datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
bash-commands-dataset
🐧 Linux Command Automation Dataset
A dataset of natural language prompts paired with their corresponding Bash command-line equivalents, designed to train or fine-tune models for automating Linux tasks via natural language.
📁 Dataset Structure
The dataset is in JSON format, structured as a flat array of objects, where each object contains:
{
"prompt": "Natural language description of a task",
"response": "Equivalent Bash command"
}
✅ Example
{… See the full description on the dataset page: https://huggingface.co/datasets/aelhalili/bash-commands-dataset.bash-instruct-III-55k
Bash Instruct III — 54,360 verified natural-language → Bash pairs
Bash Instruct III is a synthetic instruction-tuning dataset that maps natural-language
requests to correct Bash: single commands, short pipelines, and multi-line scripts. It
is built for supervised fine-tuning of small and mid-size LLMs that must turn a plain
request into shell code that actually runs.
Every row is a three-turn chat conversation (system / user / assistant) with metadata
for slicing (category… See the full description on the dataset page: https://huggingface.co/datasets/Frost2o24/bash-instruct-III-55k.trilingual-parallel-phrasebooks-bgpu
Bashkir Trilingual Parallel Phrasebooks
9,857 phrases aligned across three languages — Bashkir, Russian and one of Altai, Arabic, Kazakh, Yakut (Sakha), Chinese — from five phrasebooks published by M. Akmulla Bashkir State Pedagogical University. One row is one phrase in all three languages: a parallel corpus for machine translation and cross-lingual work with a low-resource Turkic language. Each phrasebook is a separate file and a separate config, because the third language… See the full description on the dataset page: https://huggingface.co/datasets/bashkorttele/trilingual-parallel-phrasebooks-bgpu.SensorCaps
SensorCaps
SensorCaps is an LLM-assisted softly-labelled IMU sensor data captioning dataset with feature summarizations and narrations of human activities.
Abstract
Wearable systems can recognize activities from IMU data but often fail to explain their underlying causes or contextual significance. To address this limitation, we introduce two large-scale resources: SensorCap, comprising 35,960 IMU--caption pairs, and OpenSQA, with 199,701 question--answer pairs designed… See the full description on the dataset page: https://huggingface.co/datasets/BASH-Lab/SensorCaps.bash-instruct-II-55k
Bash Instruct II — 54,803 verified natural-language → Bash pairs
⚠️ Superseded by Bash Instruct III
III is a corrected rebuild of this dataset with shell antipatterns removed at the
generator level (99.8% ShellCheck-clean vs 98.5% here). New work should use III.
The two share 92.5% of their (request, command) pairs, so they must never be
concatenated. This card is kept for reproducibility and citation of published results.
Bash Instruct II is a synthetic… See the full description on the dataset page: https://huggingface.co/datasets/Frost2o24/bash-instruct-II-55k.shellminator-bash-sft105kLLM_Bash
Description of LLM Bash
Prompt designed to convert natural language to bash command.
Inputs
This is a description of the inputs that the prompt expects.
question: User question to be answered by writing a bash command.
Usage
Below is a code snippet for how to use the prompt.
from langchain.prompts import load_prompt
from langchain.chains import LLMBashChain
llm = ...
prompt = load_prompt('lc://prompts/llm_bash/<file-name>')
chain = LLMBashChain(llm=llm… See the full description on the dataset page: https://huggingface.co/datasets/LangChainHub-Prompts/LLM_Bash.bash-instruct-55k
Bash Instruct I — 55,000 verified natural-language → Bash pairs
⚠️ Superseded by Bash Instruct III
III doubles the utility vocabulary (89 → 182), adds grouped equivalent answers, real
human phrasing from tldr-pages, validation by execution on real Linux, and a style pass
that removes shell antipatterns. New work should use III.
This version remains useful for one specific purpose: it overlaps III by only ~29%, so
it is the one generation that can be deduplicated… See the full description on the dataset page: https://huggingface.co/datasets/Frost2o24/bash-instruct-55k.shellminator-bash-combinedcommand-linux-bash-balanced-sftshellminator-bash-sft100kshellminator-bash-datasetbash-agent-grpo-pairs
Bash Agent GRPO Pairs
Single-turn (intent → shell command) pairs for training a small, local, Claude-Code-style
bash agent with SFT or GRPO. Each record pairs a natural-language objective with exactly
one verifiable bash command, framed as a single-tool bash(command, description) call.
