mmcontext-pubmedbert-gs10k

MMContext Model Information

This model uses a custom MMContextEncoder architecture for multimodal embedding generation, combining text and omics data representations.

⚠️ Important: Loading Instructions

This model requires trust_remote_code=True to load properly.

from sentence_transformers import SentenceTransformer

# ✅ CORRECT: Load with trust_remote_code=True
model = SentenceTransformer('jo-mengr/mmcontext-pubmedbert-gs10k-cxg', trust_remote_code=True)

# Generate embeddings
texts = ["Cell type annotation", "Another description"]
embeddings = model.encode(texts)
print(f"Embeddings shape: {embeddings.shape}")

Model Details

• Architecture: MMContextEncoder (custom multimodal architecture)
• Text Encoder: NeuML/pubmedbert-base-embeddings
• Omics Embedding Method: gs10k
• Output Dimension: 2048
• Pooling Strategy: mean

Omics Embedding Method: GS10K

Gene Set enrichment-based embeddings (10k genes)

Usage Tutorial

📓 Tutorial Notebook: usage_tutorial.ipynb - Detailed usage examples and best practices

Model Architecture

The MMContextEncoder combines:

• Text Branch: NeuML/pubmedbert-base-embeddings with optional adapter layers
• Omics Branch: Lookup-based encoder with precomputed gs10k embeddings
• Adapters: Feed-forward projection layers for dimensionality alignment
• Pooling: mean pooling for sentence-level embeddings

Files in this Repository

• mmcontextencoder.py: Main model implementation
• adapters.py: Adapter modules for dimensionality mapping
• omicsencoder.py: Omics data encoder
• onehot.py: One-hot text encoder
• file_utils.py: Utility functions
• usage_tutorial.ipynb: Tutorial notebook with usage examples

Training Details

• Text Encoder: NeuML/pubmedbert-base-embeddings
• Embedding Method: gs10k
• Output Dimension: 2048
• Training Datasets: 2 datasets
• Text-only Datasets: 0 (None)
• Numeric Datasets: 2 (cellxgene_pseudo_bulk_full, geo_half)
• Batch Size: 512
• Learning Rate: 2e-05
• Training Epochs: 16

---

This model was trained using the MMContext framework for multimodal single-cell analysis.

---

SentenceTransformer based on NeuML/pubmedbert-base-embeddings

This is a sentence-transformers model finetuned from NeuML/pubmedbert-base-embeddings on the cellxgene_pseudo_bulk_full_cell_sentence_1_caption and geo_half_cell_sentence_1_caption datasets. It maps sentences & paragraphs to a 2048-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 <!-- at revision d6eaca8254bc229f3ca42749a5510ae287eb3486 -->
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 2048 dimensions
• Similarity Function: Cosine Similarity
• Training Datasets:
  • cellxgene_pseudo_bulk_full_cell_sentence_1_caption
  • geo_half_cell_sentence_1_caption
• Language: code

<!-- - **License:** Unknown -->

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=1024, bias=True)
        (1): ReLU(inplace=True)
        (2): Linear(in_features=1024, out_features=2048, bias=True)
        (3): BatchNorm1d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (pooling): Pooling({'word_embedding_dimension': 2048, '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})
    (omics_adapter): AdapterModule(
      (net): Sequential(
        (0): Linear(in_features=10000, out_features=1024, bias=True)
        (1): ReLU(inplace=True)
        (2): Linear(in_features=1024, out_features=2048, bias=True)
        (3): BatchNorm1d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (omics_encoder): MiniOmicsModel(
      (embeddings): Embedding(726794, 10000, padding_idx=0)
    )
  )
)

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("sentence_transformers_model_id")
# Run inference
sentences = [
    'sample_idx:census_367b55f4-d543-49aa-90e8-4765fcb8c687_969',
    "This measurement was conducted with 10x 3' v3. Neuron cell type from the thalamic complex, specifically the centromedian and parafasicular nuclei (CM and Pf), derived from a 42-year old male.",
    "This measurement was conducted with 10x 3' v3. Neuron cell type from a 42-year-old male, specifically from the thalamic complex with thalamic excitatory supercluster term, corresponding to the Thalamus (THM) - intralaminar nuclear complex (ILN) - posterior group of intralaminar nuclei (PILN) - centromedian and parafasicular nuclei - CM and Pf dissection.",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 2048]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.5200, 0.4699],
#         [0.5200, 1.0000, 0.8290],
#         [0.4699, 0.8290, 1.0000]])

