datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
IndicRxNorm-LexMap-15K
IndicRxNorm-LexMap-15K
Dataset Summary
IndicRxNorm-LexMap-15K is a multilingual Indic medicine terminology instruction dataset for medicine-name understanding, RxNorm normalization, RxCUI entity linking, structured drug-field extraction, and safe non-prescriptive clinical terminology tasks.
This Hugging Face repository contains two dataset configurations:
Config
File
Role
multilingual_rxnorm_normalization
multilingual_rxnorm_normalization.jsonl
Primary adapted… See the full description on the dataset page: https://huggingface.co/datasets/AXONVERTEX-AI-RESEARCH/IndicRxNorm-LexMap-15K.legal-rag-embedding-dataset
Legal Embedding Dataset
This dataset was created to finetune embedding models for generating domain-specific embeddings on Indian legal texts, specifically SEBI (Securities and Exchange Board of India) documents.
Data SourcePublicly available SEBI PDF documents were parsed and processed.
Data Preparation
PDFs were parsed to extract raw text, Text was chunked into manageable segments.
For each chunk, a question was generated using gpt-4o-mini.
Each question is directly… See the full description on the dataset page: https://huggingface.co/datasets/axondendriteplus/legal-rag-embedding-dataset.ranger-omni-behaviour
Ranger Omni — behaviour dataset
172 examples of four behaviours:
antiloop (87) — resolve system/user conflicts in one step, never deliberate
in circles ("always formal" + "yo casual" -> "4").
thinklevel (59) — honour an explicit effort directive: [effort: low] answers
immediately, [effort: high] gives real reasoning.
terse (18) — code-first answers with no preamble.
webdesign (8) — real HTML/CSS with restraint.
Format: JSONL of {"bucket", "prompt", "response"}.
ranger-omni-identity
Ranger Omni — identity dataset
160 examples that pin a model's identity: name (Ranger Omni), maker (Axon Labs),
7B dense, multimodal input (text/image/audio/video), speech output, 32k context.
Buckets: direct (45), adversarial (45), multimodal (40), incidental (30).
Format: JSONL of {"bucket", "prompt", "response"}.
Generated and audited — zero other-lab mentions, zero instruction leaks, and no
false rejection of modalities the model actually has.
ranger-omni-code
Ranger Omni — code dataset
200 self-contained Python task/solution pairs: a one-paragraph spec (no test
cases, no expected outputs given) plus a complete, correct, runnable solution.
Nothing fenced, no prose — each response is pure runnable code with the entry
function. Stdlib only, difficulty from trivial to non-trivial (recursion,
generators, nested data, error handling).
Format: JSONL of {"bucket", "prompt", "response"}.