CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 accelerate launch examples/training/nli/training_nli_v2.py --model_name bigscience/bloom-1b3 --freezenonbias --train_batch_size 128 --lr 32e-5 --pooling weightedmean --wandb --wandbwatchlog gradients --gradcache --chunksize 4

Evaluation Results

{'askubuntu': 57.44, 'cqadupstack': 14.18, 'twitterpara': 73.99, 'scidocs': 74.74, 'avg': 55.087500000000006}

Training

The model was trained with the parameters:

DataLoader:

sentence_transformers.datasets.NoDuplicatesDataLoader.NoDuplicatesDataLoader of length 4403 with parameters:

{'batch_size': 128}

The model uses BitFit, weighted-mean pooling & GradCache, for details see: https://arxiv.org/abs/2202.08904

Loss:

sentence_transformers.losses.MultipleNegativesRankingLoss.MNRLGradCache

Parameters of the fit()-Method:

{
    "epochs": 1,
    "evaluation_steps": 440,
    "evaluator": "sentence_transformers.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'transformers.optimization.AdamW'>",
    "optimizer_params": {
        "lr": 0.00032
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 441,
    "weight_decay": 0.01
}

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 75, 'do_lower_case': False}) with Transformer model: BloomModel 
  (1): Pooling({'word_embedding_dimension': 2048, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': True, 'pooling_mode_lasttoken': False})
)

Citing & Authors

bibtex

@article{muennighoff2022sgpt,
  title={SGPT: GPT Sentence Embeddings for Semantic Search},
  author={Muennighoff, Niklas},
  journal={arXiv preprint arXiv:2202.08904},
  year={2022}
}

bigscience-data/sgpt-bloom-1b7-nli

sgpt-bloom-1b7-nli

Usage

Evaluation Results

Training

Full Model Architecture

Citing & Authors

bigscience-data/
sgpt-bloom-1b7-nli