philipp-zettl/GGU-CLF-xx

Model Card for GGU-CLF

Model Details

Model Description

This is a simple classification model trained on a custom dataset.

Please note that this model, although it is implemented in the transformers library. Is not a usual transformer. It combines the underlying embedding model with the required tokenizer into a simple-to-use pipeline for sequence classification.

It is used to classify user text into the following classes:

• 0: Greeting
• 1: Gratitude
• 2: Unknown

Note: To use this model please remember the following things

1. The model is an XLMRoberta model based on BAAI/bge-m3.
2. The required tokenizer is baked into the classifier implementation.

• Developed by: philipp-zettl
• Funded by [optional]: [More Information Needed]
• Shared by [optional]: [More Information Needed]
• Model type: [More Information Needed]
• Language(s) (NLP): multilingual
• License: mit
• Finetuned from model [optional]: BAAI/bge-m3

Model Sources [optional]

• Repository: philipp-zettl/GGU-CLF
• Paper [optional]: [More Information Needed]
• Demo [optional]: [More Information Needed]

Uses

Direct Use

Use this model to classify messages from natural language chats.

Downstream Use [optional]

[More Information Needed]

Out-of-Scope Use

The model was not trained on multi-sentence samples. You should avoid those.

Oficially tested and supported languages are english and german any other language is considered out of scope.

Bias, Risks, and Limitations

[More Information Needed]

Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

How to Get Started with the Model

Use the code below to get started with the model.

from transformers import AutoModel

model = AutoModel.from_pretrained("philipp-zettl/GGU-xx").to(torch.float16).to('cuda')

model([
    'Hi wie gehts?',
    'Dannke dir mein freund!',
    'Merci freundchen, send mir mal ein paar Machine Learning jobs.',
    'Works as expected, cheers!',
    'How you doin my boy',
    'send me immediately some matching jobs, thanks',
    'wer's eigentlich tom selleck?',
    'sprichst du deutsch?',
    'sprechen sie deutsch sie hurensohn?',
    'vergeltsgott',
    'heidenei dank dir recht herzlich',
    'grazie mille bambino, come estas'
]).argmax(dim=1)

Training Details

Training Data

This model was trained using the philipp-zettl/GGU-xx dataset.

You can find it's performance metrics under Evaluation Results.

Training Procedure

Preprocessing [optional]

The following code was used to create the data set as well as split the data set into training and validation sets.

from datasets import load_dataset

class Dataset:
    def __init__(self, dataset, target_names=None):
        self.data = list(map(lambda x: x[0], dataset))
        self.target = list(map(lambda x: x[1], dataset))
        self.target_names = target_names


ds = load_dataset('philipp-zettl/GGU-xx')
data = Dataset([[e['sample'], e['label']] for e in ds['train']], ['greeting', 'gratitude', 'unknown'])
X_train, X_test, y_train, y_test = train_test_split(data.data, data.target, test_size=0.2, random_state=42)

Training Hyperparameters

• Training regime: [More Information Needed]

Speeds, Sizes, Times [optional]

[More Information Needed]

Evaluation

Testing Data, Factors & Metrics

Testing Data

[More Information Needed]

Factors

[More Information Needed]

Metrics

[More Information Needed]

Results

[More Information Needed]

Summary

Model Examination [optional]

You can find the initial implementation of the classification model here:

from transformers import PreTrainedModel, PretrainedConfig, AutoModel, AutoTokenizer
import torch
import torch.nn as nn


class EmbeddingClassifierConfig(PretrainedConfig):
    model_type = 'xlm-roberta'

    def __init__(self, num_classes=3, base_model='BAAI/bge-m3', tokenizer='BAAI/bge-m3', dropout=0.0, l2_reg=0.01, torch_dtype=torch.float16, **kwargs):
        self.num_classes = num_classes
        self.base_model = base_model
        self.tokenizer = tokenizer
        self.dropout = dropout
        self.l2_reg = l2_reg
        self.torch_dtype = torch_dtype
        super().__init__(**kwargs)


class EmbeddingClassifier(PreTrainedModel):
    config_class = EmbeddingClassifierConfig

    def __init__(self, config):
        super().__init__(config)
        base_model = config.base_model
        tokenizer = config.tokenizer

        if base_model is None or isinstance(tokenizer, str):
            base_model = AutoModel.from_pretrained(base_model)#, torch_dtype=config.torch_dtype)
        if tokenizer is None or isinstance(tokenizer, str):
            tokenizer = AutoTokenizer.from_pretrained(tokenizer)

        self.tokenizer = tokenizer
        self.base = base_model
        self.fc = nn.Linear(base_model.config.hidden_size, config.num_classes)#, torch_dtype=config.torch_dtype)
        self.do = nn.Dropout(config.dropout)#, torch_dtype=config.torch_dtype)
        self.l2_reg = config.l2_reg

        self.to(config.torch_dtype)

    def forward(self, X):
        encoding = self.tokenizer(
            X, return_tensors='pt',
            padding=True, truncation=True
        ).to(self.device)
        input_ids = encoding['input_ids']
        attention_mask = encoding['attention_mask']
        emb = self.base(
            input_ids,
            attention_mask=attention_mask,
            return_dict=True,
            output_hidden_states=True
        ).last_hidden_state[:, 0, :]
        return self.fc(self.do(emb))

    def train(self, set_val=True):
        self.base.train(False)
        for param in self.base.parameters():
            param.requires_grad = False
        for param in self.fc.parameters():
            param.requires_grad = set_val

    def get_l2_loss(self):
        l2_loss = torch.tensor(0.).to('cuda')
        for param in self.parameters():
            if param.requires_grad:
                l2_loss += torch.norm(param, 2)
        return self.l2_reg * l2_loss

Environmental Impact

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

• Hardware Type: [More Information Needed]
• Hours used: [More Information Needed]
• Cloud Provider: [More Information Needed]
• Compute Region: [More Information Needed]
• Carbon Emitted: [More Information Needed]

Technical Specifications [optional]

Model Architecture and Objective

[More Information Needed]

Compute Infrastructure

[More Information Needed]

Hardware

[More Information Needed]

Software

[More Information Needed]

Citation [optional]

BibTeX:

[More Information Needed]

APA:

[More Information Needed]

Glossary [optional]

[More Information Needed]

More Information [optional]

[More Information Needed]

Model Card Authors [optional]

[More Information Needed]

Model Card Contact

[More Information Needed]