Modelpublic

Derify/ChemRanker-alpha-qed-cutoff-sim

sourceHugging Faceapache-2.0updated 7mo agoView on Hugging Face

1likes9downloads

Model Card

Derify/ChemRanker-alpha-qed-cutoff-sim

This Cross Encoder is finetuned from Derify/ModChemBERT-IR-BASE using hard-negative triplets derived from Derify/pubchem_10m_genmol_similarity. Positive SMILES pairs are first filtered by quality and similarity constraints, then reduced to one strongest positive target per anchor molecule to create a high-signal training set for reranking. The model computes relevance scores for pairs of SMILES strings, enabling SMILES reranking and molecular semantic search.

For this variant, the positives are selected with a composite ranking criterion that combines high QED and similarity, where the similarity contribution is explicitly capped at 0.75 to prevent similarity from dominating the ranking score. The quality stage uses strict inequality filtering (QED > 0.85, similarity > 0.5, with similarity also bounded below 1.0), and then keeps the top-scoring pair per anchor molecule.

Hard negatives are mined with Sentence Transformers using Derify/ChemMRL-beta as the teacher model and a TopK-PercPos-style margin setting based on NV-Retriever, with relative_margin=0.05 and max_negative_score_threshold = pos_score * percentage_margin. Training uses triplet-format samples with 5 mined negatives per anchor-positive pair and optimizes a multiple-negatives ranking objective, while reranking evaluation uses n-tuple samples with 30 mined negatives per query.

Model Details

Model Description

—Model Type: Cross Encoder
—Base model: Derify/ModChemBERT-IR-BASE
—Maximum Sequence Length: 512 tokens
—Number of Output Labels: 1 label
—Training Dataset:
—Derify/pubchem_10m_genmol_similarity Mined Hard Negatives
—License: apache-2.0

Model Sources

—Documentation: Sentence Transformers Documentation
—Documentation: Cross Encoder Documentation
—Repository: Sentence Transformers on GitHub
—Hugging Face: Cross Encoders on Hugging Face

Usage

Direct Usage (Sentence Transformers)

First install the Transformers and Sentence Transformers libraries:

bash

pip install -U "transformers>=4.57.1,<5.0.0"
pip install -U sentence-transformers

Then you can load this model and run inference.

python

from sentence_transformers import CrossEncoder

# Download from the 🤗 Hub
model = CrossEncoder("Derify/ChemRanker-alpha-qed-cutoff-sim")
# Get scores for pairs of texts
pairs = [
    ['c1snnc1C[NH2+]Cc1cc2c(s1)CCC2', 'c1snnc1CCC[NH2+]Cc1cc2c(s1)CCC2'],
    ['c1sc2c(c1-c1nc(C3CCOC3)no1)CCCC2', 'O=C([O-])Cc1noc(-c2csc3c2CCCC3)n1'],
    ['c1sc(C[NH2+]C2CC2)nc1C[NH+]1CCN2CCCC2C1', 'FC(F)[NH2+]Cc1nc(C[NH+]2CCN3CCCC3C2)cs1'],
    ['c1sc(CC[NH+]2CCOCC2)nc1C[NH2+]C1CC1', 'CCc1nc(C[NH2+]C2CC2)cs1'],
    ['c1sc(CC2CCC[NH2+]2)nc1C1CCCO1', 'c1sc(CC2CCC[NH2+]2)nc1C1CCCC1'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'c1snnc1C[NH2+]Cc1cc2c(s1)CCC2',
    [
        'c1snnc1CCC[NH2+]Cc1cc2c(s1)CCC2',
        'O=C([O-])Cc1noc(-c2csc3c2CCCC3)n1',
        'FC(F)[NH2+]Cc1nc(C[NH+]2CCN3CCCC3C2)cs1',
        'CCc1nc(C[NH2+]C2CC2)cs1',
        'c1sc(CC2CCC[NH2+]2)nc1C1CCCC1',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

<!--

Direct Usage (Transformers)

<details><summary>Click to see the direct usage in Transformers</summary>

</details> -->

<!--

Downstream Usage (Sentence Transformers)

You can finetune this model on your own dataset.

