chowder-rerank (MiniLM reranker fine-tuned on the chowder corpus)

This is a Cross Encoder model finetuned from cross-encoder/ms-marco-MiniLM-L6-v2 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/ms-marco-MiniLM-L6-v2
Maximum Sequence Length: 512 tokens
Number of Output Labels: 1 label
Supported Modality: Text
Language: en

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': 'sequence-classification', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'logits'}}, 'module_output_name': 'scores', 'architecture': 'BertForSequenceClassification'})
)

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 = [
    ['Estimating the safe mud weight window for drilling operations through pre-stack seismic inversion, a case study in one of the Southwestern Iran oil fields', 'High-temperature drilling environments often intensify rock–fluid interactions, leading to substantial formation damage primarily due to excessive mud filtrate invasion. Conventional additives employed to mitigate this issue frequently exhibit thermal degradation, limiting their effectiveness under elevated temperatures. Notably, the literature reveals a significant gap concerning the invasion behavior of bentonite-free mud filtrate within payzone sections which is considered a critical aspect for optimizing hydrocarbon recovery. This study investigates the influence of drilling mud filtrate–rock interactions on rock strength and porosity at high temperatures through an integrated approach combining laboratory experiments and computational fluid dynamics (CFD) simulations. Rock strength of sandstone and limestone was quantified using unconfined compressive strength (UCS) tests. Experimental findings indicate a significant reduction in strength following filtrate exposure. Bandera Brown exhibited a 33.42% strength reduction after 24 h at 300 ℉. Micro-computed tomography (µCT) imaging facilitated pore-scale characterization, revealing significant alterations in porosity and pore network architecture under prolonged exposure. Specifically, Bandera Brown’s porosity increased by 16.5%, whereas Boise Sandstone demonstrated a 37.64% decline over the same period. CFD simulations further interpreted invasion dynamics, showing that high porosity mitigates penetration depth, while increased permeability, extended exposure duration, and higher pressure differentials promote deeper filtrate infiltration. These results highlight the critical role of pressure management in minimizing formation damage. This research provides novel insights into the complex interplay between mud filtrate and reservoir rock properties under high-temperature conditions, offering practical implications for drilling fluid design and wellbore stability in the oil and gas industry.'],
    ['Achieving EF1 and Epistemic EFX Guarantees Simultaneously', 'The Efimov effect is a quantum phenomenon in which three particles form an infinite series of low-energy bound states at resonance, previously observed in three-dimensional systems. The authors extend this celebrated effect to long-range quantum spin systems, demonstrating long-range spin chains can also host Efimov states with widely tunable scaling ratios controlled by the interaction range.'],
    ['Study on dye-sensitized solar cell efficiency improvement using methyl orange dye', 'Abstract In this work, different varieties of dye sensitized solar cells are fabricated by simple fabrication process. In this fabrication extract of butea monosperma flower, methylene blue and methyl orange dyes are used as sensitizers. The photovoltaic performance of dye sensitized solar cells (DSSCs) has been studied. The performances of two different types of photo-electrodes are also tested in this work. The morphology and bandgap of TiO 2 (titanium dioxide) and ZnO (Zinc oxide) was observed from XRD, FTIR spectroscopy and UV-vis Spectrum. It is found that TiO 2 based DSSCs have better performance. It also observed that the current density and efficiency was increased from 7.46 to 12.9 mA/cm 2 and from 1.34 to 6.8% respectively when using methyl orange as a dye. Hence it can be said that methyl orange dye enhanced the photovoltaic performance of DSSC.'],
