CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 accelerate launch examples/training/ms_marco/train_bi-encoder_mnrl.py --model_name bigscience/bloom-7b1 --train_batch_size 32 --eval_batch_size 16 --freezenonbias --specb --lr 4e-4 --wandb --wandbwatchlog gradients --pooling weightedmean --gradcache --chunksize 8

Evaluation Results

{"ndcgs": {"sgpt-bloom-7b1-msmarco": {"scifact": {"NDCG@10": 0.71824}, "nfcorpus": {"NDCG@10": 0.35748}, "arguana": {"NDCG@10": 0.47281}, "scidocs": {"NDCG@10": 0.18435}, "fiqa": {"NDCG@10": 0.35736}, "cqadupstack": {"NDCG@10": 0.3708525}, "quora": {"NDCG@10": 0.74655}, "trec-covid": {"NDCG@10": 0.82731}, "webis-touche2020": {"NDCG@10": 0.2365}}}

See the evaluation folder or MTEB for more results.

Training

The model was trained with the parameters:

DataLoader:

torch.utils.data.dataloader.DataLoader of length 15600 with parameters:

{'batch_size': 32, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}

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": 10,
    "evaluation_steps": 0,
    "evaluator": "NoneType",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'transformers.optimization.AdamW'>",
    "optimizer_params": {
        "lr": 0.0004
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 1000,
    "weight_decay": 0.01
}

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 300, 'do_lower_case': False}) with Transformer model: BloomModel 
  (1): Pooling({'word_embedding_dimension': 4096, '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/sgpt-bloom-7b1-msmarco

Usage

Evaluation Results

Training

Full Model Architecture

Citing & Authors

bigscience/
sgpt-bloom-7b1-msmarco