<details><summary>Click to expand</summary>

</details> -->

<!--

Out-of-Scope Use

List how the model may foreseeably be misused and address what users ought not to do with the model. -->

Evaluation

Metrics

Cross Encoder Reranking

—Evaluated with <code>CrossEncoderRerankingEvaluator</code> with these parameters:

json

  {
      "at_k": 10
  }

Metric	Value
map	0.4232
mrr@10	0.6667
ndcg@10	0.6873

<!--

Bias, Risks and Limitations

What are the known or foreseeable issues stemming from this model? You could also flag here known failure cases or weaknesses of the model. -->

<!--

Recommendations

What are recommendations with respect to the foreseeable issues? For example, filtering explicit content. -->

Training Details

Training Dataset

GenMol Similarity Hard Negatives

—Dataset: GenMol Similarity Hard Negatives
—Size: 3,212,363 training samples
—Columns: <code>smilesa</code>, <code>smilesb</code>, and <code>negative</code>
—Approximate statistics based on the first 1000 samples: | | smilesa | smilesb | negative | | :------ | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :-------------------------------------------------------------------------------------------- | | type | string | string | string | | details | <ul><li>min: 19 characters</li><li>mean: 33.59 characters</li><li>max: 65 characters</li></ul> | <ul><li>min: 20 characters</li><li>mean: 34.26 characters</li><li>max: 48 characters</li></ul> | <ul><li>min: 19 characters</li><li>mean: 33.3 characters</li><li>max: 57 characters</li></ul> |
—Samples: | smilesa | smilesb | negative | | :---------------------------------------------- | :------------------------------------------------- | :------------------------------------------------- | | <code>c1sc2cc3c(cc2c1CC[NH2+]C1CC1)OCCO3</code> | <code>FC(F)(F)[NH2+]CCc1csc2cc3c(cc12)OCCO3</code> | <code>[NH3+]CCCc1cc2c(cc1C1CC1)OCO2</code> | | <code>c1sc2cc3c(cc2c1CC[NH2+]C1CC1)OCCO3</code> | <code>FC(F)(F)[NH2+]CCc1csc2cc3c(cc12)OCCO3</code> | <code>O=c1[nH]c2cc3c(cc2cc1CNC1CCCCC1)OCCO3</code> | | <code>c1sc2cc3c(cc2c1CC[NH2+]C1CC1)OCCO3</code> | <code>FC(F)(F)[NH2+]CCc1csc2cc3c(cc12)OCCO3</code> | <code>NCCc1c2c(cc3c1OCCC3)OCCC2</code> |
—Loss: <code>MultipleNegativesRankingLoss</code> with these parameters:

json

  {
      "scale": 10.0,
      "num_negatives": 4,
      "activation_fn": "torch.nn.modules.activation.Sigmoid"
  }

