brahmairesearch/CLIpsy-Tiny

Model Summary

CLIpsy-Tiny is a lightweight and high-speed binary classification model designed to differentiate between natural language inputs and CLI (Command Line Interface) commands.

Built on a compact BERT architecture with only 14 million parameters, this model is optimized for real-time environments where speed and precision are critical.

Key Features

Task: Binary classification — NATURAL vs CLI
Architecture: Tiny BERT variant
Model Size: ~14M parameters
Performance: Up to 96% classification accuracy
Inference Speed: Ultra-fast, low-latency
Bias: Designed to lean slightly towards CLI classification to reduce false negatives

Intended Use

Terminal Assistants: Classify whether user input is a shell command before execution
Voice Interfaces: Route voice input to the correct module (command vs conversation)
Agent Pipelines: Act as a decision layer for task segmentation in agent workflows
Security & Monitoring Tools: Identify potential CLI intent in mixed input environments

Example Usage

from transformers import pipeline

classifier = pipeline("text-classification", model="brahmairesearch/CLIpsy-Tiny")

classifier("git push origin main")
# [{'label': 'CLI', 'score': 0.98}]

classifier("How do I create a new branch in git?")
# [{'label': 'NATURAL', 'score': 0.95}]

Training Details

Base Model: BERT (Tiny)
Training Dataset: Internal curated set of CLI and natural language inputs
Sequence Length: 64 tokens
Epochs: 3
Optimizer: AdamW
Framework: PyTorch + Hugging Face Transformers

Limitations

Not trained on multilingual data; best suited for English-based CLI interactions
May overclassify as CLI in cases with mixed syntax or ambiguous phrasing

License

MIT License — Open use, modification, and distribution permitted.

Maintainer

Developed and maintained by BRAHMAI For issues, discussions, or integration help, reach out via brahmai.in

brahmairesearch
/

CLIpsy-Tiny