Model Card for st-polish-kartonberta-base-alpha-v1

This sentence transformer model is designed to convert text content into a 768-float vector space, ensuring an effective representation. It aims to be proficient in tasks involving sentence / document similarity.

The model has been released in its alpha version. Numerous potential enhancements could boost its performance, such as adjusting training hyperparameters or extending the training duration (currently limited to only one epoch). The main reason is limited GPU.

Model Description

• Developed by: Bartłomiej Orlik, https://www.linkedin.com/in/bartłomiej-orlik/
• Model type: RoBERTa Sentence Transformer
• Language: Polish
• License: LGPL-3.0
• Trained from model: sdadas/polish-roberta-base-v2: https://huggingface.co./sdadas/polish-roberta-base-v2

How to Get Started with the Model

Use the code below to get started with the model.

Using Sentence-Transformers

You can use the model with sentence-transformers:

pip install -U sentence-transformers

from sentence_transformers import SentenceTransformer

model = SentenceTransformer('OrlikB/st-polish-kartonberta-base-alpha-v1')

text_1 = 'Jestem wielkim fanem opakowań tekturowych'
text_2 = 'Bardzo podobają mi się kartony'

embeddings_1 = model.encode(text_1, normalize_embeddings=True)
embeddings_2 = model.encode(text_2, normalize_embeddings=True)

similarity = embeddings_1 @ embeddings_2.T
print(similarity)

Using HuggingFace Transformers

from transformers import AutoTokenizer, AutoModel
import torch
import numpy as np

def encode_text(text):
    encoded_input = tokenizer(text, padding=True, truncation=True, return_tensors='pt', max_length=512)
    with torch.no_grad():
        model_output = model(**encoded_input)
        sentence_embeddings = model_output[0][:, 0]
        sentence_embeddings = torch.nn.functional.normalize(sentence_embeddings, p=2, dim=1)
    return  sentence_embeddings.squeeze().numpy()

cosine_similarity = lambda a, b: np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))


tokenizer = AutoTokenizer.from_pretrained('OrlikB/st-polish-kartonberta-base-alpha-v1')
model = AutoModel.from_pretrained('OrlikB/st-polish-kartonberta-base-alpha-v1')
model.eval()

text_1 = 'Jestem wielkim fanem opakowań tekturowych'
text_2 = 'Bardzo podobają mi się kartony'

embeddings_1 = encode_text(text_1)
embeddings_2 = encode_text(text_2)

print(cosine_similarity(embeddings_1, embeddings_2))

*Note: You can use the encode_text function for demonstration purposes. For the best experience, it's recommended to process text in batches.

Evaluation

MTEB for Polish Language

Rank	Model	Model Size (GB)	Embedding Dimensions	Sequence Length	Average (26 datasets)	Classification Average (7 datasets)	Clustering Average (1 datasets)	Pair Classification Average (4 datasets)	Retrieval Average (11 datasets)	STS Average (3 datasets)
1	multilingual-e5-large	2.24	1024	514	58.25	60.51	24.06	84.58	47.82	67.52
2	st-polish-kartonberta-base-alpha-v1	0.5	768	514	56.92	60.44	32.85	87.92	42.19	69.47
3	multilingual-e5-base	1.11	768	514	54.18	57.01	18.62	82.08	42.5	65.07
4	multilingual-e5-small	0.47	384	512	53.15	54.35	19.64	81.67	41.52	66.08
5	st-polish-paraphrase-from-mpnet	0.5	768	514	53.06	57.49	25.09	87.04	36.53	67.39
6	st-polish-paraphrase-from-distilroberta	0.5	768	514	52.65	58.55	31.11	87	33.96	68.78

More Information

I developed this model as a personal scientific initiative.

I plan to start the development on a new ST model. However, due to limited computational resources, I suspended further work to create a larger or enhanced version of current model.