Evaluation Dataset

GenMol Similarity Hard Negatives

—Dataset: GenMol Similarity Hard Negatives
—Size: 165,968 evaluation samples
—Columns: <code>smilesa</code>, <code>smilesb</code>, <code>negative1</code>, <code>negative2</code>, <code>negative3</code>, <code>negative4</code>, <code>negative5</code>, <code>negative6</code>, <code>negative7</code>, <code>negative8</code>, <code>negative9</code>, <code>negative10</code>, <code>negative11</code>, <code>negative12</code>, <code>negative13</code>, <code>negative14</code>, <code>negative15</code>, <code>negative16</code>, <code>negative17</code>, <code>negative18</code>, <code>negative19</code>, <code>negative20</code>, <code>negative21</code>, <code>negative22</code>, <code>negative23</code>, <code>negative24</code>, <code>negative25</code>, <code>negative26</code>, <code>negative27</code>, <code>negative28</code>, <code>negative29</code>, and <code>negative30</code>
—Approximate statistics based on the first 1000 samples: | | smilesa | smilesb | negative1 | negative2 | negative3 | negative4 | negative5 | negative6 | negative7 | negative8 | negative9 | negative10 | negative11 | negative12 | negative13 | negative14 | negative15 | negative16 | negative17 | negative18 | negative19 | negative20 | negative21 | negative22 | negative23 | negative24 | negative25 | negative26 | negative27 | negative28 | negative29 | negative30 | | :------ | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :-------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :-------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- | | type | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | string | | details | <ul><li>min: 17 characters</li><li>mean: 37.57 characters</li><li>max: 96 characters</li></ul> | <ul><li>min: 14 characters</li><li>mean: 34.39 characters</li><li>max: 70 characters</li></ul> | <ul><li>min: 18 characters</li><li>mean: 35.68 characters</li><li>max: 77 characters</li></ul> | <ul><li>min: 12 characters</li><li>mean: 35.11 characters</li><li>max: 77 characters</li></ul> | <ul><li>min: 14 characters</li><li>mean: 35.24 characters</li><li>max: 81 characters</li></ul> | <ul><li>min: 17 characters</li><li>mean: 35.41 characters</li><li>max: 73 characters</li></ul> | <ul><li>min: 14 characters</li><li>mean: 35.04 characters</li><li>max: 70 characters</li></ul> | <ul><li>min: 14 characters</li><li>mean: 35.17 characters</li><li>max: 84 characters</li></ul> | <ul><li>min: 16 characters</li><li>mean: 35.04 characters</li><li>max: 64 characters</li></ul> | <ul><li>min: 13 characters</li><li>mean: 35.24 characters</li><li>max: 90 characters</li></ul> | <ul><li>min: 11 characters</li><li>mean: 35.12 characters</li><li>max: 90 characters</li></ul> | <ul><li>min: 15 characters</li><li>mean: 35.22 characters</li><li>max: 70 characters</li></ul> | <ul><li>min: 12 characters</li><li>mean: 35.38 characters</li><li>max: 74 characters</li></ul> | <ul><li>min: 