README.md

metadata

base_model: Snowflake/snowflake-arctic-embed-xs
language:
  - en
library_name: sentence-transformers
license: apache-2.0
pipeline_tag: sentence-similarity
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:416298
  - loss:MultipleNegativesRankingLoss
widget:
  - source_sentence: >-
      The radial profiles using frank for the seven targets can be seen in
      Figure 6.
    sentences:
      - >-
        At longer radio wavelengths, we selected the newest observations of the
        appropriate resolution from the VLA archive.
      - >-
        The radial profiles using frank for the seven targets can be seen in
        Figure 6.
      - "For further information on observation and data calibration, refer to Hunt et\_al. (2021)."
  - source_sentence: >-
      They are extragalactic scaled up versions of galactic Ultra Compact (UC)
      H ii regions, which are typically excited by a single massive star and are
      ≲less-than-or-similar-to\lesssim 0.1 pc in size (Wood & Churchwell, 1989).
    sentences:
      - >-
        They are extragalactic scaled up versions of galactic Ultra Compact (UC)
        H ii regions, which are typically excited by a single massive star and
        are ≲less-than-or-similar-to\lesssim 0.1 pc in size (Wood & Churchwell,
        1989).
      - >-
        The LMT is a project operated by the Instituto Nacional de Astrófisica,
        Óptica, y Electrónica (Mexico) and the University of Massachusetts at
        Amherst (USA).
      - >-
        We measure the detection confidence in the resolved image as the ratio
        between the local mean posterior and the local posterior standard
        deviation of the estimated circular polarization, evaluated based on
        1000 images drawn from the posterior distribution.
  - source_sentence: >-
      The flux density calibrator was 3C286, and the complex gain calibrator was
      J0836-2016.
    sentences:
      - >-
        The flux density calibrator was 3C286, and the complex gain calibrator
        was J0836-2016.
      - >-
        While rcsubscript𝑟cr_{\rm c} has a clear dependence on
        Dmaxsubscript𝐷maxD_{\rm max}, xMMSNsubscript𝑥MMSNx_{\rm MMSN} and
        tagesubscript𝑡aget_{\rm age}, ΣcsubscriptΣc\Sigma_{\rm c} only has weak
        dependence on Dmaxsubscript𝐷maxD_{\rm max}, and so is mostly sensitive
        to the scaling of the total initial planetesimal mass,
        xMMSNsubscript𝑥MMSNx_{\rm MMSN} and tagesubscript𝑡aget_{\rm age}.
      - "20 is valid only at r=rc𝑟subscript𝑟cr=r_{\\rm c}, it has been shown that the surface density of dust at r>rc𝑟subscript𝑟cr>r_{\\rm c} is expected to be flat for a primordial surface density exponent (−α𝛼-\\alpha) of -3/2, or more generally proportional to r−0.6​α+0.9superscript𝑟0.6𝛼0.9r^{-0.6\\alpha+0.9} (Schüppler et\_al., 2016; Marino et\_al., 2017b; Geiler & Krivov, 2017)."
  - source_sentence: >-
      We would like to thank A. Deller and W. Brisken for EHT-specific support
      with the use of DiFX.
    sentences:
      - >-
        Ice has one of the weakest strengths, and thus if we had assumed
        stronger solids the derived values of Dmaxsubscript𝐷D_{\max} and
        xMMSNsubscript𝑥MMSNx_{\rm MMSN} would be lower.
      - >-
        We would like to thank A. Deller and W. Brisken for EHT-specific support
        with the use of DiFX.
      - >-
        The wsmoothsubscript𝑤smoothw_{\rm smooth} chosen parameter ranged from
        10−2superscript10210^{-2} to 10−4superscript10410^{-4} depending on the
        disc.
  - source_sentence: >-
      New higher resolution images and our parametric modelling confirmed this
      finding.
    sentences:
      - >-
        New higher resolution images and our parametric modelling confirmed this
        finding.
      - >-
        With the 3-bit correlator configuration, we obtained a total bandwidth
        of ∼similar-to\sim8 GHz across Ka-band.
      - >-
        Pan & Schlichting, 2012) and thus could slightly affect the surface
        density slope.