<!-- ### 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

Triplet

• Datasets: cellxgene_pseudo_bulk_full_cell_sentence_1_caption and geo_half_cell_sentence_1_caption
• Evaluated with <code>TripletEvaluator</code>

<!-- ## 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 Datasets

cellxgene_pseudo_bulk_full_cell_sentence_1_caption

• Dataset: cellxgene_pseudo_bulk_full_cell_sentence_1_caption at 55717c1
• Size: 306,003 training samples
• Columns: <code>anchor</code>, <code>positive</code>, <code>negative_1</code>, and <code>negative_2</code>
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative_1	negative_2
  type	string	string	string	string
  details	<ul><li>min: 56 characters</li><li>mean: 58.69 characters</li><li>max: 60 characters</li></ul>	<ul><li>min: 22 tokens</li><li>mean: 47.33 tokens</li><li>max: 165 tokens</li></ul>	<ul><li>min: 22 tokens</li><li>mean: 49.45 tokens</li><li>max: 120 tokens</li></ul>	<ul><li>min: 56 characters</li><li>mean: 58.63 characters</li><li>max: 59 characters</li></ul>

• Samples:

  anchor	positive	negative_1	negative_2
  <code>sample_idx:census_9d5df009-eb76-43a3-b6cd-22017cc53700_231</code>	<code>This measurement was conducted with 10x 3' v3. Gut endothelial cell derived from proximal colon of a male human fetus at 13th week post-fertilization stage.</code>	<code>This measurement was conducted with 10x 3' v3. Mesothelial cell derived from the proximal colon of a male human at 23rd week post-fertilization stage.</code>	<code>sample_idx:census_9d5df009-eb76-43a3-b6cd-22017cc53700_521</code>
  <code>sample_idx:census_367b55f4-d543-49aa-90e8-4765fcb8c687_132</code>	<code>This measurement was conducted with 10x 3' v3. Sample is an oligodendrocyte cell from a 29-year-old male human, specifically from the thalamic complex, with European self-reported ethnicity.</code>	<code>This measurement was conducted with 10x 3' v3. Neuron cell type from the thalamic complex, specifically the centromedian and parafasicular nuclei (CM and Pf), derived from a 42-year old male human donor.</code>	<code>sample_idx:census_367b55f4-d543-49aa-90e8-4765fcb8c687_134</code>
  <code>sample_idx:census_1e6a6ef9-7ec9-4c90-bbfb-2ad3c3165fd1_9964</code>	<code>This measurement was conducted with Smart-seq2. Neutrophil cell type derived from the lung tissue of a 37-year old male with advanced stage non-small cell lung cancer (NSCLC), stage IV, who has never smoked. The cells exhibit an ALK mutation, with no mutations detected in BRAF, EGFR, ERBB2, KRAS, ROS, or TP53.</code>	<code>This measurement was conducted with 10x 3' v2. Myeloid cell derived from the lung tissue of a 65-year old male, located in normal adjacent tissue, with advanced non-small cell lung cancer (NSCLC), stage III.</code>	<code>sample_idx:census_1e6a6ef9-7ec9-4c90-bbfb-2ad3c3165fd1_972</code>

• Loss: <code>mmcontext.utils.PerDatasetLossLogger</code>

geo_half_cell_sentence_1_caption

• Dataset: geo_half_cell_sentence_1_caption at bc13ae5
• Size: 348,046 training samples
• Columns: <code>anchor</code>, <code>positive</code>, <code>negative_1</code>, and <code>negative_2</code>
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative_1	negative_2
  type	string	string	string	string
  details	<ul><li>min: 20 characters</li><li>mean: 20.0 characters</li><li>max: 20 characters</li></ul>	<ul><li>min: 17 tokens</li><li>mean: 36.82 tokens</li><li>max: 130 tokens</li></ul>	<ul><li>min: 19 tokens</li><li>mean: 34.25 tokens</li><li>max: 88 tokens</li></ul>	<ul><li>min: 20 characters</li><li>mean: 20.0 characters</li><li>max: 20 characters</li></ul>