15 characters</li><li>mean: 35.38 characters</li><li>max: 73 characters</li></ul> | <ul><li>min: 13 characters</li><li>mean: 35.24 characters</li><li>max: 67 characters</li></ul> | <ul><li>min: 10 characters</li><li>mean: 34.9 characters</li><li>max: 77 characters</li></ul> | <ul><li>min: 17 characters</li><li>mean: 35.23 characters</li><li>max: 77 characters</li></ul> | <ul><li>min: 18 characters</li><li>mean: 35.11 characters</li><li>max: 72 characters</li></ul> | <ul><li>min: 14 characters</li><li>mean: 35.4 characters</li><li>max: 65 characters</li></ul> | <ul><li>min: 14 characters</li><li>mean: 35.35 characters</li><li>max: 65 characters</li></ul> | <ul><li>min: 18 characters</li><li>mean: 35.25 characters</li><li>max: 62 characters</li></ul> | <ul><li>min: 14 characters</li><li>mean: 35.55 characters</li><li>max: 65 characters</li></ul> | <ul><li>min: 18 characters</li><li>mean: 35.53 characters</li><li>max: 81 characters</li></ul> | <ul><li>min: 17 characters</li><li>mean: 35.42 characters</li><li>max: 68 characters</li></ul> | <ul><li>min: 14 characters</li><li>mean: 35.53 characters</li><li>max: 68 characters</li></ul> | <ul><li>min: 14 characters</li><li>mean: 35.49 characters</li><li>max: 64 characters</li></ul> | <ul><li>min: 16 characters</li><li>mean: 35.29 characters</li><li>max: 83 characters</li></ul> | <ul><li>min: 17 characters</li><li>mean: 35.77 characters</li><li>max: 77 characters</li></ul> | <ul><li>min: 11 characters</li><li>mean: 35.43 characters</li><li>max: 77 characters</li></ul> | <ul><li>min: 16 characters</li><li>mean: 35.55 characters</li><li>max: 64 characters</li></ul> | <ul><li>min: 16 characters</li><li>mean: 35.45 characters</li><li>max: 69 characters</li></ul> | <ul><li>min: 16 characters</li><li>mean: 35.33 characters</li><li>max: 77 characters</li></ul> |
—Samples: | smilesa | smilesb | negative1 | negative2 | negative3 | negative4 | negative5 | negative6 | negative7 | negative8 | negative9 | negative10 | negative11 | negative12 | negative13 | negative14 | negative15 | negative16 | negative17 | negative18 | negative19 | negative20 | negative21 | negative22 | negative23 | negative24 | negative25 | negative26 | negative27 | negative28 | negative29 | negative30 | | :--------------------------------------------------- | :--------------------------------------------------- | :--------------------------------------------------- | :---------------------------------------------------- | :------------------------------------------------- | :------------------------------------------ | :-------------------------------------------------- | :----------------------------------------- | :------------------------------------------------- | :--------------------------------------------------- | :---------------------------------------------------- | :------------------------------------------------- | :------------------------------------------------------- | :----------------------------------------------- | :------------------------------------------------------ | :------------------------------------------------------ | :----------------------------------------------- | :--------------------------------------------------- | :------------------------------------------------ | :---------------------------------------------------- | :------------------------------------------------- | :----------------------------------------------------- | :---------------------------------------------------- | :---------------------------------------------------- | :----------------------------------------------------- | :------------------------------------------------ | :--------------------------------------------- | :--------------------------------------------------------- | :------------------------------------------------ | :--------------------------------------------------- | :------------------------------------------------------- | :----------------------------------------------- | | <code>c1snnc1C[NH2+]Cc1cc2c(s1)CCC2</code> | <code>c1snnc1CCC[NH2+]Cc1cc2c(s1)CCC2</code> | <code>c1snnc1CCC[NH2+]Cc1cc2c(s1)CCC2</code> | <code>Cn1cc(C[NH2+]Cc2cc3c(s2)CCC3)nn1</code> | <code>Cn1cc(CC[NH2+]Cc2cc3c(s2)CCC3)nn1</code> | <code>Cc1cc(C[NH2+]Cc2csnn2)sc1C</code> | <code>NC(=O)c1csc(C[NH2+]Cc2cc3c(s2)CCC3)c1</code> | <code>Cc1cc(CC[NH2+]Cc2csnn2)sc1C</code> | <code>Ic1ccc(C[NH2+]Cc2cc3c(s2)CCC3)o1</code> | <code>c1ncc(C[NH2+]Cc2csnn2)s1</code> | <code>FC(F)c1csc(C[NH2+]Cc2cc3c(s2)CCC3)c1</code> | <code>c1c(C[NH2+]CC2CC2)sc2c1CSCC2</code> | <code>N#Cc1cc(F)cc(C[NH2+]Cc2cc3c(s2)CCC3)c1</code> | <code>c1cc(C[NH2+]Cc2nc3c(s2)CCC3)no1</code> | <code>CCc1ccc(C[NH2+]Cc2csnn2)s1</code> | <code>CNH+Cc1nnc(-c2cc3c(s2)CCCC3)o1</code> | <code>Fc1cc(C[NH2+]Cc2cc3c(s2)CCC3)ccc1Br</code> | <code>FC(F)(F)C[NH2+]Cc1cc2c(s1)CCSC2</code> | <code>c1cc(C[NH2+]Cc2cc3c(s2)CCC3)c[nH]1</code> | <code>Cc1cc(C)c(CC[NH2+]Cc2cc3c(s2)CCC3)c(C)c1</code> | <code>Oc1ccc(C[NH2+]Cc2cc3c(s2)CCC3)cc1Br</code> | <code>O=C([O-])c1ccc(CC[NH2+]Cc2cc3c(s2)CCC3)s1</code> | <code>c1c(C[NH2+]CC2CCCC2)sc2c1CCC2</code> | <code>O=C([O-])c1ccc(C[NH2+]Cc2cc3c(s2)CCC3)s1</code> | <code>COc1cc(C)cc(C[NH2+]Cc2cc3c(s2)CCC3)c1</code> | <code>OCc1ccc(Br)cc1C[NH2+]Cc1cc2c(s1)CCC2</code> | <code>CCc1cnc(C[NH2+]Cc2csnn2)s1</code> | <code>Clc1cc(C[NH2+]Cc2cc3c(s2)CCC3)ccc1Br</code> | <code>c1c(C[NH2+]CC2CC2)sc2c1CCCCC2</code> | <code>Cc1ccccc1C[NH2+]Cc1cc2c(s1)CCC2</code> | <code>c1cc(C[NH+]2CCCC2)sc1C[NH2+]Cc1cc2c(s1)CCC2</code> | <code>Cc1cc(C[NH2+]Cc2cc3c(s2)CCC3)ccc1F</code> | | <code>c1sc2c(c1-c1nc(C3CCOC3)no1)CCCC2</code> | <code>O=C([O-])Cc1noc(-c2csc3c2CCCC3)n1</code> | <code>Nc1sc2c(c1-c1nc(C3CCOC3)no1)CCCC2</code> | <code>Nc1sc2c(c1-c1nc(C3CCC3)no1)CCCC2</code> | <code>c1c(-c2nc(C3CCCNC3)no2)sc2c1CCCCCC2</code> | <code>Nc1sccc1-c1nc(C2CCCOC2)no1</code> | <code>Nc1sc2c(c1-c1nc(C3CCCO3)no1)CCCC2</code> | <code>