This model is designed for intent classification tasks and has been converted to ONNX format for efficient deployment in various environments, including web browsers using Transformers.js.

Model Variants

This repository contains multiple ONNX model variants optimized for different use cases:

Model File	Description	Use Case
`model.onnx`	Original ONNX model	Best accuracy, larger size
`model_fp16.onnx`	16-bit floating point	Good balance of accuracy and speed
`model_int8.onnx`	8-bit integer quantized	Faster inference, smaller size
`model_q4.onnx`	4-bit quantized	Very fast, very small
`model_q4f16.onnx`	4-bit with FP16	Optimized for specific hardware
`model_quantized.onnx`	Standard quantized	General purpose optimization
`model_uint8.onnx`	Unsigned 8-bit	Mobile/edge deployment
`model_bnb4.onnx`	BitsAndBytes 4-bit	Advanced quantization

Quick Start

Using with Transformers.js (Browser)

javascript

import { pipeline } from '@xenova/transformers';

// Load the model
const classifier = await pipeline('text-classification', 'kousik-2310/intent-classifier-minilm');

// Classify text
const result = await classifier('I want to book a flight to New York');
console.log(result);

Using with Python/Transformers

python

from transformers import AutoTokenizer, AutoModelForSequenceClassification
from transformers import pipeline

# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("kousik-2310/intent-classifier-minilm")
model = AutoModelForSequenceClassification.from_pretrained("kousik-2310/intent-classifier-minilm")

# Create pipeline
classifier = pipeline("text-classification", model=model, tokenizer=tokenizer)

# Classify text
result = classifier("I want to book a flight to New York")
print(result)

Using ONNX Runtime

python

import onnxruntime as ort
from transformers import AutoTokenizer

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("kousik-2310/intent-classifier-minilm")

# Load ONNX model
session = ort.InferenceSession("onnx/model_int8.onnx")

# Tokenize input
text = "I want to book a flight to New York"
inputs = tokenizer(text, return_tensors="np", padding=True, truncation=True)

# Run inference
outputs = session.run(None, {
    "input_ids": inputs["input_ids"],
    "attention_mask": inputs["attention_mask"]
})

# Process results
predictions = outputs[0]

Model Architecture

Base Model: MiniLM architecture
Task: Text Classification (Intent Recognition)
Framework: PyTorch → ONNX
Quantization: Multiple variants available

Performance

The model provides different performance characteristics based on the variant used:

Accuracy: Best with model.onnx, good with quantized versions
Speed: Fastest with model_q4.onnx and model_int8.onnx
Size: Smallest with quantized variants (4-bit, 8-bit)

Intended Use

This model is intended for:

Intent classification in chatbots and virtual assistants
Text classification tasks
Real-time inference in web applications
Edge deployment scenarios

Training Details

The model has been fine-tuned for intent classification and converted to multiple ONNX formats for optimal deployment flexibility.

Limitations and Bias

The model performance depends on the similarity between your use case and the training data
Quantized models may have slightly reduced accuracy compared to the full precision model
Performance may vary based on the deployment environment

How to Cite

bibtex

@misc{intent-classifier-minilm,
  title={Intent Classifier MiniLM},
  author={kousik-2310},
  year={2024},
  url={https://huggingface.co/kousik-2310/intent-classifier-minilm}
}

License

This model is released under the Apache 2.0 License.