CrossEncoder based on cross-encoder/ettin-reranker-68m-v1

This is a Cross Encoder model finetuned from cross-encoder/ettin-reranker-68m-v1 using the sentence-transformers library. It computes scores for pairs of texts, which can be used for text reranking and semantic search.

Model Details

Model Description

Model Type: Cross Encoder
Base model: cross-encoder/ettin-reranker-68m-v1
Maximum Sequence Length: 7999 tokens
Number of Output Labels: 1 label
Supported Modality: Text

Model Sources

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

Full Model Architecture

CrossEncoder(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'ModernBertModel'})
  (1): Pooling({'embedding_dimension': 512, 'pooling_mode': 'cls', 'include_prompt': True})
  (2): Dense({'in_features': 512, 'out_features': 512, 'bias': False, 'activation_function': 'torch.nn.modules.activation.GELU', 'module_input_name': 'sentence_embedding', 'module_output_name': 'sentence_embedding'})
  (3): LayerNorm({'dimension': 512})
  (4): Dense({'in_features': 512, 'out_features': 1, 'bias': True, 'activation_function': 'torch.nn.modules.linear.Identity', 'module_input_name': 'sentence_embedding', 'module_output_name': 'scores'})
)

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 CrossEncoder

# Download from the 🤗 Hub
model = CrossEncoder("cross_encoder_model_id")
# Get scores for pairs of inputs
pairs = [
    ['Bill Garrison Hiatus interview podcast 2023 2024 2025', 'The Summer Rhubarb Tour. The first of many non-radio APHC tours, one-nighters, covering the country. Sue Scott, Fred Newman, and I were on some of the first of these tours and it was deluxe: jetting to a city in a private plane (Pearl Jam’s touring plane), doing a show then flying to the next city, a stay in a hotel for the day, then doing the next show, then rinse and repeat. Later Garrison took just Fred, the musicians, and a guest singer like Suzy Bogguss or Sara Watkins for a couple of weeks on a magical mystery bus tour with a different theme each summer — two buses, lots of adventures.\n\nJENNIFER HOWE, APHC OFFICE MANAGER AND PERSONAL ASSISTANT TO GARRISON'],
    ['Bill Garrison Hiatus interview podcast 2023 2024 2025', "where it all falls in a couple of weeks to when you need 270 electoral votes and there was just a fight recently about one state that splits up its votes Nebraska and a single District that Democrats are favored to win and Republicans were trying to flip this rules in Nebraska so all of the electoral votes would have gone to Trump the effort failed but this was about that blue wall because if you didn't flip that and Harris just won those three states Michigan Wisconsin and Pennsylvania she and he would end at a 269 tie like it's not even unlikely they were talking about plans to get to a tie that's how close this RAC and that's why the Trump folks wanted to change the rules in Nebraska they wanted in the event of a tie for that final tiebreaking vote to not go to k Harris they wanted a"],
    ['Bill Garrison Hiatus interview podcast 2023 2024 2025', 'Spaghetti in Saint Paul, 1974\n\nReturning to Minneapolis from 14 months doing national service (VISTA/Western Nebraska) in the fall of 1974, I got a call from fellow performers Bill Hinkley and Judy Larson that a writer and radio DJ named Garrison Keillor had just started a variety radio show, based in part on the Grand Ole Opry. They both knew my music — like “Getting in the Cows” and “Winter on the Farm.” The songs had a rural theme that Garrison was after at the time, and Bill and Judy pitched me to Garrison thinking I might be a good fit. He invited me and my wife to dinner at his house for spaghetti, and afterward, hearing a few of my songs clinched the deal. I was a featured guest on the second or third broadcast, and a fairly regular performer for years after.'],
    ['Bill Garrison Hiatus interview podcast 2023 2024 2025', 'This is not the career I was trained for. I decided to become a writer when I was 14 and so went to a liberal arts college and studied Humanities, which was a tragic mistake. My cousins went into medicine, architecture, psychology, engineering, education — cheerful occupations intent on making the world better — and my most successful cousin, Wayne, went into raising flowers and vegetables and built enormous greenhouses and did good on a massive scale, carrying on in the tradition of my aunt Josephine who was a passionate gardener, whereas I got trapped in the sorrows of the 19th century, which dominated humanities in the 20th, gloomy philosophers like Kafka, Kierkegaard, Thoreau.'],
    ['Bill Garrison Hiatus interview podcast 2023 2024 2025', 'Garrison’s willingness to accept outside speaking engagements — in addition to writing, producing, and hosting a weekly show — was remarkable. In most cases, a request for Garrison to “stop by and greet volunteers of a hospital or church during their lunch break” would turn into a full-blown two-hour benefit performance, in a theater that held 2,000 people. Garrison would insist that all proceeds go to the charity the volunteers were working for. It is amazing how Garrison gives of his time and talents.'],
]
scores = model.predict(pairs)
print(scores)
# [-0.5195 -0.3984 -0.625  -0.6953  0.127 ]