Cc1csc(-c2nc(C3CCOCC3)no2)c1N</code> | <code>Cc1oc2c(c1-c1nc(C3CCOC3)no1)C(=O)CCC2</code> | <code>c1c(-c2nc(C3C[NH2+]CCO3)no2)sc2c1CCCCC2</code> | <code>O=C([O-])Nc1sc2c(c1-c1nc(C3CC3)no1)CCCC2</code> | <code>c1cc2c(s1)CCCC2c1nc(C2CC2)no1</code> | <code>CC(=O)N1CCCC(c2noc(-c3cc4c(s3)CCCCCC4)n2)C1</code> | <code>Cc1cc(-c2nc([C@@H]3CCOC3)no2)c(N)s1</code> | <code>c1cc2c(nc1-c1noc(C3CCCOC3)n1)CCCC2</code> | <code>Nc1sccc1-c1nc(C2CCCC2)no1</code> | <code>c1cc2c(nc1-c1noc(C3CCOCC3)n1)CCCC2</code> | <code>[NH3+]C(c1noc(-c2cc3c(s2)CCCC3)n1)C1CC1</code> | <code>c1cc2c(c(-c3nc(C4CCOCC4)no3)c1)CCCN2</code> | <code>c1c(-c2nc(C3CC3)no2)nn2c1CCCC2</code> | <code>CN1CC(c2noc(-c3cc4c(s3)CCCC4)n2)CC1=O</code> | <code>O=C([O-])Cc1noc(-c2csc3c2CCCC3)n1</code> | <code>Oc1c(-c2nc(C3CCC(F)(F)C3)no2)ccc2c1CCCC2</code> | <code>Cc1cc(=O)c(-c2noc(C3CCCOC3)n2)c2n1CCC2</code> | <code>O=C([O-])CNc1sc2c(c1-c1nc(C3CC3)no1)CCCC2</code> | <code>CC1CCc2c(sc(N)c2-c2nc(C3CC3)no2)C1</code> | <code>Cn1nc(-c2nc(C3CCCO3)no2)c2c1CCCC2</code> | <code>O=C(Nc1sc2c(c1-c1nc(C3CC3)no1)COCC2)C1=CCCCC1</code> | <code>Cc1cscc1-c1noc(C2CCOCC2)n1</code> | <code>CC1(C)CCCc2sc(N)c(-c3nc(C4CC4)no3)c21</code> | <code>Clc1cc2c(c(-c3nc(C4CCOC4)no3)c1)OCC2</code> | <code>Nc1sc2c(c1-c1nnc(C3CC3)o1)CCCC2</code> | | <code>c1sc(C[NH2+]C2CC2)nc1C[NH+]1CCN2CCCC2C1</code> | <code>FC(F)[NH2+]Cc1nc(C[NH+]2CCN3CCCC3C2)cs1</code> | <code>FC(F)[NH2+]Cc1nc(C[NH+]2CCN3CCCC3C2)cs1</code> | <code>CC(C)[NH2+]Cc1nc(C[NH+]2CCC3CCCCC3C2)cs1</code> | <code>CN1C2CCC1CNH+n1)CC2</code> | <code>Nc1nc(CC[NH+]2CCCN3CCCC3C2)cs1</code> | <code>CC1CNH+n2)CCN1C</code> | <code>Oc1csc(CN2CCCC3C[NH2+]CC32)n1</code> | <code>CCc1nc(C[NH+]2CCCC3CCCCC32)cs1</code> | <code>C[NH2+]Cc1csc(N2CC[NH+]3CCCC3C2)n1</code> | <code>[NH3+]Cc1nc(C[NH+]2CCC3CCCCC32)cs1</code> | <code>CC1CN2CCCCC2C[NH+]1Cc1csc(CC[NH3+])n1</code> | <code>CCCc1nc(CN2CCCC2C2CCC[NH2+]2)cs1</code> | <code>ClCCc1nc(CN2CCCC2C2CCC[NH2+]2)cs1</code> | <code>c1cc(C[NH2+]C2CC2)c(C[NH+]2CCN3CCCCC3C2)o1</code> | <code>O=C(Cc1nc(CCl)cs1)N1CCC[NH+]2CCCC2C1</code> | <code>N#CCc1nc(C[NH+]2CCCC3CCCCC32)cs1</code> | <code>CC[NH2+]Cc1csc(N2CCC3C(CCC[NH+]3C)C2)n1</code> | <code>c1sc(C[NH2+]C2CC2)nc1C[NH+]1CCCCC1</code> | <code>[NH3+]Cc1nc(C[NH+]2CCCC2C2CCCC2)cs1</code> | <code>Cc1csc(C[NH+]2CCC3C[NH2+]CC3C2)n1</code> | <code>ClOCc1csc(C[NH+]2CC3C[NH2+]CC3C2)n1</code> | <code>c1cc(C[NH+]2CCCN3CCCC3C2)nc(C2CC2)n1</code> | <code>Cc1ccsc1C[NH2+]CCN1CCN2CCCC2C1</code> | <code>c1sc(C[NH2+]C2CCCC2)nc1C[NH+]1CCCCC1</code> | <code>Brc1csc(C[NH2+]CCN2CCN3CCCCC3C2)c1</code> | <code>Cc1nc(CCC[NH2+]C2CCN3CCCCC23)cs1</code> | <code>CCOC(=O)c1nc(CN2CC3CCC[NH2+]C3C2)cs1</code> | <code>CCCC(=O)c1nc(CN2CC3CCC[NH2+]C3C2)cs1</code> | <code>CC(C)(C)c1csc(CN2CCC[NH2+]C(C3CC3)C2)n1</code> | <code>COCc1nc(CN2CCC([NH3+])C2)cs1</code> | <code>CCC[NH2+]Cc1nc(C[NH+]2CC3CCC2C3)cs1</code> |
—Loss: <code>MultipleNegativesRankingLoss</code> with these parameters:

