bge-base-financial-matryoshka / README.md

phucvt0302

Add new SentenceTransformer model

c19723e verified 5 months ago

preview code

raw

history blame contribute delete

30.1 kB

metadata

language:
  - en
license: apache-2.0
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:6300
  - loss:MatryoshkaLoss
  - loss:MultipleNegativesRankingLoss
base_model: BAAI/bge-base-en-v1.5
widget:
  - source_sentence: >-
      What is the anticipated total capital investment range for fiscal year
      2024 related to property and equipment?
    sentences:
      - >-
        Apollo, which we began offering as a commercial solution in 2021, is a
        cloud-agnostic, single control layer that coordinates ongoing delivery
        of new features, security updates, and platform configurations, helping
        to ensure the continuous operation of critical systems.
      - >-
        During fiscal year 2024, we expect to use our existing cash and cash
        equivalents, our marketable securities, and the cash generated by our
        operations to fund our capital investments of approximately $1.10
        billion to $1.30 billion related to property and equipment.
      - >-
        Joseph F. Wayland was appointed as General Counsel and Secretary of
        Chubb Limited in July 2013.
  - source_sentence: >-
      What was the effective price per share of class A common stock for fiscal
      2023 under the U.S. retrospective responsibility plan?
    sentences:
      - We operated 692 gas stations at the end of 2023.
      - >-
        The effective price per share for class A common stock under the U.S.
        retrospective responsibility plan for fiscal 2023 was $221.33,
        calculated using the weighted-average price based on the volume-weighted
        average price during the pricing period.
      - >-
        For a description of our material pending legal proceedings, see Legal
        Matters in Array 10 of the Notes to Consolidated Financial Statements
        included in Part II, Item 8 of this Annual Report on...
  - source_sentence: >-
      What financial challenge did the company face in 2017 and how did it
      impact them legally?
    sentences:
      - >-
        In 2017, we experienced a material cybersecurity incident following a
        criminal attack on our systems that involved the theft of personal
        information of consumers. As a result of the 2017 cybersecurity
        incident, we were subject to proceedings and investigations.
      - >-
        The overall net effect on our gross margin from inventory provisions and
        sales of items previously written down was an unfavorable impact of 7.5%
        in fiscal year 2023 and 0.9% in fiscal year 2022.
      - >-
        The adjusted after-tax return on invested capital (ROIC) is computed by
        dividing the after-tax operating profit, excluding rent expenses, by the
        total invested capital which factors in the capitalization of operating
        leases.
  - source_sentence: What is the title of Item 8 in the document?
    sentences:
      - >-
        Research and development expenses for fiscal year 2023 increased by $142
        million over the previous fiscal year.
      - >-
        As of January 28, 2023, the company was authorized to issue up to
        10,000,000 preferred shares, each with a par value of $0.01. However,
        there were no preferred shares issued and outstanding as of January 28,
        2023 and the previous year as well.
      - >-
        Item 8 of the document is titled 'Financial Statements and Supplementary
        Data'.
  - source_sentence: >-
      How many shares were outstanding at the beginning of 2023 and what was
      their aggregate intrinsic value?
    sentences:
      - >-
        At the beginning of 2023, there were 355 shares outstanding with an
        aggregate intrinsic value of $142,916.
      - >-
        In IBM’s 2023 Annual Report to Stockholders, the Financial Statements
        and Supplementary Data are included on pages 44 through 121.
      - >-
        Privacy and data protection regulations affect operations by dictating
        how data is used and handled, impacting product offering and operation.
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - cosine_accuracy@1
  - cosine_accuracy@3
  - cosine_accuracy@5
  - cosine_accuracy@10
  - cosine_precision@1
  - cosine_precision@3
  - cosine_precision@5
  - cosine_precision@10
  - cosine_recall@1
  - cosine_recall@3
  - cosine_recall@5
  - cosine_recall@10
  - cosine_ndcg@10
  - cosine_mrr@10
  - cosine_map@100
model-index:
  - name: BGE base Financial Matryoshka
    results:
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 768
          type: dim_768
        metrics:
          - type: cosine_accuracy@1
            value: 0.7171428571428572
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8414285714285714
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.87
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.9071428571428571
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.7171428571428572
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.28047619047619043
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.174
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09071428571428569
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.7171428571428572
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8414285714285714
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.87
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.9071428571428571
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8148141512407042
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7849903628117916
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7887661106132665
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 512
          type: dim_512
        metrics:
          - type: cosine_accuracy@1
            value: 0.7171428571428572
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8471428571428572
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8671428571428571
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.9085714285714286
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.7171428571428572
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.28238095238095234
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.1734285714285714
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09085714285714284
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.7171428571428572
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8471428571428572
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8671428571428571
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.9085714285714286
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8146547679133024
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7843815192743763
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7879908541996482
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 256
          type: dim_256
        metrics:
          - type: cosine_accuracy@1
            value: 0.7171428571428572
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8414285714285714
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8728571428571429
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.9014285714285715
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.7171428571428572
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.28047619047619043
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.17457142857142854
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09014285714285714
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.7171428571428572
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8414285714285714
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8728571428571429
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.9014285714285715
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.81391710609103
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7854092970521545
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7894965954567308
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 128
          type: dim_128
        metrics:
          - type: cosine_accuracy@1
            value: 0.7128571428571429
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.8242857142857143
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8585714285714285
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.9028571428571428
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.7128571428571429
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.2747619047619047
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.1717142857142857
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09028571428571427
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.7128571428571429
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.8242857142857143
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8585714285714285
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.9028571428571428
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.8071617730563218
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7765963718820862
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7803238233984962
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 64
          type: dim_64
        metrics:
          - type: cosine_accuracy@1
            value: 0.6914285714285714
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.7942857142857143
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.8314285714285714
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.8728571428571429
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.6914285714285714
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.26476190476190475
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.16628571428571426
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.08728571428571427
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.6914285714285714
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.7942857142857143
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.8314285714285714
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.8728571428571429
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.7805313652937041
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.7512346938775507
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.7553555070551027
            name: Cosine Map@100