    ['Consciousness as Uncommon Self-Knowledge: A Synergistic Information Framework', "The slogan ``information is physical,'' introduced by Rolf Landauer and developed through quantum information theory and black-hole thermodynamics, has achieved near-axiomatic status in modern physics. Yet the ontological status of information remains surprisingly underexamined: most discussions either reduce information to a form of energy or treat it as a purely mathematical object. This paper proposes a third position. I argue that information is neither a physical substance nor a free-floating abstraction, but rather \\emph{the structure of physically realizable alternatives} -- a counterfactual structure that a physical system instantiates in virtue of the possibility space available to it. Building on Shannon's combinatorial definition, the Landauer principle, the no-cloning theorem, and the black-hole information paradox, I show that the informational content of any physical event is constituted by the set of outcomes that \\emph{could have occurred} but did not. This counterfactual reading dissolves several persistent confusions: it explains why erasing information dissipates heat without making information ``material,'' why quantum superposition is informationally richer than any classical mixture, and why information loss in black holes is physically significant beyond mere bookkeeping. The proposal sits within a structural-realist framework but departs from standard structural realism by locating the relevant structure in modal, not merely actual, relations. I conclude by sketching implications for the foundations of quantum mechanics, quantum gravity, and scientific ontology more broadly."],
    ['Design strategies, methods, and photophysical insights in polymeric photocatalysts for solar-driven hydrogen evolution', 'Abstract Efficient solar-to-hydrogen conversion remains challenging due to the limited visible-light activity and rapid charge recombination in conventional photocatalysts. Here, we report a plasmon-enhanced nitrogen-doped niobium pentoxide (N–Nb₂O₅) photocatalyst decorated with gold nanoparticles (Au@N–Nb₂O₅) for high-performance solar-driven photocatalytic hydrogen generation. N–Nb₂O₅ was synthesized via a simple wet-chemical route and calcined at 500 °C, forming crystalline orthorhombic nanoplates with XRD-derived crystallite sizes of 50–55 nm and lateral dimensions of ~ 150 nm (FE-SEM). Gold nanoparticles were subsequently deposited via photodeposition, extending visible-light absorption and narrowing the band gap to 2.3–2.5 eV (UV–DRS). XPS analysis confirmed successful nitrogen incorporation and surface metallization. Photoluminescence studies revealed efficient charge separation and concentration-dependent suppression of radiative recombination. The optimised Au@N–Nb₂O₅ (2 wt% Au) achieved a hydrogen evolution rate of 2168 µmol h⁻ 1 g⁻ 1 under natural sunlight, nearly fourfold higher than pristine Nb₂O₅ and N–Nb₂O₅. The ordered nanoplate morphology facilitates charge transport, complementing the interfacial effects of Au, providing a scalable strategy for designing high-performance Nb₂O₅-based photocatalysts for sustainable solar water splitting.'],
]
scores = model.predict(pairs)
print(scores)
# [-8.0625 -7.1875  9.875  -7.1875  2.9844]