json

  {
      "scale": 10.0,
      "num_negatives": 4,
      "activation_fn": "torch.nn.modules.activation.Sigmoid"
  }

Training Hyperparameters

Non-Default Hyperparameters

—eval_strategy: epoch
—per_device_train_batch_size: 256
—per_device_eval_batch_size: 256
—torch_empty_cache_steps: 1000
—learning_rate: 3e-05
—weight_decay: 1e-05
—max_grad_norm: None
—lr_scheduler_type: warmupstabledecay
—lr_scheduler_kwargs: {'numdecaysteps': 6274, 'warmuptype': 'linear', 'decaytype': '1-sqrt'}
—warmup_steps: 6274
—seed: 12
—data_seed: 24681357
—bf16: True
—bf16_full_eval: True
—tf32: True
—dataloader_num_workers: 8
—dataloader_prefetch_factor: 2
—load_best_model_at_end: True
—optim: stable_adamw
—optim_args: decouplelr=True,maxlr=3e-05
—dataloader_persistent_workers: True
—resume_from_checkpoint: False
—gradient_checkpointing: True
—torch_compile: True
—torch_compile_backend: inductor
—torch_compile_mode: max-autotune
—eval_on_start: True
—batch_sampler: no_duplicates

All Hyperparameters

<details><summary>Click to expand</summary>

—overwrite_output_dir: False
—do_predict: False
—eval_strategy: epoch
—prediction_loss_only: True
—per_device_train_batch_size: 256
—per_device_eval_batch_size: 256
—per_gpu_train_batch_size: None
—per_gpu_eval_batch_size: None
—gradient_accumulation_steps: 1
—eval_accumulation_steps: None
—torch_empty_cache_steps: 1000
—learning_rate: 3e-05
—weight_decay: 1e-05
—adam_beta1: 0.9
—adam_beta2: 0.999
—adam_epsilon: 1e-08
—max_grad_norm: None
—num_train_epochs: 3
—max_steps: -1
—lr_scheduler_type: warmupstabledecay
—lr_scheduler_kwargs: {'numdecaysteps': 6274, 'warmuptype': 'linear', 'decaytype': '1-sqrt'}
—warmup_ratio: 0.0
—warmup_steps: 6274
—log_level: passive
—log_level_replica: warning
—log_on_each_node: True
—logging_nan_inf_filter: True
—save_safetensors: True
—save_on_each_node: False
—save_only_model: False
—restore_callback_states_from_checkpoint: False
—no_cuda: False
—use_cpu: False
—use_mps_device: False
—seed: 12
—data_seed: 24681357
—jit_mode_eval: False
—bf16: True
—fp16: False
—fp16_opt_level: O1
—half_precision_backend: auto
—bf16_full_eval: True
—fp16_full_eval: False
—tf32: True
—local_rank: 0
—ddp_backend: None
—tpu_num_cores: None
—tpu_metrics_debug: False
—debug: []
—dataloader_drop_last: True
—dataloader_num_workers: 8
—dataloader_prefetch_factor: 2
—past_index: -1
—disable_tqdm: False
—remove_unused_columns: True
—label_names: None
—load_best_model_at_end: True
—ignore_data_skip: False
—fsdp: []
—fsdp_min_num_params: 0
—fsdp_config: {'minnumparams': 0, 'xla': False, 'xlafsdpv2': False, 'xlafsdpgrad_ckpt': False}
—fsdp_transformer_layer_cls_to_wrap: None
—accelerator_config: {'splitbatches': False, 'dispatchbatches': None, 'evenbatches': True, 'useseedablesampler': True, 'nonblocking': False, 'gradientaccumulationkwargs': None}
—parallelism_config: None
—deepspeed: None
—label_smoothing_factor: 0.0
—optim: stable_adamw
—optim_args: decouplelr=True,maxlr=3e-05
—adafactor: False
—group_by_length: False
—length_column_name: length
—project: huggingface
—trackio_space_id: trackio
—ddp_find_unused_parameters: None
—ddp_bucket_cap_mb: None
—ddp_broadcast_buffers: False
—dataloader_pin_memory: True
—dataloader_persistent_workers: True
—skip_memory_metrics: True
—use_legacy_prediction_loop: False
—push_to_hub: False
—resume_from_checkpoint: False
—hub_model_id: None
—hub_strategy: every_save
—hub_private_repo: None
—hub_always_push: False
—hub_revision: None
—gradient_checkpointing: True
—gradient_checkpointing_kwargs: None
—include_inputs_for_metrics: False
—include_for_metrics: []
—eval_do_concat_batches: True
—fp16_backend: auto
—push_to_hub_model_id: None
—push_to_hub_organization: None
—mp_parameters:
—auto_find_batch_size: False
—full_determinism: False
—torchdynamo: None
—ray_scope: last
—ddp_timeout: 1800
—torch_compile: True
—torch_compile_backend: inductor
—torch_compile_mode: max-autotune
—include_tokens_per_second: False
—include_num_input_tokens_seen: no
—neftune_noise_alpha: None
—optim_target_modules: None
—batch_eval_metrics: False
—eval_on_start: True
—use_liger_kernel: False
—liger_kernel_config: None
—eval_use_gather_object: False
—average_tokens_across_devices: True
—prompts: None
—batch_sampler: no_duplicates
—multi_dataset_batch_sampler: proportional
—router_mapping: {}
—learning_rate_mapping: {}