BGE base Financial Matryoshka

This is a sentence-transformers model finetuned from BAAI/bge-base-en-v1.5 on the json dataset. It maps sentences & paragraphs to a 768-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: BAAI/bge-base-en-v1.5
Maximum Sequence Length: 512 tokens
Output Dimensionality: 768 dimensions
Similarity Function: Cosine Similarity
Training Dataset:
- json
Language: en
License: apache-2.0

Model Sources

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

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': True}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Normalize()
)

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("phucvt0302/bge-base-financial-matryoshka")
# Run inference
sentences = [
    'How many shares were outstanding at the beginning of 2023 and what was their aggregate intrinsic value?',
    'At the beginning of 2023, there were 355 shares outstanding with an aggregate intrinsic value of $142,916.',
    'In IBM’s 2023 Annual Report to Stockholders, the Financial Statements and Supplementary Data are included on pages 44 through 121.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Information Retrieval

Dataset: dim_768
Evaluated with InformationRetrievalEvaluator with these parameters:
```
{
    "truncate_dim": 768
}
```

Metric	Value
cosine_accuracy@1	0.7171
cosine_accuracy@3	0.8414
cosine_accuracy@5	0.87
cosine_accuracy@10	0.9071
cosine_precision@1	0.7171
cosine_precision@3	0.2805
cosine_precision@5	0.174
cosine_precision@10	0.0907
cosine_recall@1	0.7171
cosine_recall@3	0.8414
cosine_recall@5	0.87
cosine_recall@10	0.9071
cosine_ndcg@10	0.8148
cosine_mrr@10	0.785
cosine_map@100	0.7888

Information Retrieval

Dataset: dim_512
Evaluated with InformationRetrievalEvaluator with these parameters:
```
{
    "truncate_dim": 512
}
```

Metric	Value
cosine_accuracy@1	0.7171
cosine_accuracy@3	0.8471
cosine_accuracy@5	0.8671
cosine_accuracy@10	0.9086
cosine_precision@1	0.7171
cosine_precision@3	0.2824
cosine_precision@5	0.1734
cosine_precision@10	0.0909
cosine_recall@1	0.7171
cosine_recall@3	0.8471
cosine_recall@5	0.8671
cosine_recall@10	0.9086
cosine_ndcg@10	0.8147
cosine_mrr@10	0.7844
cosine_map@100	0.788

Information Retrieval

Dataset: dim_256
Evaluated with InformationRetrievalEvaluator with these parameters:
```
{
    "truncate_dim": 256
}
```

Metric	Value
cosine_accuracy@1	0.7171
cosine_accuracy@3	0.8414
cosine_accuracy@5	0.8729
cosine_accuracy@10	0.9014
cosine_precision@1	0.7171
cosine_precision@3	0.2805
cosine_precision@5	0.1746
cosine_precision@10	0.0901
cosine_recall@1	0.7171
cosine_recall@3	0.8414
cosine_recall@5	0.8729
cosine_recall@10	0.9014
cosine_ndcg@10	0.8139
cosine_mrr@10	0.7854
cosine_map@100	0.7895

