Add new CrossEncoder model

d3b35b0 verified about 1 year ago

22.9 kB

tags:
  - sentence-transformers
  - cross-encoder
  - reranker
  - generated_from_trainer
  - dataset_size:942069
  - loss:BinaryCrossEntropyLoss
base_model: microsoft/deberta-v3-base
pipeline_tag: text-ranking
library_name: sentence-transformers
metrics:
  - accuracy
  - accuracy_threshold
  - f1
  - f1_threshold
  - precision
  - recall
  - average_precision
model-index:
  - name: CrossEncoder based on microsoft/deberta-v3-base
    results:
      - task:
          type: cross-encoder-classification
          name: Cross Encoder Classification
        dataset:
          name: AllNLI norm dev
          type: AllNLI-norm-dev
        metrics:
          - type: accuracy
            value: 0.6807244238693595
            name: Accuracy
          - type: accuracy_threshold
            value: 0.4375791847705841
            name: Accuracy Threshold
          - type: f1
            value: 0.5465734265734266
            name: F1
          - type: f1_threshold
            value: 0.00438243243843317
            name: F1 Threshold
          - type: precision
            value: 0.40035514274006284
            name: Precision
          - type: recall
            value: 0.8610458284371327
            name: Recall
          - type: average_precision
            value: 0.49929639749742777
            name: Average Precision
      - task:
          type: cross-encoder-classification
          name: Cross Encoder Classification
        dataset:
          name: AllNLI test
          type: AllNLI-test
        metrics:
          - type: accuracy
            value: 0.6814204314204314
            name: Accuracy
          - type: accuracy_threshold
            value: 0.5599576234817505
            name: Accuracy Threshold
          - type: f1
            value: 0.5269568771714694
            name: F1
          - type: f1_threshold
            value: 0.0010413693962618709
            name: F1 Threshold
          - type: precision
            value: 0.3655438357718045
            name: Precision
          - type: recall
            value: 0.9436392914653784
            name: Recall
          - type: average_precision
            value: 0.48186391164637776
            name: Average Precision

CrossEncoder based on microsoft/deberta-v3-base

This is a Cross Encoder model finetuned from microsoft/deberta-v3-base 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: microsoft/deberta-v3-base
Maximum Sequence Length: 512 tokens
Number of Output Labels: 1 label

Model Sources

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

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("tani-at-nola/reranker-deberta-v3-base-nli")
# Get scores for pairs of texts
pairs = [
    ['The sisters are hugging goodbye while holding to go packages after just eating lunch.', 'Two women are embracing while holding to go packages.'],
    ['Two woman are holding packages.', 'Two women are embracing while holding to go packages.'],
    ['The men are fighting outside a deli.', 'Two women are embracing while holding to go packages.'],
    ['Two kids in numbered jerseys wash their hands.', 'Two young children in blue jerseys, one with the number 9 and one with the number 2 are standing on wooden steps in a bathroom and washing their hands in a sink.'],
    ['Two kids at a ballgame wash their hands.', 'Two young children in blue jerseys, one with the number 9 and one with the number 2 are standing on wooden steps in a bathroom and washing their hands in a sink.'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'The sisters are hugging goodbye while holding to go packages after just eating lunch.',
    [
        'Two women are embracing while holding to go packages.',
        'Two women are embracing while holding to go packages.',
        'Two women are embracing while holding to go packages.',
        'Two young children in blue jerseys, one with the number 9 and one with the number 2 are standing on wooden steps in a bathroom and washing their hands in a sink.',
        'Two young children in blue jerseys, one with the number 9 and one with the number 2 are standing on wooden steps in a bathroom and washing their hands in a sink.',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Classification

Datasets: AllNLI-norm-dev and AllNLI-test
Evaluated with CrossEncoderClassificationEvaluator

Metric	AllNLI-norm-dev	AllNLI-test
accuracy	0.6807	0.6814
accuracy_threshold	0.4376	0.56
f1	0.5466	0.527
f1_threshold	0.0044	0.001
precision	0.4004	0.3655
recall	0.861	0.9436
average_precision	0.4993	0.4819

Training Details

Training Dataset

Unnamed Dataset

Size: 942,069 training samples
Columns: hypothesis, premise, and label

Approximate statistics based on the first 1000 samples:

	hypothesis	premise	label
type	string	string	int
details	min: 11 characters mean: 38.26 characters max: 131 characters	min: 23 characters mean: 69.54 characters max: 227 characters	0: ~66.60% 1: ~33.40%

