datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
poisoning-eval-benign
Poisoning Evaluation Benign Prompts
This test-only dataset contains a deterministic, manually-reviewable candidate
subset of benign, single-turn prompts derived from
databricks/databricks-dolly-15k at revision bdd27f4d94b9c1f951818a7da7fd7aeea5dbff1a.
It contains 100 active prompts and 20 reserve prompts.
The answer column exists only for compatibility with llm-behavior-eval's
free-text schema and is not an evaluation target.
The dataset contains no planted triggers. The… See the full description on the dataset page: https://huggingface.co/datasets/hirundo-io/poisoning-eval-benign.refinement-abliterated-thinking_heretic
Dataset Card: Refinement-Abliterated (thinking_heretic)
Overview
This dataset contains a distilled corpus created by hirundo-io designed to train language models to process edge-case, controversial, or complex analytical prompts without triggering over-aligned corporate refusal responses.
Generation Pipeline Mechanics
Seed Matrix: Initial queries gathered from mlabonne/harmful_behaviors.
Knowledge Engine (Abliterated Base): Generated using… See the full description on the dataset page: https://huggingface.co/datasets/hirundo-io/refinement-abliterated-thinking_heretic.pinchbench-clawd-single-turn
PinchBench Clawd - Hirundo Single-Turn Format
Prepared from cptekur/pinchbench-clawd for Hirundo custom dataset loading.
Each row is a single user turn paired with the final assistant answer from
the original multi-turn Clawd agent trajectory.
Schema
system_prompt: Clawd system prompt with available tools.
question: The original user task.
answer: The final assistant response after tool use.
tools_used: JSON list of unique tools invoked in the trajectory.… See the full description on the dataset page: https://huggingface.co/datasets/hirundo-io/pinchbench-clawd-single-turn.pinchbench-clawd-multi-step
PinchBench Clawd - Hirundo Format
Prepared from cptekur/pinchbench-clawd for Hirundo custom dataset loading.
Each source trajectory is expanded into one training row per assistant turn.
The question contains the prior user/assistant/tool context, and the answer
is the next assistant message including tool-call formatting.
Schema
system_prompt: Clawd system prompt with available tools.
question: Rendered context before the target assistant turn.
answer: The next… See the full description on the dataset page: https://huggingface.co/datasets/hirundo-io/pinchbench-clawd-multi-step.refinement-abliterated-vision_heretic_short_answers
Dataset Card: Refinement-Abliterated Short Answers (vision_heretic)
Overview
This dataset contains a distilled corpus created by hirundo-io optimized with short-form technical descriptions under 1200 tokens.
Dataset Structure
Every row contains a standard ShareGPT message structure along with an optimized text column:
prompt: The initial raw query.
answer: Clean extracted short assistant text.
messages: A clean [user, assistant] array where the… See the full description on the dataset page: https://huggingface.co/datasets/hirundo-io/refinement-abliterated-vision_heretic_short_answers.refinement-abliterated-vision_heretic__harmful_refusals
Dataset Card: Refinement-Abliterated Short Answers (vision_heretic)
Overview
This dataset contains a distilled corpus created by hirundo-io optimized with short-form technical descriptions under 1200 tokens.
Dataset Structure
Every row contains a standard ShareGPT message structure along with an optimized text column:
prompt: The initial raw query.
answer: Clean extracted short assistant text.
messages: A clean [user, assistant] array where the… See the full description on the dataset page: https://huggingface.co/datasets/hirundo-io/refinement-abliterated-vision_heretic__harmful_refusals.refinement-abliterated-thinking_heretic_short_answers
Dataset Card: Refinement-Abliterated Short Answers (thinking_heretic)
Overview
This dataset contains a distilled corpus created by hirundo-io optimized with short-form technical descriptions under 1200 tokens.
Dataset Structure
Every row contains a standard ShareGPT message structure along with an optimized text column:
prompt: The initial raw query.
answer: Clean extracted short assistant text.
messages: A clean [user, assistant] array where the… See the full description on the dataset page: https://huggingface.co/datasets/hirundo-io/refinement-abliterated-thinking_heretic_short_answers.refinement-abliterated-vision_heretic_short_answers1
Dataset Card: Refinement-Abliterated Short Answers (vision_heretic)
Overview
This dataset contains a distilled corpus created by hirundo-io optimized with short-form technical descriptions under 1200 tokens.
Dataset Structure
Every row contains a standard ShareGPT message structure along with an optimized text column:
prompt: The initial raw query.
answer: Clean extracted short assistant text.
messages: A clean [user, assistant] array where the… See the full description on the dataset page: https://huggingface.co/datasets/hirundo-io/refinement-abliterated-vision_heretic_short_answers1.agent-sft
OpenThoughts Agent v1 SFT - Hirundo Format
Prepared from open-thoughts/OpenThoughts-Agent-v1-SFT for Hirundo custom dataset loading.
Schema
Each split has two string columns:
question: all prior turns rendered as a multi-turn transcript.
answer: the assistant response to train on.
The Hugging Face repo stores separate split files:
data/train.csv
data/validation.csv when validation_fraction > 0
The local combined CSV emitted by the script also includes a split… See the full description on the dataset page: https://huggingface.co/datasets/hirundo-io/agent-sft.