interstellar-ice-crystal-xs

This is a sentence-transformers model finetuned from Snowflake/snowflake-arctic-embed-xs. It maps sentences & paragraphs to a 384-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: Snowflake/snowflake-arctic-embed-xs
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 384 tokens
• Similarity Function: Cosine Similarity
• Training Dataset: scraped astronomy papers at the NLP for Space Science workshop.
• 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': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, '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("SimoneAstarita/interstellar-ice-crystal-xs")
# Run inference
sentences = [
    'New higher resolution images and our parametric modelling confirmed this finding.',
    'New higher resolution images and our parametric modelling confirmed this finding.',
    'Pan & Schlichting, 2012) and thus could slightly affect the surface density slope.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

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

Training Details

Training Dataset

The dataset is made of scrapes papers in astronomy, including abstract, introduction and conclusions. They are divided into sentences using nklt. We then duplicate them and train using the same senrence for positive and anchor. We are using SimSCE.

Unnamed Dataset

• Size: 416,298 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 4 tokens mean: 42.81 tokens max: 512 tokens	min: 4 tokens mean: 42.81 tokens max: 512 tokens

• Samples:

  anchor	positive
  Resolving the inner parsec of the blazar J1924–2914 with the Event Horizon Telescope	Resolving the inner parsec of the blazar J1924–2914 with the Event Horizon Telescope
  The radio source J1924–2914 (PKS 1921–293, OV–236) is a radio-loud quasar at a redshift z=0.353𝑧0.353z=0.353 (Wills & Wills, 1981; Jones et al., 2009).	The radio source J1924–2914 (PKS 1921–293, OV–236) is a radio-loud quasar at a redshift z=0.353𝑧0.353z=0.353 (Wills & Wills, 1981; Jones et al., 2009).
  The source exhibits strong optical variability and is highly polarized (Wills & Wills, 1981; Pica et al., 1988; Worrall & Wilkes, 1990).	The source exhibits strong optical variability and is highly polarized (Wills & Wills, 1981; Pica et al., 1988; Worrall & Wilkes, 1990).