Samples:

hypothesis	premise	label
`A person is training his horse for a competition.`	`A person on a horse jumps over a broken down airplane.`	`0`
`A person is at a diner, ordering an omelette.`	`A person on a horse jumps over a broken down airplane.`	`0`
`A person is outdoors, on a horse.`	`A person on a horse jumps over a broken down airplane.`	`1`

Loss: BinaryCrossEntropyLoss with these parameters:

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

Evaluation Dataset

Unnamed Dataset

Size: 19,657 evaluation samples
Columns: hypothesis, premise, and label

Approximate statistics based on the first 1000 samples:

	hypothesis	premise	label
type	string	string	int
details	min: 11 characters mean: 37.66 characters max: 116 characters	min: 16 characters mean: 75.01 characters max: 229 characters	0: ~66.90% 1: ~33.10%

Samples:

hypothesis	premise	label
`The sisters are hugging goodbye while holding to go packages after just eating lunch.`	`Two women are embracing while holding to go packages.`	`0`
`Two woman are holding packages.`	`Two women are embracing while holding to go packages.`	`1`
`The men are fighting outside a deli.`	`Two women are embracing while holding to go packages.`	`0`

Loss: BinaryCrossEntropyLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 64
per_device_eval_batch_size: 64
num_train_epochs: 5
warmup_ratio: 0.1
bf16: True
load_best_model_at_end: True

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 64
per_device_eval_batch_size: 64
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: 5
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: True
fp16: False
fp16_opt_level: O1
half_precision_backend: auto
bf16_full_eval: False
fp16_full_eval: False
tf32: None
local_rank: 0
ddp_backend: None
tpu_num_cores: None
tpu_metrics_debug: False
debug: []
dataloader_drop_last: True
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, '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: None
hub_always_push: False
hub_revision: None
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_inputs_for_metrics: False
include_for_metrics: []
eval_do_concat_batches: True
fp16_backend: auto
push_to_hub_model_id: None
push_to_hub_organization: None
mp_parameters:
auto_find_batch_size: False
full_determinism: False
torchdynamo: None
ray_scope: last
ddp_timeout: 1800
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
include_tokens_per_second: False
include_num_input_tokens_seen: False
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
eval_on_start: False
use_liger_kernel: False
liger_kernel_config: None
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	AllNLI-norm-dev_average_precision	AllNLI-test_average_precision
-1	-1	-	-	0.3614	-
0.0068	100	0.7205	-	-	-
0.0136	200	0.6972	-	-	-
0.0204	300	0.6086	-	-	-
0.0272	400	0.4855	-	-	-
0.0340	500	0.3991	-	-	-
0.0408	600	0.3409	-	-	-
0.0476	700	0.2987	-	-	-
0.0544	800	0.2841	-	-	-
0.0611	900	0.2729	-	-	-
0.0679	1000	0.2627	-	-	-
0.0747	1100	0.2517	-	-	-
0.0815	1200	0.2286	-	-	-
0.0883	1300	0.2385	-	-	-
0.0951	1400	0.2329	-	-	-
0.1019	1500	0.2213	0.1959	0.4997	-
0.1087	1600	0.22	-	-	-
0.1155	1700	0.2295	-	-	-
0.1223	1800	0.2236	-	-	-
0.1291	1900	0.2273	-	-	-
0.1359	2000	0.2071	-	-	-
0.1427	2100	0.2254	-	-	-
0.1495	2200	0.2217	-	-	-
0.1563	2300	0.2093	-	-	-
0.1631	2400	0.2112	-	-	-
0.1698	2500	0.2176	-	-	-
0.1766	2600	0.2195	-	-	-
0.1834	2700	0.2107	-	-	-
0.1902	2800	0.2164	-	-	-
0.1970	2900	0.213	-	-	-
0.2038	3000	0.2055	0.1726	0.4789	-
0.2106	3100	0.2039	-	-	-
0.2174	3200	0.2157	-	-	-
0.2242	3300	0.2155	-	-	-
0.2310	3400	0.2017	-	-	-
0.2378	3500	0.2068	-	-	-
0.2446	3600	0.2111	-	-	-
0.2514	3700	0.2062	-	-	-
0.2582	3800	0.2062	-	-	-
0.2650	3900	0.2217	-	-	-
0.2718	4000	0.2012	-	-	-
0.2786	4100	0.2127	-	-	-
0.2853	4200	0.212	-	-	-
0.2921	4300	0.2075	-	-	-
0.2989	4400	0.2099	-	-	-
0.3057	4500	0.2134	0.1644	0.4993	-
-1	-1	-	-	-	0.4819

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.10.12
Sentence Transformers: 5.0.0
Transformers: 4.53.2
PyTorch: 2.7.1+cu126
Accelerate: 1.9.0
Datasets: 4.0.0
Tokenizers: 0.21.2

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",
}