distilbert-base-uncased-mnli

DistilBERT base model (uncased)

Table of Contents

• Model Details
• How to Get Started With the Model
• Uses
• Risks, Limitations and Biases
• Training
• Evaluation
• Environmental Impact

Model Details

Model Description: This is the uncased DistilBERT model fine-tuned on Multi-Genre Natural Language Inference (MNLI) dataset for the zero-shot classification task.

• Developed by: The Typeform team.
• Model Type: Zero-Shot Classification
• Language(s): English
• License: Unknown
• Parent Model: See the distilbert base uncased model for more information about the Distilled-BERT base model.

How to Get Started with the Model

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("typeform/distilbert-base-uncased-mnli")

model = AutoModelForSequenceClassification.from_pretrained("typeform/distilbert-base-uncased-mnli")

Uses

This model can be used for text classification tasks.

Risks, Limitations and Biases

CONTENT WARNING: Readers should be aware this section contains content that is disturbing, offensive, and can propagate historical and current stereotypes.

Significant research has explored bias and fairness issues with language models (see, e.g., Sheng et al. (2021) and Bender et al. (2021)).

Training

Training Data

This model of DistilBERT-uncased is pretrained on the Multi-Genre Natural Language Inference (MultiNLI) corpus. It is a crowd-sourced collection of 433k sentence pairs annotated with textual entailment information. The corpus covers a range of genres of spoken and written text, and supports a distinctive cross-genre generalization evaluation.

This model is also not case-sensitive, i.e., it does not make a difference between "english" and "English".

Training Procedure

Training is done on a p3.2xlarge AWS EC2 with the following hyperparameters:

$ run_glue.py \
    --model_name_or_path distilbert-base-uncased \
    --task_name mnli \
    --do_train \
    --do_eval \
    --max_seq_length 128 \
    --per_device_train_batch_size 16 \
    --learning_rate 2e-5 \
    --num_train_epochs 5 \
    --output_dir /tmp/distilbert-base-uncased_mnli/

Evaluation

Evaluation Results

When fine-tuned on downstream tasks, this model achieves the following results:

• **Epoch = ** 5.0
• Evaluation Accuracy = 0.8206875508543532
• Evaluation Loss = 0.8706700205802917
• ** Evaluation Runtime = ** 17.8278
• ** Evaluation Samples per second = ** 551.498

MNLI and MNLI-mm results:

Environmental Impact

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019). We present the hardware type based on the associated paper.

Hardware Type: 1 NVIDIA Tesla V100 GPUs

Hours used: Unknown

Cloud Provider: AWS EC2 P3

Compute Region: Unknown

Carbon Emitted: (Power consumption x Time x Carbon produced based on location of power grid): Unknown