Intento-v1

Intento-v1-edge is the ultra-compressed, production-ready companion to the Intento-v1 family. It was trained on the SID dataset and is designed specifically for low-latency 5-language intent classification on mobile and IoT devices.

Performance & Compression

This model achieves extreme compression compared to the standard transformer baseline while maintaining high accuracy for production use.

Comparison Benchmark

Metric	SOTA Transformer (3.3GB)	Edge AI Model (500kB)	Delta
Accuracy	93.98%	93.91%	Parity (-0.07%)
Weighted F1	0.9205	0.9206	+0.0001
Macro F1	0.9670	0.9688	+0.0018

Files

model.safetensors: Optimized quantized weights for high-speed edge inference.
word_vocab.json / char_vocab.json: Required for tokenizer preprocessing.

License

This model is licensed under the MIT License.

Downloads last month: -; Downloads are not tracked for this model. How to track

Safetensors

Model size

504k params

Tensor type

F32

I32

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train luigicfilho/intento-v1-edge

Collection including luigicfilho/intento-v1-edge

Intento-V1

Collection

Family of models, for intent classification, based on the article: https://www.linkedin.com/pulse/ai-journey-efficiency-frontier-achieving-sota-parity • 2 items • Updated 4 days ago • 1

Evaluation results

Validation Accuracy
self-reported

0.939