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• warmup_ratio: 0.1
• fp16: True
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: no
• prediction_loss_only: True
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• 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: 5e-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: 3
• 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: False
• fp16: True
• 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: 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
• batch_eval_metrics: False
• eval_on_start: False
• eval_use_gather_object: False
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss
0.0077	100	0.0025
0.0154	200	0.0032
0.0231	300	0.0026
0.0307	400	0.0026
0.0384	500	0.0041
0.0461	600	0.0014
0.0538	700	0.0019
0.0615	800	0.0015
0.0692	900	0.001
0.0769	1000	0.0005
0.0846	1100	0.0004
0.0922	1200	0.0013
0.0999	1300	0.0013
0.1076	1400	0.0027
0.1153	1500	0.0018
0.1230	1600	0.001
0.1307	1700	0.0014
0.1384	1800	0.0012
0.1460	1900	0.0041
0.1537	2000	0.0009
0.1614	2100	0.0005
0.1691	2200	0.0011
0.1768	2300	0.001
0.1845	2400	0.0004
0.1922	2500	0.0011
0.1998	2600	0.0044
0.2075	2700	0.0004
0.2152	2800	0.0022
0.2229	2900	0.0007
0.2306	3000	0.0006
0.2383	3100	0.0002
0.2460	3200	0.0006
0.2537	3300	0.0004
0.2613	3400	0.0013
0.2690	3500	0.0006
0.2767	3600	0.0005
0.2844	3700	0.0018
0.2921	3800	0.0023
0.2998	3900	0.0011
0.3075	4000	0.0007
0.3151	4100	0.0008
0.3228	4200	0.0013
0.3305	4300	0.0012
0.3382	4400	0.001
0.3459	4500	0.0016
0.3536	4600	0.0025
0.3613	4700	0.0015
0.3689	4800	0.0018
0.3766	4900	0.0019
0.3843	5000	0.0021
0.3920	5100	0.0018
0.3997	5200	0.0004
0.4074	5300	0.0006
0.4151	5400	0.0007
0.4228	5500	0.0009
0.4304	5600	0.0004
0.4381	5700	0.0003
0.4458	5800	0.0007
0.4535	5900	0.0013
0.4612	6000	0.0007
0.4689	6100	0.0005
0.4766	6200	0.001
0.4842	6300	0.0027
0.4919	6400	0.0018
0.4996	6500	0.0006
0.5073	6600	0.0008
0.5150	6700	0.0006
0.5227	6800	0.0007
0.5304	6900	0.001
0.5380	7000	0.0007
0.5457	7100	0.0005
0.5534	7200	0.0012
0.5611	7300	0.0012
0.5688	7400	0.0011
0.5765	7500	0.0005
0.5842	7600	0.0013
0.5919	7700	0.0012
0.5995	7800	0.0007
0.6072	7900	0.0012
0.6149	8000	0.0012
0.6226	8100	0.0003
0.6303	8200	0.0003
0.6380	8300	0.0003
0.6457	8400	0.002
0.6533	8500	0.0003
0.6610	8600	0.0016
0.6687	8700	0.0003
0.6764	8800	0.0002
0.6841	8900	0.0006
0.6918	9000	0.0005
0.6995	9100	0.0017
0.7071	9200	0.0037
0.7148	9300	0.0005
0.7225	9400	0.0006
0.7302	9500	0.0004
0.7379	9600	0.0002
0.7456	9700	0.0008
0.7533	9800	0.0005
0.7610	9900	0.0006
0.7686	10000	0.0004
0.7763	10100	0.0004
0.7840	10200	0.0006
0.7917	10300	0.0019
0.7994	10400	0.0007
0.8071	10500	0.0003
0.8148	10600	0.0003
0.8224	10700	0.0005
0.8301	10800	0.0009
0.8378	10900	0.0006
0.8455	11000	0.002
0.8532	11100	0.0018
0.8609	11200	0.0009
0.8686	11300	0.0004
0.8762	11400	0.0005
0.8839	11500	0.0008
0.8916	11600	0.0003