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'Bill Garrison Hiatus interview podcast 2023 2024 2025',
    [
        'The Summer Rhubarb Tour. The first of many non-radio APHC tours, one-nighters, covering the country. Sue Scott, Fred Newman, and I were on some of the first of these tours and it was deluxe: jetting to a city in a private plane (Pearl Jam’s touring plane), doing a show then flying to the next city, a stay in a hotel for the day, then doing the next show, then rinse and repeat. Later Garrison took just Fred, the musicians, and a guest singer like Suzy Bogguss or Sara Watkins for a couple of weeks on a magical mystery bus tour with a different theme each summer — two buses, lots of adventures.\n\nJENNIFER HOWE, APHC OFFICE MANAGER AND PERSONAL ASSISTANT TO GARRISON',
        "where it all falls in a couple of weeks to when you need 270 electoral votes and there was just a fight recently about one state that splits up its votes Nebraska and a single District that Democrats are favored to win and Republicans were trying to flip this rules in Nebraska so all of the electoral votes would have gone to Trump the effort failed but this was about that blue wall because if you didn't flip that and Harris just won those three states Michigan Wisconsin and Pennsylvania she and he would end at a 269 tie like it's not even unlikely they were talking about plans to get to a tie that's how close this RAC and that's why the Trump folks wanted to change the rules in Nebraska they wanted in the event of a tie for that final tiebreaking vote to not go to k Harris they wanted a",
        'Spaghetti in Saint Paul, 1974\n\nReturning to Minneapolis from 14 months doing national service (VISTA/Western Nebraska) in the fall of 1974, I got a call from fellow performers Bill Hinkley and Judy Larson that a writer and radio DJ named Garrison Keillor had just started a variety radio show, based in part on the Grand Ole Opry. They both knew my music — like “Getting in the Cows” and “Winter on the Farm.” The songs had a rural theme that Garrison was after at the time, and Bill and Judy pitched me to Garrison thinking I might be a good fit. He invited me and my wife to dinner at his house for spaghetti, and afterward, hearing a few of my songs clinched the deal. I was a featured guest on the second or third broadcast, and a fairly regular performer for years after.',
        'This is not the career I was trained for. I decided to become a writer when I was 14 and so went to a liberal arts college and studied Humanities, which was a tragic mistake. My cousins went into medicine, architecture, psychology, engineering, education — cheerful occupations intent on making the world better — and my most successful cousin, Wayne, went into raising flowers and vegetables and built enormous greenhouses and did good on a massive scale, carrying on in the tradition of my aunt Josephine who was a passionate gardener, whereas I got trapped in the sorrows of the 19th century, which dominated humanities in the 20th, gloomy philosophers like Kafka, Kierkegaard, Thoreau.',
        'Garrison’s willingness to accept outside speaking engagements — in addition to writing, producing, and hosting a weekly show — was remarkable. In most cases, a request for Garrison to “stop by and greet volunteers of a hospital or church during their lunch break” would turn into a full-blown two-hour benefit performance, in a theater that held 2,000 people. Garrison would insist that all proceeds go to the charity the volunteers were working for. It is amazing how Garrison gives of his time and talents.',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Training Details

Training Dataset

Unnamed Dataset

Size: 8,729,587 training samples
Columns: query, doc, and label

Approximate statistics based on the first 100 samples:

	query	doc	label
type	string	string	float
modality	text	text
details	min: 14 tokens mean: 14.0 tokens max: 14 tokens	min: 10 tokens mean: 134.07 tokens max: 257 tokens	min: -1.45 mean: -0.52 max: 1.44