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'Estimating the safe mud weight window for drilling operations through pre-stack seismic inversion, a case study in one of the Southwestern Iran oil fields',
    [
        'High-temperature drilling environments often intensify rock–fluid interactions, leading to substantial formation damage primarily due to excessive mud filtrate invasion. Conventional additives employed to mitigate this issue frequently exhibit thermal degradation, limiting their effectiveness under elevated temperatures. Notably, the literature reveals a significant gap concerning the invasion behavior of bentonite-free mud filtrate within payzone sections which is considered a critical aspect for optimizing hydrocarbon recovery. This study investigates the influence of drilling mud filtrate–rock interactions on rock strength and porosity at high temperatures through an integrated approach combining laboratory experiments and computational fluid dynamics (CFD) simulations. Rock strength of sandstone and limestone was quantified using unconfined compressive strength (UCS) tests. Experimental findings indicate a significant reduction in strength following filtrate exposure. Bandera Brown exhibited a 33.42% strength reduction after 24 h at 300 ℉. Micro-computed tomography (µCT) imaging facilitated pore-scale characterization, revealing significant alterations in porosity and pore network architecture under prolonged exposure. Specifically, Bandera Brown’s porosity increased by 16.5%, whereas Boise Sandstone demonstrated a 37.64% decline over the same period. CFD simulations further interpreted invasion dynamics, showing that high porosity mitigates penetration depth, while increased permeability, extended exposure duration, and higher pressure differentials promote deeper filtrate infiltration. These results highlight the critical role of pressure management in minimizing formation damage. This research provides novel insights into the complex interplay between mud filtrate and reservoir rock properties under high-temperature conditions, offering practical implications for drilling fluid design and wellbore stability in the oil and gas industry.',
        'The Efimov effect is a quantum phenomenon in which three particles form an infinite series of low-energy bound states at resonance, previously observed in three-dimensional systems. The authors extend this celebrated effect to long-range quantum spin systems, demonstrating long-range spin chains can also host Efimov states with widely tunable scaling ratios controlled by the interaction range.',
        'Abstract In this work, different varieties of dye sensitized solar cells are fabricated by simple fabrication process. In this fabrication extract of butea monosperma flower, methylene blue and methyl orange dyes are used as sensitizers. The photovoltaic performance of dye sensitized solar cells (DSSCs) has been studied. The performances of two different types of photo-electrodes are also tested in this work. The morphology and bandgap of TiO 2 (titanium dioxide) and ZnO (Zinc oxide) was observed from XRD, FTIR spectroscopy and UV-vis Spectrum. It is found that TiO 2 based DSSCs have better performance. It also observed that the current density and efficiency was increased from 7.46 to 12.9 mA/cm 2 and from 1.34 to 6.8% respectively when using methyl orange as a dye. Hence it can be said that methyl orange dye enhanced the photovoltaic performance of DSSC.',
        "The slogan ``information is physical,'' introduced by Rolf Landauer and developed through quantum information theory and black-hole thermodynamics, has achieved near-axiomatic status in modern physics. Yet the ontological status of information remains surprisingly underexamined: most discussions either reduce information to a form of energy or treat it as a purely mathematical object. This paper proposes a third position. I argue that information is neither a physical substance nor a free-floating abstraction, but rather \\emph{the structure of physically realizable alternatives} -- a counterfactual structure that a physical system instantiates in virtue of the possibility space available to it. Building on Shannon's combinatorial definition, the Landauer principle, the no-cloning theorem, and the black-hole information paradox, I show that the informational content of any physical event is constituted by the set of outcomes that \\emph{could have occurred} but did not. This counterfactual reading dissolves several persistent confusions: it explains why erasing information dissipates heat without making information ``material,'' why quantum superposition is informationally richer than any classical mixture, and why information loss in black holes is physically significant beyond mere bookkeeping. The proposal sits within a structural-realist framework but departs from standard structural realism by locating the relevant structure in modal, not merely actual, relations. I conclude by sketching implications for the foundations of quantum mechanics, quantum gravity, and scientific ontology more broadly.",
        'Abstract Efficient solar-to-hydrogen conversion remains challenging due to the limited visible-light activity and rapid charge recombination in conventional photocatalysts. Here, we report a plasmon-enhanced nitrogen-doped niobium pentoxide (N–Nb₂O₅) photocatalyst decorated with gold nanoparticles (Au@N–Nb₂O₅) for high-performance solar-driven photocatalytic hydrogen generation. N–Nb₂O₅ was synthesized via a simple wet-chemical route and calcined at 500 °C, forming crystalline orthorhombic nanoplates with XRD-derived crystallite sizes of 50–55 nm and lateral dimensions of ~ 150 nm (FE-SEM). Gold nanoparticles were subsequently deposited via photodeposition, extending visible-light absorption and narrowing the band gap to 2.3–2.5 eV (UV–DRS). XPS analysis confirmed successful nitrogen incorporation and surface metallization. Photoluminescence studies revealed efficient charge separation and concentration-dependent suppression of radiative recombination. The optimised Au@N–Nb₂O₅ (2 wt% Au) achieved a hydrogen evolution rate of 2168 µmol h⁻ 1 g⁻ 1 under natural sunlight, nearly fourfold higher than pristine Nb₂O₅ and N–Nb₂O₅. The ordered nanoplate morphology facilitates charge transport, complementing the interfacial effects of Au, providing a scalable strategy for designing high-performance Nb₂O₅-based photocatalysts for sustainable solar water splitting.',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Reranking