0.8993	11700	0.0002
0.9070	11800	0.0004
0.9147	11900	0.0007
0.9224	12000	0.0009
0.9301	12100	0.0007
0.9377	12200	0.0007
0.9454	12300	0.0009
0.9531	12400	0.0007
0.9608	12500	0.0009
0.9685	12600	0.0004
0.9762	12700	0.0002
0.9839	12800	0.0003
0.9915	12900	0.0002
0.9992	13000	0.0002
1.0069	13100	0.0006
1.0146	13200	0.0007
1.0223	13300	0.0007
1.0300	13400	0.0005
1.0377	13500	0.0008
1.0453	13600	0.0016
1.0530	13700	0.0007
1.0607	13800	0.0013
1.0684	13900	0.0005
1.0761	14000	0.0002
1.0838	14100	0.0001
1.0915	14200	0.0003
1.0992	14300	0.0003
1.1068	14400	0.0006
1.1145	14500	0.0002
1.1222	14600	0.0003
1.1299	14700	0.0002
1.1376	14800	0.0006
1.1453	14900	0.0011
1.1530	15000	0.0004
1.1606	15100	0.0001
1.1683	15200	0.0003
1.1760	15300	0.0001
1.1837	15400	0.0002
1.1914	15500	0.0001
1.1991	15600	0.003
1.2068	15700	0.0001
1.2145	15800	0.0002
1.2221	15900	0.0005
1.2298	16000	0.0004
1.2375	16100	0.0001
1.2452	16200	0.0003
1.2529	16300	0.0003
1.2606	16400	0.0008
1.2683	16500	0.0004
1.2759	16600	0.0001
1.2836	16700	0.0002
1.2913	16800	0.0011
1.2990	16900	0.0001
1.3067	17000	0.0001
1.3144	17100	0.0002
1.3221	17200	0.0005
1.3297	17300	0.0012
1.3374	17400	0.0003
1.3451	17500	0.0002
1.3528	17600	0.0009
1.3605	17700	0.0003
1.3682	17800	0.0005
1.3759	17900	0.0008
1.3836	18000	0.0005
1.3912	18100	0.0007
1.3989	18200	0.0002
1.4066	18300	0.0003
1.4143	18400	0.0002
1.4220	18500	0.0001
1.4297	18600	0.0001
1.4374	18700	0.0001
1.4450	18800	0.0005
1.4527	18900	0.0002
1.4604	19000	0.0001
1.4681	19100	0.0002
1.4758	19200	0.0006
1.4835	19300	0.0015
1.4912	19400	0.0012
1.4988	19500	0.0003
1.5065	19600	0.0005
1.5142	19700	0.0001
1.5219	19800	0.0002
1.5296	19900	0.0009
1.5373	20000	0.0002
1.5450	20100	0.0001
1.5527	20200	0.0003
1.5603	20300	0.0006
1.5680	20400	0.0002
1.5757	20500	0.0004
1.5834	20600	0.0006
1.5911	20700	0.0004
1.5988	20800	0.0002
1.6065	20900	0.0006
1.6141	21000	0.0006
1.6218	21100	0.0001
1.6295	21200	0.0001
1.6372	21300	0.0001
1.6449	21400	0.0008
1.6526	21500	0.0001
1.6603	21600	0.0005
1.6679	21700	0.0001
1.6756	21800	0.0001
1.6833	21900	0.0001
1.6910	22000	0.0001
1.6987	22100	0.0008
1.7064	22200	0.0014
1.7141	22300	0.0002
1.7218	22400	0.0007
1.7294	22500	0.0001
1.7371	22600	0.0001
1.7448	22700	0.0001
1.7525	22800	0.0002
1.7602	22900	0.0002
1.7679	23000	0.0001
1.7756	23100	0.0001
1.7832	23200	0.0005
1.7909	23300	0.0004
1.7986	23400	0.0002
1.8063	23500	0.0001
1.8140	23600	0.0001
1.8217	23700	0.0001
1.8294	23800	0.0004
1.8370	23900	0.0002
1.8447	24000	0.0002
1.8524	24100	0.0013
1.8601	24200	0.0004
1.8678	24300	0.0002
1.8755	24400	0.0002
1.8832	24500	0.0001
1.8909	24600	0.0001
1.8985	24700	0.0001
1.9062	24800	0.0002
1.9139	24900	0.0005
1.9216	25000	0.0001
1.9293	25100	0.0001
1.9370	25200	0.0002
1.9447	25300	0.0002
1.9523	25400	0.0006
1.9600	25500	0.0004
1.9677	25600	0.0002
1.9754	25700	0.0001
1.9831	25800	0.0001
1.9908	25900	0.0001
1.9985	26000	0.0001