Samples:

query	doc	label
`Xiaodan Tan University of Waterloo undergraduate engineering computer`	3) Advanced IRS: In addition to the IRS sensing architec-tures discussed above, several new designs have been proposed to overcome their limitations and enhance performance, includ-ing beyond diagonal IRS (BD-IRS) sensing, active IRS sensing, holographic IRS sensing, and STAR-IRS, as follows: Most existing IRS sensing studies concentrate on using a sim-ple IRS model with a diagonal phase shift matrix, where each IRS element is connected to its own reconfigurable impedance without any inter-element connections. Such a lossless IRS only supports the reflection of signals toward the same side, which limits the sensing coverage. To overcome this limitation and 6 Unauthorized radar Legitimate radar Target (b) Target-mounted IRS for secure sensing Target (a) Target-mounted IRS aided	`-0.08750010281801224`
`Xiaodan Tan University of Waterloo undergraduate engineering computer`	`Bokhee Im, Chonnam National University Alex W. Nowak, Iowa State University Jonathan D. H. Smith, Iowa State University (1150-20-594) 11:00 a.m. Asymptotic Irrelevance of Quasigroup Identities. Jonathan D.H. Smith, Iowa State University Stefanie G. Wang, Smith College (1150-20-548)`	`-1.181249976158142`
`Xiaodan Tan University of Waterloo undergraduate engineering computer`	2) Parameter estimation: In parameter estimation, IRS con-tributes by optimizing electromagnetic wave propagation paths to mitigate multipath effects and interference. It enhances mea-surement stability and reliability while facilitating more accurate extraction of target parameters. Furthermore, the IRS can be integrated with advanced signal processing algorithms to refine radar echo signals to achieve superior estimation precision. The following analysis starts with parameter estimation for the single-target case and then extends to the multi-target scenario.	`-0.30625027418136597`

Loss: MSELoss with these parameters:

{
    "activation_fn": "torch.nn.modules.linear.Identity"
}

Evaluation Dataset

Unnamed Dataset

Size: 100,000 evaluation samples
Columns: query, doc, and label

Approximate statistics based on the first 100 samples:

	query	doc	label
type	string	string	float
modality	text	text
details	min: 15 tokens mean: 15.0 tokens max: 15 tokens	min: 8 tokens mean: 149.99 tokens max: 266 tokens	min: 0.21 mean: 0.39 max: 0.7

Samples:

query	doc	label
`Bill Garrison Hiatus interview podcast 2023 2024 2025`	The Summer Rhubarb Tour. The first of many non-radio APHC tours, one-nighters, covering the country. Sue Scott, Fred Newman, and I were on some of the first of these tours and it was deluxe: jetting to a city in a private plane (Pearl Jam’s touring plane), doing a show then flying to the next city, a stay in a hotel for the day, then doing the next show, then rinse and repeat. Later Garrison took just Fred, the musicians, and a guest singer like Suzy Bogguss or Sara Watkins for a couple of weeks on a magical mystery bus tour with a different theme each summer — two buses, lots of adventures. JENNIFER HOWE, APHC OFFICE MANAGER AND PERSONAL ASSISTANT TO GARRISON	`0.4703472852706909`
`Bill Garrison Hiatus interview podcast 2023 2024 2025`	where it all falls in a couple of weeks to when you need 270 electoral votes and there was just a fight recently about one state that splits up its votes Nebraska and a single District that Democrats are favored to win and Republicans were trying to flip this rules in Nebraska so all of the electoral votes would have gone to Trump the effort failed but this was about that blue wall because if you didn't flip that and Harris just won those three states Michigan Wisconsin and Pennsylvania she and he would end at a 269 tie like it's not even unlikely they were talking about plans to get to a tie that's how close this RAC and that's why the Trump folks wanted to change the rules in Nebraska they wanted in the event of a tie for that final tiebreaking vote to not go to k Harris they wanted a	`0.4742415249347687`
`Bill Garrison Hiatus interview podcast 2023 2024 2025`	Spaghetti in Saint Paul, 1974 Returning to Minneapolis from 14 months doing national service (VISTA/Western Nebraska) in the fall of 1974, I got a call from fellow performers Bill Hinkley and Judy Larson that a writer and radio DJ named Garrison Keillor had just started a variety radio show, based in part on the Grand Ole Opry. They both knew my music — like “Getting in the Cows” and “Winter on the Farm.” The songs had a rural theme that Garrison was after at the time, and Bill and Judy pitched me to Garrison thinking I might be a good fit. He invited me and my wife to dinner at his house for spaghetti, and afterward, hearing a few of my songs clinched the deal. I was a featured guest on the second or third broadcast, and a fairly regular performer for years after.	`0.35720863938331604`