• Samples:

  anchor	positive	negative_1	negative_2
  <code>sample_idx:SRX173216</code>	<code>This measurement was conducted with Illumina HiSeq 2000. B-cells from individual GM12004, assayed using global run-on technique. These are primary cells, with no reported treatment.</code>	<code>This measurement was conducted with Illumina HiSeq 2000. 48 hour Activin treatment of H1 embryonic stem cells.</code>	<code>sample_idx:SRX189728</code>
  <code>sample_idx:SRX185041</code>	<code>This measurement was conducted with Illumina HiSeq 2000. 1000 ng of fragmented total RNA from a cultured chronic myelogenous leukemia (CML) cell line, specifically the human CML cell line K-562. This cell line is derived from a female hematological system disease, specifically a lymphoid neoplasm (leukemia) known as C.M.L., which is a type of neoplasm affecting the bone marrow. The sample has not undergone any treatment.</code>	<code>This measurement was conducted with Illumina HiSeq 2000. 1000 ng of fragmented total RNA from a cultured female human Chronic Myelogenous Leukemia (CML) cell line, K-562, which was grown in tissue culture. The sample has not received any treatment.</code>	<code>sample_idx:SRX185051</code>
  <code>sample_idx:SRX185046</code>	<code>This measurement was conducted with Illumina HiSeq 2000. 1000 ng of fragmented total RNA from a cultured female human Chronic Myelogenous Leukemia (CML) cell line, K-562, which was grown in tissue culture. The sample has not received any treatment.</code>	<code>This measurement was conducted with Illumina HiSeq 2000. 1000 ng of fragmented total RNA from a cultured chronic myelogenous leukemia (CML) cell line, specifically the human CML cell line K-562. This cell line is derived from a female hematological system disease, specifically a lymphoid neoplasm (leukemia) known as C.M.L., which is a type of neoplasm affecting the bone marrow. The sample has not undergone any treatment.</code>	<code>sample_idx:SRX185051</code>

• Loss: <code>mmcontext.utils.PerDatasetLossLogger</code>

Evaluation Datasets

cellxgene_pseudo_bulk_full_cell_sentence_1_caption

• Dataset: cellxgene_pseudo_bulk_full_cell_sentence_1_caption at 55717c1
• Size: 33,937 evaluation samples
• Columns: <code>anchor</code>, <code>positive</code>, <code>negative_1</code>, and <code>negative_2</code>
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative_1	negative_2
  type	string	string	string	string
  details	<ul><li>min: 56 characters</li><li>mean: 58.7 characters</li><li>max: 60 characters</li></ul>	<ul><li>min: 21 tokens</li><li>mean: 47.32 tokens</li><li>max: 147 tokens</li></ul>	<ul><li>min: 21 tokens</li><li>mean: 44.03 tokens</li><li>max: 88 tokens</li></ul>	<ul><li>min: 56 characters</li><li>mean: 58.76 characters</li><li>max: 60 characters</li></ul>

• Samples:

  anchor	positive	negative_1	negative_2
  <code>sample_idx:census_7db0c178-b0a4-442f-ba54-e9e1633a84bb_763</code>	<code>This measurement was conducted with 10x 3' v3. Oligodendrocyte cell sample taken from the cerebral cortex (Cx) of a 42-year-old male, specifically from the human A43 region.</code>	<code>This measurement was conducted with 10x 3' v3. Neuron cell type from a 50-year old male, specifically an MGE interneuron, located in the cerebral cortex, parietal operculum, gustatory cortex, A43 region.</code>	<code>sample_idx:census_7db0c178-b0a4-442f-ba54-e9e1633a84bb_541</code>
  <code>sample_idx:census_1e6a6ef9-7ec9-4c90-bbfb-2ad3c3165fd1_18907</code>	<code>This measurement was conducted with 10x 3' v2. Endothelial cell, specifically a vein endothelial cell, derived from normal adjacent lung tissue of a 71-year-old female patient with early stage NSCLC (stage I) who has a history of smoking.</code>	<code>This measurement was conducted with 10x 3' v2. Endothelial cell derived from the lymphatic vessel of a 69-year-old male with early stage non-small cell lung cancer (NSCLC), stage II. The patient has a history of smoking and the cell was obtained from the primary tumor site.</code>	<code>sample_idx:census_1e6a6ef9-7ec9-4c90-bbfb-2ad3c3165fd1_16402</code>
  <code>sample_idx:census_fd072bc3-2dfb-46f8-b4e3-467cb3223182_3695</code>	<code>This measurement was conducted with 10x 3' v2. Endothelial cells collected from the spleen of a male human fetus at 15 weeks post-fertilization.</code>	<code>This measurement was conducted with 10x 5' v1. A native cell from the skin of a female human fetus at 11 weeks post-fertilization, identified as a doublet of endothelial and erythrocyte lineage.</code>	<code>sample_idx:census_fd072bc3-2dfb-46f8-b4e3-467cb3223182_6661</code>