Datasets: NanoSCIDOCS_R100, NanoSciFact_R100, NanoNFCorpus_R100 and chowder-val

Evaluated with CrossEncoderRerankingEvaluator with these parameters:

{
    "at_k": 10,
    "always_rerank_positives": true
}

Metric	NanoSCIDOCS_R100	NanoSciFact_R100	NanoNFCorpus_R100	chowder-val
map	0.3296 (+0.0553)	0.7490 (+0.0792)	0.3709 (+0.1099)	0.9965 (+0.8283)
mrr@10	0.6280 (+0.0685)	0.7594 (+0.0813)	0.6541 (+0.1542)	0.9965 (+0.8634)
ndcg@10	0.4096 (+0.0745)	0.7919 (+0.0820)	0.4283 (+0.1032)	0.9974 (+0.7864)

Cross Encoder Nano BEIR

Dataset: NanoBEIR_R100_mean

Evaluated with CrossEncoderNanoBEIREvaluator with these parameters:

{
    "dataset_names": [
        "scidocs",
        "scifact",
        "nfcorpus"
    ],
    "dataset_id": "sentence-transformers/NanoBEIR-en",
    "rerank_k": 100,
    "at_k": 10,
    "always_rerank_positives": true
}

Metric	Value
map	0.4832 (+0.0815)
mrr@10	0.6805 (+0.1014)
ndcg@10	0.5433 (+0.0866)

Training Details

Training Dataset

Unnamed Dataset

Size: 424,009 training samples
Columns: anchor, positive, and label

Approximate statistics based on the first 100 samples:

	anchor	positive	label
type	string	string	int
modality	text	text
details	min: 5 tokens mean: 20.72 tokens max: 40 tokens	min: 61 tokens mean: 297.62 tokens max: 512 tokens	0: ~80.77% 1: ~19.23%

Samples:

anchor	positive	label
`Excitation of non-modal perturbations in hypersonic boundary layers by free stream forcing. Part II: asymptotic theory and key mechanisms`	Recently, Zhao & Dong (J. Fluid Mech. 2025, vol. 1013: A44) developed a high-efficiency, high-accuracy numerical framework, the shock-fitting harmonic linearised Navier-Stokes (SF-HLNS) approach, which enables a systematic study of the receptivity of non-modal perturbations in hypersonic blunt-body boundary layers over a wide parameter range. In this Part II, we employ a high-Reynolds-number asymptotic analysis to elucidate the physical mechanism of the receptivity process. A distinct slow-down convection mechanism is identified in the nose region, amplifying the perturbation streamwise vorticity from the post-shock position to the boundary layer around the stagnation point by a factor of O(\sqrt{R}), where R is the Reynolds number based on nose radius. Downstream, the lift-up mechanism further leads to a transient growth of the perturbation streamwise velocity up to an amplitude of O(R). Based on these mechanisms, a reduced model is developed to predict the downstream evolution of the...	`1`
`A non-intrusive approach to index-aware learning`	Large language models (LLMs) offer a natural-language interface for interpreting Internet of Things (IoT) sensor data in smart environments; however, cloud deployment introduces latency, privacy, and connectivity concerns. Local LLMs can reduce these limitations, but compact edge-deployable models often show weaker numerical reasoning when raw sensor readings are provided directly. This paper investigates whether prompt-side preprocessing can improve the accuracy-latency trade-off of local LLMs for environmental monitoring. We propose a structured prompt construction framework that transforms raw air-quality and thermal-comfort measurements into progressively enriched textual representations: raw sensor values, threshold-aware descriptions, and compact environmental summary flags. The approach is evaluated using indoor Raspberry Pi/BME680 datasets from Tampere University and outdoor air-quality datasets from Helsinki, Katowice, and Warsaw. We construct a binary LLM query dataset coveri...	`0`
`A note on the transcendental basepoint-free conjecture for Calabi-Yau manifolds`	We study the Hulek--Verrill families of Calabi--Yau threefolds. They are birationally equivalent to fibred products of elliptic surfaces, so we expect to be able to compute periods on these threefolds by integrating products of elliptic periods over a contour on $\mathbb{P}^1$. We numerically verify this in several examples. This article was submitted to MATRIX Annals (2024) for inclusion in the proceedings of the conference "The Geometry of Moduli Spaces in String Theory", held 2--13 September 2024.	`0`

Loss: BinaryCrossEntropyLoss with these parameters:

{
    "activation_fn": "torch.nn.modules.linear.Identity",
    "pos_weight": 4.9989399909973145
}

Evaluation Dataset

Unnamed Dataset

Size: 500 evaluation samples
Columns: anchor, positive, and label

Approximate statistics based on the first 100 samples:

	anchor	positive	label
type	string	string	int
modality	text	text
details	min: 6 tokens mean: 21.94 tokens max: 48 tokens	min: 33 tokens mean: 289.3 tokens max: 512 tokens	0: ~82.69% 1: ~17.31%

Samples:

anchor	positive	label
`Estimating the safe mud weight window for drilling operations through pre-stack seismic inversion, a case study in one of the Southwestern Iran oil fields`	High-temperature drilling environments often intensify rock–fluid interactions, leading to substantial formation damage primarily due to excessive mud filtrate invasion. Conventional additives employed to mitigate this issue frequently exhibit thermal degradation, limiting their effectiveness under elevated temperatures. Notably, the literature reveals a significant gap concerning the invasion behavior of bentonite-free mud filtrate within payzone sections which is considered a critical aspect for optimizing hydrocarbon recovery. This study investigates the influence of drilling mud filtrate–rock interactions on rock strength and porosity at high temperatures through an integrated approach combining laboratory experiments and computational fluid dynamics (CFD) simulations. Rock strength of sandstone and limestone was quantified using unconfined compressive strength (UCS) tests. Experimental findings indicate a significant reduction in strength following filtrate exposure. Bandera Brown...	`0`
`Achieving EF1 and Epistemic EFX Guarantees Simultaneously`	`The Efimov effect is a quantum phenomenon in which three particles form an infinite series of low-energy bound states at resonance, previously observed in three-dimensional systems. The authors extend this celebrated effect to long-range quantum spin systems, demonstrating long-range spin chains can also host Efimov states with widely tunable scaling ratios controlled by the interaction range.`	`0`
`Study on dye-sensitized solar cell efficiency improvement using methyl orange dye`	Abstract In this work, different varieties of dye sensitized solar cells are fabricated by simple fabrication process. In this fabrication extract of butea monosperma flower, methylene blue and methyl orange dyes are used as sensitizers. The photovoltaic performance of dye sensitized solar cells (DSSCs) has been studied. The performances of two different types of photo-electrodes are also tested in this work. The morphology and bandgap of TiO 2 (titanium dioxide) and ZnO (Zinc oxide) was observed from XRD, FTIR spectroscopy and UV-vis Spectrum. It is found that TiO 2 based DSSCs have better performance. It also observed that the current density and efficiency was increased from 7.46 to 12.9 mA/cm 2 and from 1.34 to 6.8% respectively when using methyl orange as a dye. Hence it can be said that methyl orange dye enhanced the photovoltaic performance of DSSC.	`1`

Loss: BinaryCrossEntropyLoss with these parameters:

{
    "activation_fn": "torch.nn.modules.linear.Identity",
    "pos_weight": 4.9989399909973145
}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 64
learning_rate: 2e-05
warmup_steps: 0.1
weight_decay: 0.01
bf16: True
per_device_eval_batch_size: 64
load_best_model_at_end: True
seed: 12

All Hyperparameters

Click to expand

per_device_train_batch_size: 64
num_train_epochs: 3
max_steps: -1
learning_rate: 2e-05
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_steps: 0.1
optim: adamw_torch_fused
optim_args: None
weight_decay: 0.01
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: 64
prediction_loss_only: True
eval_on_start: False
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: True
ignore_data_skip: False
restore_callback_states_from_checkpoint: False
full_determinism: False
seed: 12
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
dataloader_multiprocessing_context: None
dataloader_in_order: True
remove_unused_columns: True
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: None
fsdp_config: None
deepspeed: None
debug: []
skip_memory_metrics: True
do_predict: False
resume_from_checkpoint: None
local_rank: -1
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}
warmup_ratio: None