Loss: MSELoss with these parameters:

{
    "activation_fn": "torch.nn.modules.linear.Identity"
}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 64
max_steps: 15000
learning_rate: 2e-05
bf16: True
per_device_eval_batch_size: 256
eval_on_start: True
remove_unused_columns: False
warmup_ratio: 0.0

All Hyperparameters

Click to expand

per_device_train_batch_size: 64
num_train_epochs: 3.0
max_steps: 15000
learning_rate: 2e-05
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_steps: 0.0
optim: adamw_torch
optim_args: None
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
optim_target_modules: None
gradient_accumulation_steps: 1
average_tokens_across_devices: True
max_grad_norm: 1.0
label_smoothing_factor: 0.0
bf16: True
fp16: False
bf16_full_eval: False
fp16_full_eval: False
tf32: None
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
use_liger_kernel: False
liger_kernel_config: None
use_cache: False
neftune_noise_alpha: None
torch_empty_cache_steps: None
auto_find_batch_size: False
log_on_each_node: True
logging_nan_inf_filter: True
include_num_input_tokens_seen: no
log_level: passive
log_level_replica: warning
disable_tqdm: False
project: huggingface
trackio_space_id: None
trackio_bucket_id: None
trackio_static_space_id: None
per_device_eval_batch_size: 256
prediction_loss_only: True
eval_on_start: True
eval_do_concat_batches: True
eval_use_gather_object: False
eval_accumulation_steps: None
include_for_metrics: []
batch_eval_metrics: False
save_only_model: False
save_on_each_node: False
enable_jit_checkpoint: False
push_to_hub: False
hub_private_repo: None
hub_model_id: None
hub_strategy: every_save
hub_always_push: False
hub_revision: None
load_best_model_at_end: False
ignore_data_skip: False
restore_callback_states_from_checkpoint: False
full_determinism: False
seed: 42
data_seed: None
use_cpu: False
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
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_pin_memory: True
dataloader_persistent_workers: False
dataloader_prefetch_factor: None
remove_unused_columns: False
label_names: None
train_sampling_strategy: random
length_column_name: length
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
ddp_static_graph: None
ddp_backend: None
ddp_timeout: 1800
fsdp: []
fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
deepspeed: None
debug: []
skip_memory_metrics: True
do_predict: False
resume_from_checkpoint: None
warmup_ratio: 0.0
local_rank: -1
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Click to expand

