Dl26/Veyra-20M

Veyra-20M: Compact Many-Label Intent Classification

Veyra-20M is a small, encoder-only text classification model built by Dl26. It is designed for fast English banking-intent and topic-style classification using a BERT-style bidirectional Transformer encoder.

The released checkpoint, Dl26/Veyra-20M, is a 20M-parameter-class classifier trained from random initialization. It uses the standard Transformers bert model type, so it loads with AutoModelForSequenceClassification and does not require trust_remote_code=True.

Why this model

• —Compact encoder-only classifier
• —Many-label classification setup with 77 labels
• —Standard Hugging Face Transformers compatibility
• —No custom architecture Python files
• —Fast batch inference on CPU or GPU
• —Built for intent routing, query classification, triage, and lightweight encoder research

Model details

Supported labels

Installation

pip install -U transformers torch accelerate

Quick start

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_id = "Dl26/Veyra-20M"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)
model.eval()

text = "Can you help me reset my password?"
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=64)

with torch.no_grad():
    logits = model(**inputs).logits

label_id = int(logits.argmax(dim=-1))
print(model.config.id2label[label_id])

Batch inference

texts = [
    "Book me a flight to Zurich tomorrow morning.",
    "What is the weather like today?",
    "Please cancel my card.",
]

inputs = tokenizer(
    texts,
    return_tensors="pt",
    truncation=True,
    padding=True,
    max_length=64,
)

with torch.no_grad():
    logits = model(**inputs).logits

for text, label_id in zip(texts, logits.argmax(dim=-1).tolist()):
    print(model.config.id2label[label_id], "-", text)

Evaluation highlights

Intended use

Veyra-20M is intended for:

• —intent classification
• —query routing
• —customer-support triage
• —lightweight text classification
• —many-class encoder experiments
• —CPU-friendly classifier deployments

Limitations

• —The model is specialized for the supported intent labels.
• —It is trained for English short utterances.
• —It may fail on long documents or out-of-domain language.
• —Confidence scores should be calibrated for production systems.
• —It is not a generative model or a semantic embedding model.

Citation

@misc{dl26_2026_veyra_20m,
  title        = {Veyra-20M: Compact Many-Label Intent Classification},
  author       = {Dl26},
  year         = {2026},
  url          = {https://huggingface.co/Dl26/Veyra-20M}
}