SentenceTransformer based on NeuML/pubmedbert-base-embeddings

This is a sentence-transformers model finetuned from NeuML/pubmedbert-base-embeddings on the cellxgene_pseudo_bulk_100k_multiplets_natural_language_annotation and geo_70k_multiplets_natural_language_annotation datasets. It maps sentences & paragraphs to a 1024-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: NeuML/pubmedbert-base-embeddings
Maximum Sequence Length: 512 tokens
Output Dimensionality: 1024 dimensions
Similarity Function: Cosine Similarity
Training Datasets:
- cellxgene_pseudo_bulk_100k_multiplets_natural_language_annotation
- geo_70k_multiplets_natural_language_annotation
Language: code

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): MMContextEncoder(
    (text_encoder): BertModel(
      (embeddings): BertEmbeddings(
        (word_embeddings): Embedding(30522, 768, padding_idx=0)
        (position_embeddings): Embedding(512, 768)
        (token_type_embeddings): Embedding(2, 768)
        (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
        (dropout): Dropout(p=0.1, inplace=False)
      )
      (encoder): BertEncoder(
        (layer): ModuleList(
          (0-11): 12 x BertLayer(
            (attention): BertAttention(
              (self): BertSdpaSelfAttention(
                (query): Linear(in_features=768, out_features=768, bias=True)
                (key): Linear(in_features=768, out_features=768, bias=True)
                (value): Linear(in_features=768, out_features=768, bias=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
              (output): BertSelfOutput(
                (dense): Linear(in_features=768, out_features=768, bias=True)
                (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
                (dropout): Dropout(p=0.1, inplace=False)
              )
            )
            (intermediate): BertIntermediate(
              (dense): Linear(in_features=768, out_features=3072, bias=True)
              (intermediate_act_fn): GELUActivation()
            )
            (output): BertOutput(
              (dense): Linear(in_features=3072, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
        )
      )
      (pooler): BertPooler(
        (dense): Linear(in_features=768, out_features=768, bias=True)
        (activation): Tanh()
      )
    )
    (text_adapter): AdapterModule(
      (net): Sequential(
        (0): Linear(in_features=768, out_features=512, bias=True)
        (1): ReLU(inplace=True)
        (2): Linear(in_features=512, out_features=1024, bias=True)
        (3): BatchNorm1d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (pooling): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  )
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("jo-mengr/mmcontext-pubmedbert-v2")
# Run inference
sentences = [
    'census_1b9d8702-5af8-4142-85ed-020eb06ec4f6_20229',
    "This measurement was conducted with 10x 3' v3. Terminally differentiated CD8+ T cells from the lung tissue of a male individual in his sixties.",
    "This measurement was conducted with 10x 5' v2. Sample contains regulatory T cells (Tregs), specifically T cells, from a female individual in her eighth decade, isolated from a thoracic lymph node.",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.4047, 0.4202],
#         [0.4047, 1.0000, 0.9129],
#         [0.4202, 0.9129, 1.0000]])

Evaluation

Metrics

Triplet

Datasets: cellxgene_pseudo_bulk_100k_multiplets_natural_language_annotation_cell_sentence_1 and geo_70k_multiplets_natural_language_annotation_cell_sentence_1
Evaluated with TripletEvaluator

Metric	cellxgene_pseudo_bulk_100k_multiplets_natural_language_annotation_cell_sentence_1	geo_70k_multiplets_natural_language_annotation_cell_sentence_1
cosine_accuracy	0.5292	0.7119

Training Details

Training Datasets

cellxgene_pseudo_bulk_100k_multiplets_natural_language_annotation

Dataset: cellxgene_pseudo_bulk_100k_multiplets_natural_language_annotation at b141493
Size: 81,143 training samples
Columns: anchor, positive, negative_1, and negative_2

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative_1	negative_2
type	string	string	string	string
details	min: 45 characters mean: 47.72 characters max: 49 characters	min: 92 characters mean: 216.13 characters max: 900 characters	min: 101 characters mean: 215.14 characters max: 870 characters	min: 45 characters mean: 47.75 characters max: 49 characters