</details>

Training Logs

Epoch	Step	Training Loss	Validation Loss	ndcg@10
0	0	-	3.7029	0.1171
0.0002	1	1.8166	-	-
0.1594	1000	0.2008	-	-
0.3188	2000	0.0206	-	-
0.4782	3000	0.0118	-	-
0.6376	4000	0.0085	-	-
0.7969	5000	0.0066	-	-
0.9563	6000	0.0055	-	-
1.0	6274	-	1.7016	0.6775
1.1157	7000	0.0048	-	-
1.2751	8000	0.0042	-	-
1.4345	9000	0.0037	-	-
1.5939	10000	0.0035	-	-
1.7533	11000	0.0033	-	-
1.9127	12000	0.0031	-	-
2.0	12548	-	1.6824	0.6840
2.0720	13000	0.0029	-	-
2.2314	14000	0.0027	-	-
2.3908	15000	0.0026	-	-
2.5502	16000	0.0025	-	-
2.7096	17000	0.0024	-	-
2.8690	18000	0.0024	-	-
3.0	18822	-	1.6982	0.6873

—The bold row denotes the saved checkpoint.

Environmental Impact

Carbon emissions were measured using CodeCarbon.

—Energy Consumed: 19.343 kWh
—Carbon Emitted: 3.970 kg of CO2
—Hours Used: 32.183 hours

Training Hardware

—On Cloud: No
—GPU Model: 2 x NVIDIA GeForce RTX 3090
—CPU Model: AMD Ryzen 7 3700X 8-Core Processor
—RAM Size: 62.70 GB

Framework Versions

—Python: 3.13.7
—Sentence Transformers: 5.1.2
—Transformers: 4.57.1
—PyTorch: 2.9.0+cu128
—Accelerate: 1.11.0
—Datasets: 4.4.1
—Tokenizers: 0.22.1

Citation

BibTeX

Sentence Transformers

bibtex

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

NV-Retriever

bibtex

@misc{moreira2025nvretrieverimprovingtextembedding,
      title={NV-Retriever: Improving text embedding models with effective hard-negative mining}, 
      author={Gabriel de Souza P. Moreira and Radek Osmulski and Mengyao Xu and Ronay Ak and Benedikt Schifferer and Even Oldridge},
      year={2025},
      eprint={2407.15831},
      archivePrefix={arXiv},
      primaryClass={cs.IR},
      url={https://arxiv.org/abs/2407.15831}, 
}

<!--

Glossary

Clearly define terms in order to be accessible across audiences. -->

<!--

Model Card Authors

Lists the people who create the model card, providing recognition and accountability for the detailed work that goes into its construction. -->

<!--

Model Card Contact

Provides a way for people who have updates to the Model Card, suggestions, or questions, to contact the Model Card authors. -->