Information Retrieval

Dataset: dim_128
Evaluated with InformationRetrievalEvaluator with these parameters:
```
{
    "truncate_dim": 128
}
```

Metric	Value
cosine_accuracy@1	0.7129
cosine_accuracy@3	0.8243
cosine_accuracy@5	0.8586
cosine_accuracy@10	0.9029
cosine_precision@1	0.7129
cosine_precision@3	0.2748
cosine_precision@5	0.1717
cosine_precision@10	0.0903
cosine_recall@1	0.7129
cosine_recall@3	0.8243
cosine_recall@5	0.8586
cosine_recall@10	0.9029
cosine_ndcg@10	0.8072
cosine_mrr@10	0.7766
cosine_map@100	0.7803

Information Retrieval

Dataset: dim_64
Evaluated with InformationRetrievalEvaluator with these parameters:
```
{
    "truncate_dim": 64
}
```

Metric	Value
cosine_accuracy@1	0.6914
cosine_accuracy@3	0.7943
cosine_accuracy@5	0.8314
cosine_accuracy@10	0.8729
cosine_precision@1	0.6914
cosine_precision@3	0.2648
cosine_precision@5	0.1663
cosine_precision@10	0.0873
cosine_recall@1	0.6914
cosine_recall@3	0.7943
cosine_recall@5	0.8314
cosine_recall@10	0.8729
cosine_ndcg@10	0.7805
cosine_mrr@10	0.7512
cosine_map@100	0.7554

Training Details

Training Dataset

json

Dataset: json
Size: 6,300 training samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 7 tokens
mean: 20.53 tokens
max: 46 tokens

min: 8 tokens
mean: 44.95 tokens
max: 512 tokens

	anchor	positive
type	string	string
details	min: 7 tokens mean: 20.53 tokens max: 46 tokens	min: 8 tokens mean: 44.95 tokens max: 512 tokens

Samples:

anchor	positive
`What does the No Surprises Act require providers to develop and disclose?`	`Under the No Surprises Act, which went into effect January 1, 2022, certain providers, including DaVita, are required to develop and disclose a 'Good Faith Estimate' that details the expected charges for furnishing certain items or services.`
`What does Gross Merchandise Volume (GMV) represent in financial terms?`	`GMV consists of the total value of all paid transactions between users on our platforms during the applicable period inclusive of shipping fees and taxes.`
`What was the pre-tax restructuring charge for the fiscal year 2023 related to the discontinuation of certain R&D programs?`	`The pre-tax restructuring charge of approximately $0.5 billion in the fiscal year 2023 included the termination of partnered and non-partnered program costs and asset impairments.`

Loss: MatryoshkaLoss with these parameters:

{
    "loss": "MultipleNegativesRankingLoss",
    "matryoshka_dims": [
        768,
        512,
        256,
        128,
        64
    ],
    "matryoshka_weights": [
        1,
        1,
        1,
        1,
        1
    ],
    "n_dims_per_step": -1
}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 32
per_device_eval_batch_size: 16
gradient_accumulation_steps: 16
learning_rate: 2e-05
num_train_epochs: 4
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: True
tf32: True
load_best_model_at_end: True
optim: adamw_torch_fused
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
prediction_loss_only: True
per_device_train_batch_size: 32
per_device_eval_batch_size: 16
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 16
eval_accumulation_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: 4
max_steps: -1
lr_scheduler_type: cosine
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
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: True
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: True
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, 'gradient_accumulation_kwargs': 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
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: False
hub_always_push: False
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_inputs_for_metrics: False
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
dispatch_batches: None
split_batches: None
include_tokens_per_second: False
include_num_input_tokens_seen: False
neftune_noise_alpha: None
optim_target_modules: None
prompts: None
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	dim_768_cosine_ndcg@10	dim_512_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.9697	6	-	0.7993	0.7967	0.7941	0.7845	0.7431
1.6162	10	2.3395	-	-	-	-	-
1.9394	12	-	0.8089	0.8086	0.8108	0.8007	0.7669
2.9091	18	-	0.8158	0.8134	0.8144	0.8066	0.7761
3.2323	20	1.0419	-	-	-	-	-
3.8788	24	-	0.8148	0.8147	0.8139	0.8072	0.7805

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.12.6
Sentence Transformers: 4.1.0
Transformers: 4.40.2
PyTorch: 2.6.0+cu124
Accelerate: 1.6.0
Datasets: 2.19.1
Tokenizers: 0.19.1

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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning},
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

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}
}