The dataset is designed to be rewardable: the ground-truth command is a deterministic target,
so a shell-equivalence reward (canonical program + flag set + argument comparison) can score… See the full description on the dataset page: https://huggingface.co/datasets/adeelahmad/bash-agent-grpo-pairs.BashCoder
🌀 Claude-opus-4.7-TraceInversion-5000x
v1.0 Release
A High-Fidelity Reconstructed CoT Dataset Saturated with the 'Opus Deep Logic Style' via Trace Inversion
📊 5,000 Samples
🧬 Trace Inversion & Negentropy
🛠 SFT & DPO Ready
🔥 Claude 4.7-Max Distillation
🌐 English & Multilingual
💡 What is Trace Inversion?
In Large Language Model (LLM) reasoning distillation, proprietary API models (such as GPT-4/5 and Claude)… See the full description on the dataset page: https://huggingface.co/datasets/Coding-With-Bashir/BashCoder.bash
Bash & Linux CLI Dataset
What This Dataset Is
A supervised fine-tuning (SFT) dataset that teaches a language model to produce correct Bash commands, shell scripts, and Linux CLI operations — including Arch Linux and Ubuntu/Debian specifics. The assistant responses are executable commands or scripts that directly accomplish the user's stated task. Explanatory rows describe what a given command does, its key flags, and caveats.
Sources (provenance-tracked)… See the full description on the dataset page: https://huggingface.co/datasets/Euroswarms/bash.rlvr-bash-terminal-bench
rlvr-bash-terminal-bench
RLVR (Reinforcement Learning with Verifiable Rewards) dataset for bash scripting, generated from Terminal-Bench tasks.
Stats
Metric
Value
Total samples
1,120
Unique tasks
88
Avg samples/task
12.7
Average reward
0.249
Perfect solutions (reward=1.0)
10.4%
Partial solutions (0<reward<1)
28.8%
Zero reward
60.8%
Tasks fully solved
13.6%
Format
{
"task_id": "string",
"prompt": "string",
"completion":… See the full description on the dataset page: https://huggingface.co/datasets/lisayan/rlvr-bash-terminal-bench.bashkir-russian-parallel
Dataset Card for Bashkir-Russian Parallel Corpus
Dataset Details
Dataset Description
Bashkir-Russian Parallel Corpus is a large-scale sentence-aligned parallel corpus for the Bashkir–Russian language pair, assembled from authentic human-created translations. It contains 3,040,085 unique parallel sentence pairs, where each Bashkir sentence is aligned with its Russian counterpart.
The corpus combines data from three open parallel corpora: TIL-MT… See the full description on the dataset page: https://huggingface.co/datasets/BashkirNLPWorld/bashkir-russian-parallel.Medical_v1abashkir-lexicon
Dataset Card for Bashkir Lexicon (Machine Fund)
Dataset Details
Dataset Description
Bashkir Lexicon (Machine Fund) is a machine-readable lexical database of the Bashkir language, compiled from the open digital resource Machine Fund of the Bashkir Language (mfbl2.ru). It contains 34,008 unique lexical entries covering dialectal word forms with their part of speech, dialect, subdialect, literary norm, and Russian translation.
The dataset preserves… See the full description on the dataset page: https://huggingface.co/datasets/BashkirNLPWorld/bashkir-lexicon.bash-commands-dataset
🐧 Linux Command Automation Dataset
A dataset of natural language prompts paired with their corresponding Bash command-line equivalents, designed to train or fine-tune models for automating Linux tasks via natural language.
📁 Dataset Structure
The dataset is in JSON format, structured as a flat array of objects, where each object contains:
{
"prompt": "Natural language description of a task",
"response": "Equivalent Bash command"
}
✅ Example… See the full description on the dataset page: https://huggingface.co/datasets/dravidmathavan07/bash-commands-dataset.bashorgbash-toolcallsshellminator-bash-cleandataset_zabih_3k_samplestest_bashlego_grey_test1my-sft-datasetaeon-ouroboros_recursive-survivability-tracesÆon-Ouroboros: Recursive Survivability Traces
Summary
This dataset contains non-dialogic survivability cycles documenting how a system responds to perturbation under explicit constraints.
The focus is recovery, coherence, and ethical load management, not output quality or task performance.
These records are intended to support research into recursive systems, constraint-first intelligence, and survivability under stress.
What This Dataset Is
A collection of cycle-level traces, not samples… See the full description on the dataset page: https://huggingface.co/datasets/basheuvel/aeon-ouroboros_recursive-survivability-traces.Arabic_MCQ_Baseline_120BArabic_QWEN3_1_7B_Basee