datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
1c-bsl-dataset
1C:Enterprise (BSL) dataset
Собрано для построения retrieval-инструмента (MCP-сервер), который даёт AI-агенту
знание платформы 1C:Enterprise (язык BSL): понимание чужого legacy-кода и
написание нового по образцу реальных примеров + официальным рекомендациям.
Два источника знаний:
Код — процедуры/функции, извлечённые из открытых репозиториев GitHub.
Доки — стандарты разработки its.1c.ru и статьи хаба "1C" на Habr, для
концептуальных вопросов "как это правильно делать".… See the full description on the dataset page: https://huggingface.co/datasets/pupkaZALUPKA/1c-bsl-dataset.month2-agent-runs
month2-agent-runs
20 prompts run via POST http://51.79.71.160/agent/run (VPS :8001 via Caddy, redis:7, DatasetTool BSLBSL/month1-spam-sample) with openai/gpt-oss-20b via Groq free (custom key) + heuristic 90-100% p50 27ms.
See https://huggingface.co/spaces/BSLBSL/ai-lab-m2-agent and https://github.com/BrunosGits/ai-lab/tree/main/month2
bslm-triplet-2.6M-qp-data
BSLM Triplet 2.6M Query-Positive-Hard-Negative Dataset
Query/product-title triplet dataset used to fine-tune mjaliz/bslm-triplet-2.6M-qp with Sentence Transformers TripletLoss after the pair-training stage.
Each row contains a query, a positive product title, and a lexical/confuser hard-negative product title from a different top-level domain/category.
Files
File
Rows
Purpose
train.jsonl
2,673,575
Triplet training rows
dev.jsonl
26,592
Triplet… See the full description on the dataset page: https://huggingface.co/datasets/mjaliz/bslm-triplet-2.6M-qp-data.bslm-pair-206k-data
BSLM Pair 206k Query-Title Dataset
Query/product-title pair dataset used to fine-tune mjaliz/bslm-pair-206k with Sentence Transformers MultipleNegativesRankingLoss.
The dataset was built from ecommerce query-to-anchor candidates by selecting one positive product title per query (top1) to reduce false negatives in in-batch negative training.
Files
File
Rows
Purpose
train.jsonl
202,149
Pair training rows
dev.jsonl
2,063
Pair validation rows
test.jsonl… See the full description on the dataset page: https://huggingface.co/datasets/mjaliz/bslm-pair-206k-data.