2.0061	26100	0.0002
2.0138	26200	0.0007
2.0215	26300	0.0003
2.0292	26400	0.0001
2.0369	26500	0.0011
2.0446	26600	0.0002
2.0523	26700	0.0001
2.0600	26800	0.0002
2.0676	26900	0.0004
2.0753	27000	0.0001
2.0830	27100	0.0001
2.0907	27200	0.0001
2.0984	27300	0.0002
2.1061	27400	0.0001
2.1138	27500	0.0001
2.1214	27600	0.0001
2.1291	27700	0.0001
2.1368	27800	0.0003
2.1445	27900	0.0012
2.1522	28000	0.0001
2.1599	28100	0.0001
2.1676	28200	0.0001
2.1752	28300	0.0001
2.1829	28400	0.0001
2.1906	28500	0.0001
2.1983	28600	0.0014
2.2060	28700	0.0001
2.2137	28800	0.0001
2.2214	28900	0.0002
2.2291	29000	0.0
2.2367	29100	0.0001
2.2444	29200	0.0001
2.2521	29300	0.0001
2.2598	29400	0.0001
2.2675	29500	0.0001
2.2752	29600	0.0001
2.2829	29700	0.0001
2.2905	29800	0.0001
2.2982	29900	0.0001
2.3059	30000	0.0001
2.3136	30100	0.0001
2.3213	30200	0.0002
2.3290	30300	0.0011
2.3367	30400	0.0001
2.3444	30500	0.0001
2.3520	30600	0.0005
2.3597	30700	0.0001
2.3674	30800	0.0001
2.3751	30900	0.0006
2.3828	31000	0.0001
2.3905	31100	0.0001
2.3982	31200	0.0002
2.4058	31300	0.0001
2.4135	31400	0.0001
2.4212	31500	0.0001
2.4289	31600	0.0001
2.4366	31700	0.0001
2.4443	31800	0.0004
2.4520	31900	0.0001
2.4596	32000	0.0001
2.4673	32100	0.0002
2.4750	32200	0.0002
2.4827	32300	0.0004
2.4904	32400	0.0008
2.4981	32500	0.0001
2.5058	32600	0.0001
2.5135	32700	0.0001
2.5211	32800	0.0001
2.5288	32900	0.0006
2.5365	33000	0.0001
2.5442	33100	0.0001
2.5519	33200	0.0002
2.5596	33300	0.0001
2.5673	33400	0.0002
2.5749	33500	0.0001
2.5826	33600	0.0001
2.5903	33700	0.0001
2.5980	33800	0.0001
2.6057	33900	0.0001
2.6134	34000	0.0007
2.6211	34100	0.0
2.6287	34200	0.0001
2.6364	34300	0.0001
2.6441	34400	0.0006
2.6518	34500	0.0001
2.6595	34600	0.0001
2.6672	34700	0.0001
2.6749	34800	0.0
2.6826	34900	0.0001
2.6902	35000	0.0001
2.6979	35100	0.0005
2.7056	35200	0.0006
2.7133	35300	0.0001
2.7210	35400	0.0005
2.7287	35500	0.0001
2.7364	35600	0.0001
2.7440	35700	0.0001
2.7517	35800	0.0001
2.7594	35900	0.0001
2.7671	36000	0.0001
2.7748	36100	0.0001
2.7825	36200	0.0005
2.7902	36300	0.0001
2.7978	36400	0.0001
2.8055	36500	0.0001
2.8132	36600	0.0001
2.8209	36700	0.0001
2.8286	36800	0.0001
2.8363	36900	0.0001
2.8440	37000	0.0001
2.8517	37100	0.0013
2.8593	37200	0.0001
2.8670	37300	0.0001
2.8747	37400	0.0001
2.8824	37500	0.0001
2.8901	37600	0.0001
2.8978	37700	0.0001
2.9055	37800	0.0001
2.9131	37900	0.0002
2.9208	38000	0.0001
2.9285	38100	0.0001
2.9362	38200	0.0001
2.9439	38300	0.0001
2.9516	38400	0.0004
2.9593	38500	0.0001
2.9669	38600	0.0001
2.9746	38700	0.0001
2.9823	38800	0.0001
2.9900	38900	0.0001
2.9977	39000	0.0001

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.1.0
• Transformers: 4.44.2
• PyTorch: 2.4.0+cu121
• Accelerate: 0.34.2
• Datasets: 3.0.0
• 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",
}

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}
}

#Add SimSCE reference