all-MiniLM-L12-v2-embedding-all

This model is a fine-tuned version of all-MiniLM-L12-v2 on the following datasets: squad, newsqa, LLukas22/cqadupstack, LLukas22/fiqa, LLukas22/scidocs, deepset/germanquad, LLukas22/nq.

Usage (Sentence-Transformers)

Using this model becomes easy when you have sentence-transformers installed:

pip install -U sentence-transformers

Then you can use the model like this:

from sentence_transformers import SentenceTransformer
sentences = ["This is an example sentence", "Each sentence is converted"]

model = SentenceTransformer('LLukas22/all-MiniLM-L12-v2-embedding-all')
embeddings = model.encode(sentences)
print(embeddings)

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 1E+00
per device batch size: 60
effective batch size: 180
seed: 42
optimizer: AdamW with betas (0.9,0.999) and eps 1E-08
weight decay: 2E-02
D-Adaptation: True
Warmup: True
number of epochs: 20
mixed_precision_training: bf16

Training results

Epoch	Train Loss	Validation Loss
0	0.0708	0.0619
1	0.0609	0.0567
2	0.0531	0.0542
3	0.0475	0.0528
4	0.0428	0.0521
5	0.0389	0.0513
6	0.0352	0.0508
7	0.0322	0.0494
8	0.0289	0.0485
9	0.0264	0.0483
10	0.0242	0.0466
11	0.0221	0.0459
12	0.0204	0.0469
13	0.0189	0.0459

Evaluation results

Epoch	top_1	top_3	top_5	top_10	top_25
0	0.507	0.665	0.721	0.784	0.847
1	0.501	0.661	0.719	0.783	0.846
2	0.508	0.669	0.726	0.789	0.851
3	0.507	0.665	0.722	0.785	0.85
4	0.506	0.667	0.724	0.788	0.851
5	0.511	0.673	0.731	0.795	0.857
6	0.51	0.674	0.732	0.794	0.856
7	0.512	0.674	0.732	0.796	0.859
8	0.515	0.678	0.736	0.799	0.861
9	0.514	0.679	0.737	0.8	0.862
10	0.52	0.683	0.741	0.803	0.864
11	0.522	0.686	0.744	0.806	0.866
12	0.519	0.683	0.741	0.804	0.864
13	0.522	0.685	0.743	0.806	0.865

Framework versions

Transformers: 4.25.1
PyTorch: 2.0.0.dev20230210+cu118
PyTorch Lightning: 1.8.6
Datasets: 2.7.1
Tokenizers: 0.13.1
Sentence Transformers: 2.2.2

Additional Information

This model was trained as part of my Master's Thesis 'Evaluation of transformer based language models for use in service information systems'. The source code is available on Github.

LLukas22
/

all-MiniLM-L12-v2-embedding-all