Epoch	Step	Training Loss	Validation Loss
0	0	-	16.5714
0.0004	50	2.8778	-
0.0007	100	0.3278	-
0.0011	150	0.2539	-
0.0015	200	0.2421	-
0.0018	250	0.2145	-
0.0022	300	0.2320	-
0.0026	350	0.2132	-
0.0029	400	0.2535	-
0.0033	450	0.2066	-
0.0037	500	0.1772	-
0.0040	550	0.1944	-
0.0044	600	0.1805	-
0.0048	650	0.1779	-
0.0051	700	0.2133	-
0.0055	750	0.1860	-
0.0059	800	0.1908	-
0.0062	850	0.1968	-
0.0066	900	0.1715	-
0.0070	950	0.1799	-
0.0073	1000	0.1726	-
0.0077	1050	0.1754	-
0.0081	1100	0.1660	-
0.0084	1150	0.1617	-
0.0088	1200	0.1640	-
0.0092	1250	0.1621	-
0.0095	1300	0.1642	-
0.0099	1350	0.1493	-
0.0103	1400	0.1606	-
0.0106	1450	0.1489	-
0.0110	1500	0.1536	-
0.0114	1550	0.1813	-
0.0117	1600	0.1462	-
0.0121	1650	0.1493	-
0.0125	1700	0.1804	-
0.0128	1750	0.1571	-
0.0132	1800	0.1783	-
0.0136	1850	0.1461	-
0.0139	1900	0.1568	-
0.0143	1950	0.1475	-
0.0147	2000	0.1459	-
0.0150	2050	0.1440	-
0.0154	2100	0.1513	-
0.0158	2150	0.1427	-
0.0161	2200	0.1574	-
0.0165	2250	0.1393	-
0.0169	2300	0.1360	-
0.0172	2350	0.1398	-
0.0176	2400	0.1388	-
0.0180	2450	0.1651	-
0.0183	2500	0.1367	-
0.0187	2550	0.1396	-
0.0191	2600	0.1538	-
0.0194	2650	0.1368	-
0.0198	2700	0.1351	-
0.0202	2750	0.1408	-
0.0205	2800	0.1329	-
0.0209	2850	0.1262	-
0.0213	2900	0.1394	-
0.0216	2950	0.1444	-
0.0220	3000	0.1405	1.5620
0.0224	3050	0.1297	-
0.0227	3100	0.1362	-
0.0231	3150	0.1365	-
0.0235	3200	0.1217	-
0.0238	3250	0.1318	-
0.0242	3300	0.1592	-
0.0246	3350	0.1329	-
0.0249	3400	0.1269	-
0.0253	3450	0.1460	-
0.0257	3500	0.1272	-
0.0260	3550	0.1513	-
0.0264	3600	0.1262	-
0.0268	3650	0.1280	-
0.0271	3700	0.1206	-
0.0275	3750	0.1187	-
0.0279	3800	0.1446	-
0.0282	3850	0.1273	-
0.0286	3900	0.1268	-
0.0290	3950	0.1467	-
0.0293	4000	0.1354	-
0.0297	4050	0.1164	-
0.0301	4100	0.1266	-
0.0304	4150	0.1204	-
0.0308	4200	0.1234	-
0.0312	4250	0.1189	-
0.0315	4300	0.1237	-
0.0319	4350	0.1213	-
0.0323	4400	0.1194	-
0.0326	4450	0.1406	-
0.0330	4500	0.1153	-
0.0334	4550	0.1408	-
0.0337	4600	0.1185	-
0.0341	4650	0.1562	-
0.0345	4700	0.1178	-
0.0348	4750	0.1137	-
0.0352	4800	0.1099	-
0.0356	4850	0.1362	-
0.0359	4900	0.1147	-
0.0363	4950	0.1181	-
0.0367	5000	0.1175	-
0.0370	5050	0.1175	-
0.0374	5100	0.1388	-
0.0378	5150	0.1140	-
0.0381	5200	0.1070	-
0.0385	5250	0.1184	-
0.0389	5300	0.1314	-
0.0392	5350	0.1122	-
0.0396	5400	0.1349	-
0.0400	5450	0.1167	-
0.0403	5500	0.1110	-
0.0407	5550	0.1105	-
0.0411	5600	0.1492	-
0.0414	5650	0.1307	-
0.0418	5700	0.1104	-
0.0422	5750	0.1150	-
0.0425	5800	0.1356	-
0.0429	5850	0.1104	-
0.0433	5900	0.1293	-
0.0436	5950	0.1160	-
0.0440	6000	0.1104	1.6807
0.0444	6050	0.1113	-
0.0447	6100	0.1109	-
0.0451	6150	0.1090	-
0.0455	6200	0.1241	-
0.0458	6250	0.1336	-
0.0462	6300	0.1130	-
0.0466	6350	0.1168	-
0.0469	6400	0.1076	-
0.0473	6450	0.1096	-
0.0477	6500	0.1178	-
0.0480	6550	0.1088	-
0.0484	6600	0.1062	-
0.0488	6650	0.1091	-
0.0491	6700	0.1002	-
0.0495	6750	0.1096	-
0.0499	6800	0.1635	-
0.0502	6850	0.1043	-