Samples:

anchor	positive	negative_1	negative_2
`census_218acb0f-9f2f-4f76-b90b-15a4b7c7f629_26009`	`This measurement was conducted with 10x 3' v2. A proliferating lymphocyte cell sample, obtained from a 34-year-old female Asian individual, derived from peripheral blood mononuclear cells.`	`This measurement was conducted with 10x 3' v2. Sample is a 25-year-old female with European ethnicity, having CD8-positive, alpha-beta T cell type. This cell type exhibits elevated expression of type 1 interferon-stimulated genes (ISGs) in monocytes, reduction of naïve CD4+ T cells correlating with monocyte ISG expression, and expansion of repertoire-restricted cytotoxic GZMH+ CD8+ T cells.`	`census_218acb0f-9f2f-4f76-b90b-15a4b7c7f629_14165`
`census_1b9d8702-5af8-4142-85ed-020eb06ec4f6_6333`	`This measurement was conducted with 10x 5' v1. Sample is a cell from the omentum tissue, specifically an effector memory CD4-positive, alpha-beta T cell, from a female in her sixth decade.`	`This measurement was conducted with 10x 5' v2. Conventional dendritic cell from the jejunal epithelium of a female in her eighth decade.`	`census_1b9d8702-5af8-4142-85ed-020eb06ec4f6_2714`
`census_adda0684-f8ea-4403-b393-2a25607077c4_271`	`This measurement was conducted with 10x 3' v3. Neuron cell type from a 29-year-old male, specifically from the thalamic complex, specifically the thalamus (THM) - posterior nuclear complex of thalamus (PoN) - medial geniculate nuclei (MG).`	`This measurement was conducted with 10x 3' v3. Neuron from the thalamic complex (thalamus, posterior nuclear complex of thalamus, medial geniculate nuclei) of a 42-year-old male, identified as a midbrain-derived inhibitory neuron.`	`census_adda0684-f8ea-4403-b393-2a25607077c4_425`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim"
}

geo_70k_multiplets_natural_language_annotation

Dataset: geo_70k_multiplets_natural_language_annotation at 4c62cd1
Size: 61,911 training samples
Columns: anchor, positive, negative_1, and negative_2

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative_1	negative_2
type	string	string	string	string
details	min: 9 characters mean: 9.29 characters max: 10 characters	min: 83 characters mean: 189.5 characters max: 698 characters	min: 100 characters mean: 165.46 characters max: 465 characters	min: 9 characters mean: 9.1 characters max: 10 characters

Samples:

anchor	positive	negative_1	negative_2
`SRX083304`	`This measurement was conducted with Illumina HiSeq 2000. 5-day HeLa cell line with ELAVL1/HuR siRNA1 knockdown, 120 hours post-transfection.`	`This measurement was conducted with Illumina HiSeq 2000. BJ fibroblast cells in a proliferative stage, with polyA RNA subtype.`	`SRX105303`
`SRX105302`	`This measurement was conducted with Illumina HiSeq 2000. BJ fibroblast cells in a proliferative stage, with polyA RNA subtype.`	`This measurement was conducted with Illumina HiSeq 2000. 5-day HeLa cell line with ELAVL1/HuR siRNA1 knockdown, 120 hours post-transfection.`	`SRX105303`
`SRX105303`	`This measurement was conducted with Illumina HiSeq 2000. BJ fibroblast cells at a confluent growth stage, with polyA RNA subtype.`	`This measurement was conducted with Illumina HiSeq 2000. 5-day HeLa cell line with ELAVL1/HuR siRNA1 knockdown, 120 hours post-transfection.`	`SRX105302`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim"
}

Evaluation Datasets

cellxgene_pseudo_bulk_100k_multiplets_natural_language_annotation

Dataset: cellxgene_pseudo_bulk_100k_multiplets_natural_language_annotation at b141493
Size: 9,011 evaluation samples
Columns: anchor, positive, negative_1, and negative_2

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative_1	negative_2
type	string	string	string	string
details	min: 45 characters mean: 47.73 characters max: 49 characters	min: 99 characters mean: 209.99 characters max: 941 characters	min: 102 characters mean: 213.87 characters max: 981 characters	min: 45 characters mean: 47.73 characters max: 49 characters