Training Logs

Epoch	Step	Training Loss	Validation Loss	NanoSCIDOCS_R100_ndcg@10	NanoSciFact_R100_ndcg@10	NanoNFCorpus_R100_ndcg@10	NanoBEIR_R100_mean_ndcg@10	chowder-val_ndcg@10
-1	-1	-	-	0.3527 (+0.0176)	0.7424 (+0.0325)	0.3925 (+0.0675)	0.4959 (+0.0392)	0.9979 (+0.7869)
0.0002	1	0.7125	-	-	-	-	-	-
0.0300	199	0.2834	-	-	-	-	-	-
0.0601	398	0.2766	-	-	-	-	-	-
0.0901	597	0.2554	-	-	-	-	-	-
0.1201	796	0.2408	-	-	-	-	-	-
0.1502	995	0.1754	-	-	-	-	-	-
0.1802	1194	0.1818	-	-	-	-	-	-
0.2102	1393	0.1616	-	-	-	-	-	-
0.2403	1592	0.1621	-	-	-	-	-	-
0.2703	1791	0.1467	-	-	-	-	-	-
0.3	1988	-	0.1059	0.4096 (+0.0745)	0.7919 (+0.0820)	0.4283 (+0.1032)	0.5433 (+0.0866)	-
0.3003	1990	0.1526	-	-	-	-	-	-
0.3304	2189	0.1658	-	-	-	-	-	-
0.3604	2388	0.1504	-	-	-	-	-	-
0.3904	2587	0.1384	-	-	-	-	-	-
0.4205	2786	0.1490	-	-	-	-	-	-
0.4505	2985	0.1576	-	-	-	-	-	-
0.4805	3184	0.1317	-	-	-	-	-	-
0.5106	3383	0.1365	-	-	-	-	-	-
0.5406	3582	0.1301	-	-	-	-	-	-
0.5706	3781	0.1424	-	-	-	-	-	-
0.6001	3976	-	0.0889	0.4179 (+0.0828)	0.7597 (+0.0498)	0.4247 (+0.0997)	0.5341 (+0.0774)	-
0.6007	3980	0.1291	-	-	-	-	-	-
0.6307	4179	0.1469	-	-	-	-	-	-
0.6607	4378	0.1253	-	-	-	-	-	-
0.6908	4577	0.1302	-	-	-	-	-	-
0.7208	4776	0.1475	-	-	-	-	-	-
0.7508	4975	0.1370	-	-	-	-	-	-
0.7809	5174	0.1307	-	-	-	-	-	-
0.8109	5373	0.1409	-	-	-	-	-	-
0.8409	5572	0.1305	-	-	-	-	-	-
0.8710	5771	0.1363	-	-	-	-	-	-
0.9001	5964	-	0.0749	0.4243 (+0.0891)	0.7458 (+0.0359)	0.4299 (+0.1048)	0.5333 (+0.0766)	-
0.9010	5970	0.1263	-	-	-	-	-	-
0.9310	6169	0.1270	-	-	-	-	-	-
0.9611	6368	0.1334	-	-	-	-	-	-
0.9911	6567	0.1416	-	-	-	-	-	-
1.0211	6766	0.1064	-	-	-	-	-	-
1.0512	6965	0.1132	-	-	-	-	-	-
1.0812	7164	0.1002	-	-	-	-	-	-
1.1112	7363	0.0942	-	-	-	-	-	-
1.1413	7562	0.1065	-	-	-	-	-	-
1.1713	7761	0.0965	-	-	-	-	-	-
1.2001	7952	-	0.0809	0.4138 (+0.0787)	0.7118 (+0.0019)	0.4326 (+0.1075)	0.5194 (+0.0627)	-
-1	-1	-	-	0.4096 (+0.0745)	0.7919 (+0.0820)	0.4283 (+0.1032)	0.5433 (+0.0866)	0.9974 (+0.7864)

The bold row denotes the saved checkpoint.

Training Time

Training: 41.1 minutes
Evaluation: 1.1 minutes
Total: 42.3 minutes

Framework Versions

Python: 3.14.7
Sentence Transformers: 5.7.0
Transformers: 5.15.0
PyTorch: 2.13.0+cu130
Accelerate: 1.14.0
Datasets: 5.0.1
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",
}