• Loss: <code>mmcontext.utils.PerDatasetLossLogger</code>

geo_half_cell_sentence_1_caption

• Dataset: geo_half_cell_sentence_1_caption at bc13ae5
• Size: 38,807 evaluation samples
• Columns: <code>anchor</code>, <code>positive</code>, <code>negative_1</code>, and <code>negative_2</code>
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative_1	negative_2
  type	string	string	string	string
  details	<ul><li>min: 20 characters</li><li>mean: 20.19 characters</li><li>max: 21 characters</li></ul>	<ul><li>min: 16 tokens</li><li>mean: 37.63 tokens</li><li>max: 119 tokens</li></ul>	<ul><li>min: 16 tokens</li><li>mean: 54.71 tokens</li><li>max: 111 tokens</li></ul>	<ul><li>min: 20 characters</li><li>mean: 20.04 characters</li><li>max: 21 characters</li></ul>

• Samples:

  anchor	positive	negative_1	negative_2
  <code>sample_idx:SRX185061</code>	<code>This measurement was conducted with Illumina HiSeq 2000. 1000 ng of fragmented total RNA from a cultured chronic myelogenous leukemia (CML) cell line (K-562) derived from a female hematological system disease (CML). The cells were grown in tissue culture and have undergone Ribo-Zero treatment.</code>	<code>This measurement was conducted with Illumina HiSeq 1000. The sample is a cell line (OCI-LY1) derived from a diffuse large B-cell lymphoma (DLBCL), a type of non-Hodgkin lymphoma that affects the lymphatic system. The cells have been treated with siNT (a non-coding siRNA) for 48 hours.</code>	<code>sample_idx:SRX188847</code>
  <code>sample_idx:SRX185895</code>	<code>This measurement was conducted with Illumina HiSeq 1000. The sample is a cell line (OCI-LY1) derived from a diffuse large B-cell lymphoma (DLBCL), a type of non-Hodgkin lymphoma that affects the lymphatic system. The cells have been treated with siNT (a non-coding siRNA) for 48 hours.</code>	<code>This measurement was conducted with Illumina HiSeq 2000. 1000 ng of fragmented total RNA from a cultured chronic myelogenous leukemia (CML) cell line (K-562) derived from a female hematological system disease (CML). The cells were grown in tissue culture and have undergone Ribo-Zero treatment.</code>	<code>sample_idx:SRX188847</code>
  <code>sample_idx:SRX188847</code>	<code>This measurement was conducted with Illumina HiSeq 2000. 48-hour Activin-treated H1 embryonic stem cells.</code>	<code>This measurement was conducted with Illumina HiSeq 2000. 1000 ng of fragmented total RNA from a cultured chronic myelogenous leukemia (CML) cell line (K-562) derived from a female hematological system disease (CML). The cells were grown in tissue culture and have undergone Ribo-Zero treatment.</code>	<code>sample_idx:SRX185895</code>

• Loss: <code>mmcontext.utils.PerDatasetLossLogger</code>

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 512
• per_device_eval_batch_size: 512
• learning_rate: 2e-05
• num_train_epochs: 16
• warmup_ratio: 0.1
• bf16: True

All Hyperparameters

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

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 512
• per_device_eval_batch_size: 512
• 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: 2e-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: 16
• 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
• 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}
• parallelism_config: None
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch_fused
• optim_args: None
• 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: 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: False
• 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: no
• 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: True
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

</details>

Training Logs

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

Framework Versions

• Python: 3.12.8
• Sentence Transformers: 5.1.2
• Transformers: 4.57.2
• PyTorch: 2.9.1+cu128
• Accelerate: 1.12.0
• Datasets: 3.6.0
• Tokenizers: 0.22.1

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",
}

<!-- ## 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.* -->