Samples:

anchor	positive	negative_1	negative_2
`census_0b4a15a7-4e9e-4555-9733-2423e5c66469_490`	`This measurement was conducted with 10x 3' v3. Cell sample from the cortex of kidney, taken from a 43-year-old male of European ethnicity with a reported history of kidney cancer. The cell type is identified as a kidney collecting duct intercalated cell.`	`This measurement was conducted with 10x 3' v3. Kidney collecting duct intercalated cell from a 43-year old European male with kidney cancer, taken from the cortex of kidney and cryopreserved for further analysis.`	`census_0b4a15a7-4e9e-4555-9733-2423e5c66469_9`
`census_4976b234-9028-4b4b-8a2f-8ac59d636219_269`	`This measurement was conducted with 10x 3' v3. Neuron cell type from a 29-year-old male cerebellum, specifically from the Cerebellar Vermis - CBV region, with European self-reported ethnicity, analyzed at the nucleus level.`	`This measurement was conducted with 10x 3' v3. Endothelial cells derived from the cerebellum (specifically, cerebellar vermis) of a 42-year-old male, classified under the vascular supercluster term.`	`census_4976b234-9028-4b4b-8a2f-8ac59d636219_923`
`census_44882825-0da1-4547-b721-2c6105d4a9d1_10258`	`This measurement was conducted with 10x 5' v1. Cell sample from the tonsil of a 9-year-old female with recurrent tonsillitis, characterized as a centroblast B cell with IGLC2, IGLV7-43, IGLJ3 immunoglobulin genes expressed.`	`This measurement was conducted with 10x 5' v1. Centroblast cells derived from a 3-year-old male human tonsil sample, with obstructive sleep apnea and recurrent tonsillitis, undergoing affinity maturation and differentiation into memory or plasma cells.`	`census_44882825-0da1-4547-b721-2c6105d4a9d1_9654`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim"
}

geo_70k_multiplets_natural_language_annotation

Dataset: geo_70k_multiplets_natural_language_annotation at 4c62cd1
Size: 6,901 evaluation samples
Columns: anchor, positive, negative_1, and negative_2

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative_1	negative_2
type	string	string	string	string
details	min: 9 characters mean: 10.35 characters max: 11 characters	min: 78 characters mean: 191.46 characters max: 983 characters	min: 90 characters mean: 217.63 characters max: 702 characters	min: 10 characters mean: 10.04 characters max: 11 characters

Samples:

anchor	positive	negative_1	negative_2
`SRX2244363`	`This measurement was conducted with Illumina HiSeq 2000. 15-year-old male HepG2 immortalized cell line with hepatocellular carcinoma, transiently expressing shRNA targeting PKM2 for RNA-seq study.`	`This measurement was conducted with Illumina HiSeq 2000. 15-year-old male patient with hepatocellular carcinoma; HNRNPC knocked down via shRNA in HepG2 (immortalized cell line) for RNA-seq analysis.`	`SRX5457055`
`SRX3136447`	`This measurement was conducted with Illumina HiSeq 2000. 16-year-old female's T cells from a control group, stimulated with ag85 at timepoint 0, and primary cells.`	`This measurement was conducted with Illumina HiSeq 2000. 17-year-old male's monocytes stimulated with mTb, taken at 180 days post-stimulation, as part of the control group in a study.`	`SRX3137689`
`SRX2734845`	`This measurement was conducted with Illumina HiSeq 2500. UM-UC18 bladder cancer cell line, a type of urinary bladder cancer cell line, cultured for study of bladder disease, cancer cell proliferation, and neoplasm.`	`This measurement was conducted with NextSeq 500. HeLa cells with PARP knockdown treatment.`	`SRX3130770`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim"
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 256
per_device_eval_batch_size: 256
learning_rate: 0.05
num_train_epochs: 4
warmup_ratio: 0.1
bf16: True
gradient_checkpointing: True

