README.md

---
library_name: transformers
tags:
- text-classification
- zero-shot-classification
- echo-dsrn
base_model:
- ethicalabs/Echo-DSRN-114M-v0.1.2
new_version: ethicalabs/Echo-DSRN-v0.1.3-Intent-CLF
---

# Echo-SmolTools-114M-Intent-CLF-Gen

[![GitHub](https://img.shields.io/badge/GitHub-ethicalabs.ai-black.svg)](https://github.com/ethicalabs-ai/Echo-DSRN/)
[![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
[![Python](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)
[![Model Collection](https://img.shields.io/badge/Echo--DSRN-HuggingFace-yellow.svg)](https://huggingface.co/collections/ethicalabs/echo-dsrn)
[![Hybrid Collection](https://img.shields.io/badge/Echo--Hybrid-HuggingFace-red.svg)](https://huggingface.co/collections/ethicalabs/echo-dsrn-hybrid)
[![Working Paper](https://img.shields.io/badge/Working--Paper-Echo_DSRN-green.svg)](https://github.com/ethicalabs-ai/Echo-DSRN/blob/main/PAPER.md)

> [!WARNING]
> This repository contains experimental models designed strictly for academic evaluation and research purposes.
> 
> Critical Constraints:
> * **No Production Deployment:** Experimental models must not be deployed in commercial, enterprise, or mission-critical environments under any circumstances. 
> * **No Liability:** Experimental models are provided "as-is" without warranties of any kind. The developers assume zero liability for downstream consequences, system integration failures, or regulatory non-compliance resulting from unauthorized deployment.

This is a **generative** sequence classification model based on the **Echo-DSRN** architecture.
It was merged from the base model [`ethicalabs/Echo-DSRN-114M-v0.1.2`](https://huggingface.co/ethicalabs/Echo-DSRN-114M-v0.1.2)
and the PEFT adapter [`ethicalabs/Echo-SmolTools-114M-Intent-PEFT`](https://huggingface.co/ethicalabs/Echo-SmolTools-114M-Intent-PEFT).

No additional linear head is trained — the adapter's generative knowledge is used directly via
**constrained next-token scoring**: for each candidate label the model sums the log-probability
of each of its tokens, then picks the highest-scoring one.

## Model Details

- **Architecture:** `EchoForGenerativeClassification`
- **Base model:** `ethicalabs/Echo-DSRN-114M-v0.1.2`
- **Adapter:** `ethicalabs/Echo-SmolTools-114M-Intent-PEFT`
- **Labels:** 60 Amazon MASSIVE intents (51 languages)
- **Dtype:** `bfloat16`
- **Constraint Method:** Next-token generative scoring

## Usage

This model requires `trust_remote_code=True` to load the custom architecture.

```python
import torch
from transformers import AutoTokenizer
from echo_dsrn.modeling_generative_clf import EchoForGenerativeClassification

model_id = "ethicalabs/Echo-SmolTools-114M-Intent-CLF-Gen" # or your hub path

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = EchoForGenerativeClassification.from_pretrained(
    model_id,
    trust_remote_code=True,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

# Single utterance
label, probs = model.classify("Enter your text here", tokenizer)
print(f"Prediction: {label}")
```