0.0506	6900	0.1048	-
0.0510	6950	0.1174	-
0.0513	7000	0.1090	-
0.0517	7050	0.1026	-
0.0521	7100	0.1084	-
0.0524	7150	0.1096	-
0.0528	7200	0.1066	-
0.0532	7250	0.1142	-
0.0535	7300	0.1201	-
0.0539	7350	0.1049	-
0.0543	7400	0.1310	-
0.0546	7450	0.1328	-
0.0550	7500	0.1060	-
0.0554	7550	0.1068	-
0.0557	7600	0.1191	-
0.0561	7650	0.1245	-
0.0565	7700	0.1207	-
0.0568	7750	0.0960	-
0.0572	7800	0.1228	-
0.0576	7850	0.1011	-
0.0579	7900	0.1039	-
0.0583	7950	0.1027	-
0.0587	8000	0.0954	-
0.0590	8050	0.0991	-
0.0594	8100	0.1071	-
0.0598	8150	0.0997	-
0.0601	8200	0.1041	-
0.0605	8250	0.0968	-
0.0609	8300	0.1089	-
0.0612	8350	0.0980	-
0.0616	8400	0.0975	-
0.0620	8450	0.0999	-
0.0623	8500	0.1017	-
0.0627	8550	0.1006	-
0.0630	8600	0.0970	-
0.0634	8650	0.0964	-
0.0638	8700	0.1031	-
0.0641	8750	0.0986	-
0.0645	8800	0.1029	-
0.0649	8850	0.0981	-
0.0652	8900	0.1017	-
0.0656	8950	0.0989	-
0.0660	9000	0.0967	1.7543
0.0663	9050	0.0936	-
0.0667	9100	0.0959	-
0.0671	9150	0.0963	-
0.0674	9200	0.0995	-
0.0678	9250	0.1002	-
0.0682	9300	0.1029	-
0.0685	9350	0.0921	-
0.0689	9400	0.0896	-
0.0693	9450	0.1027	-
0.0696	9500	0.0953	-
0.0700	9550	0.0967	-
0.0704	9600	0.0922	-
0.0707	9650	0.0963	-
0.0711	9700	0.1044	-
0.0715	9750	0.0953	-
0.0718	9800	0.0918	-
0.0722	9850	0.0891	-
0.0726	9900	0.0954	-
0.0729	9950	0.1024	-
0.0733	10000	0.0952	-
0.0737	10050	0.0933	-
0.0740	10100	0.0933	-
0.0744	10150	0.1215	-
0.0748	10200	0.0937	-
0.0751	10250	0.1088	-
0.0755	10300	0.1082	-
0.0759	10350	0.0938	-
0.0762	10400	0.1044	-
0.0766	10450	0.1000	-
0.0770	10500	0.0937	-
0.0773	10550	0.0944	-
0.0777	10600	0.1356	-
0.0781	10650	0.0964	-
0.0784	10700	0.0913	-
0.0788	10750	0.1047	-
0.0792	10800	0.0879	-
0.0795	10850	0.0944	-
0.0799	10900	0.0973	-
0.0803	10950	0.0909	-
0.0806	11000	0.0944	-
0.0810	11050	0.1437	-
0.0814	11100	0.1228	-
0.0817	11150	0.0894	-
0.0821	11200	0.0923	-
0.0825	11250	0.0923	-
0.0828	11300	0.1092	-
0.0832	11350	0.0888	-
0.0836	11400	0.0873	-
0.0839	11450	0.0907	-
0.0843	11500	0.0885	-
0.0847	11550	0.0925	-
0.0850	11600	0.0933	-
0.0854	11650	0.0888	-
0.0858	11700	0.0944	-
0.0861	11750	0.0962	-
0.0865	11800	0.0996	-
0.0869	11850	0.0889	-
0.0872	11900	0.0906	-
0.0876	11950	0.0880	-
0.0880	12000	0.0935	1.6985

Training Time

Training: 43.9 minutes
Evaluation: 28.5 minutes
Total: 1.2 hours

Framework Versions

Python: 3.12.12
Sentence Transformers: 5.5.1
Transformers: 5.8.1
PyTorch: 2.6.0+cu124
Accelerate: 1.13.0
Datasets: 4.8.5
Tokenizers: 0.22.2

Additional Resources

Training and Finetuning Reranker Models with Sentence Transformers: the end-to-end guide for training or finetuning Cross Encoder (reranker) models.
Multimodal Embedding & Reranker Models with Sentence Transformers: use text, image, audio, and video reranker models through the same API.
Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers: training multimodal Cross Encoders.

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",
}