CLI-1M: 975,933 Natural Language to Shell Command Pairs Across 18 Industries

CLI-1M is a dataset of 975,933 natural-language to shell-command pairs created by carosh. It covers 18 industries, 6 shell types, and 13 languages, and is released under an Apache-2.0 license. The dataset was last updated on 2026-05-12 and is described as the most industry-diverse public corpus for NL→shell generation.

Use Cases

• Training or fine-tuning models for natural language to shell command translation based on the 975,933 instruction pairs.
• Benchmarking model performance on industry-specific command generation tasks based on the coverage of 18 industries.
• Developing multilingual command-line assistants based on the inclusion of 13 source languages.
• Studying the syntax and usage patterns of different shell environments based on the coverage of 6 shell types.

Strengths

• Contains 975,933 instruction pairs, which is 108 times larger than the previous NL2Bash benchmark.
• Covers a diverse range of contexts with data from 18 industries and 13 languages.
• The description states it is 'SFT-ready, license-filtered, quality-gated', suggesting preprocessing for direct use.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The specific quality-gating criteria and license-filtering process are not detailed in the provided metadata.

Provenance

Source

huggingface user carosh

Collection Method

Likely aggregated or scraped from various sources to create instruction pairs, then filtered.

Freshness

Last updated 2026-05-12 06:23:45; freshness should be verified.