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
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: None
learning_rate: 0.05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 4
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.1
warmup_steps: 0
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: 42
data_seed: None
jit_mode_eval: False
use_ipex: False
bf16: True
fp16: False
fp16_opt_level: O1
half_precision_backend: auto
bf16_full_eval: False
fp16_full_eval: False
tf32: None
local_rank: 0
ddp_backend: None
tpu_num_cores: None
tpu_metrics_debug: False
debug: []
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_prefetch_factor: None
past_index: -1
disable_tqdm: False
remove_unused_columns: True
label_names: None
load_best_model_at_end: False
ignore_data_skip: False
fsdp: []
fsdp_min_num_params: 0
fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
fsdp_transformer_layer_cls_to_wrap: None
accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch
optim_args: None
adafactor: False
group_by_length: False
length_column_name: length
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
dataloader_pin_memory: True
dataloader_persistent_workers: False
skip_memory_metrics: True
use_legacy_prediction_loop: False
push_to_hub: False
resume_from_checkpoint: None
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: False
torch_compile_backend: None
torch_compile_mode: None
include_tokens_per_second: False
include_num_input_tokens_seen: False
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
eval_on_start: False
use_liger_kernel: False
liger_kernel_config: None
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	cellxgene pseudo bulk 100k multiplets natural language annotation loss	geo 70k multiplets natural language annotation loss	cellxgene_pseudo_bulk_100k_multiplets_natural_language_annotation_cell_sentence_1_cosine_accuracy	geo_70k_multiplets_natural_language_annotation_cell_sentence_1_cosine_accuracy
0.1789	100	5.4042	12.7258	18.5172	0.5005	0.4127
0.3578	200	4.4018	21.3994	27.5368	0.5012	0.4662
0.5367	300	4.274	13.9052	15.7111	0.5054	0.5134
0.7156	400	3.9977	17.2145	18.8384	0.5060	0.6522
0.8945	500	3.8001	18.0511	20.1693	0.5058	0.3982
1.0733	600	3.7527	15.4862	20.6695	0.5064	0.4026
1.2522	700	3.7414	15.5879	13.7452	0.5089	0.3620
1.4311	800	3.4425	14.7486	11.9465	0.5069	0.5199
1.6100	900	3.3452	13.9171	14.1143	0.5113	0.4123
1.7889	1000	3.2576	15.2234	12.7155	0.5143	0.4120
1.9678	1100	3.322	14.6208	10.3553	0.5262	0.4456
2.1467	1200	3.1823	12.7034	12.2282	0.5236	0.5434
2.3256	1300	3.1449	11.2867	9.9116	0.5292	0.6111
2.5045	1400	3.0859	10.7462	9.3380	0.5349	0.6641
2.6834	1500	3.0582	12.2004	8.9558	0.5298	0.6866
2.8623	1600	2.9614	11.8808	8.8887	0.5317	0.6728
3.0411	1700	3.008	12.4199	8.4042	0.5290	0.6911
3.2200	1800	2.9739	10.9099	8.9717	0.5379	0.6656
3.3989	1900	2.9152	11.6201	8.5289	0.5314	0.6954
3.5778	2000	2.9668	11.9039	8.4831	0.5318	0.6860
3.7567	2100	3.0303	11.2059	8.9941	0.5368	0.6696
3.9356	2200	2.936	12.0965	7.9045	0.5292	0.7119

Framework Versions

Python: 3.11.6
Sentence Transformers: 5.0.0
Transformers: 4.55.0.dev0
PyTorch: 2.5.1+cu121
Accelerate: 1.9.0
Datasets: 2.19.1
Tokenizers: 0.21.4

Citation

BibTeX

Sentence Transformers

@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",
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for jo-mengr/mmcontext-pubmedbert-v2

Base model

microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract-fulltext

Finetuned

NeuML/pubmedbert-base-embeddings

Finetuned

(77)

this model

Evaluation results

Cosine Accuracy on cellxgene pseudo bulk 100k multiplets natural language annotation cell sentence 1
self-reported

0.529
Cosine Accuracy on geo 70k multiplets natural language annotation cell sentence 1
self-reported

0.712

View on Papers With Code