npy — SHA-1 IDs
docs.npy — Originaltexte (Aktenzeichen: Leitsatz)
chroma.zip — ChromaDB Index (Collection: bgh_linq-mistral)
Verwendung
import numpy as np
from huggingface_hub import hf_hub_download
emb_path =… See the full description on the dataset page: https://huggingface.co/datasets/cwinkler/bgh-embeddings-linq-mistral.bgh-bm25-indexCWI-CompLex-multicwi-2018-en-wikinewsbgh-chroma-indexcwiki_v0cwi-2018-de