123BERT / README.md

kselight

Add new CrossEncoder model

bd35c30 verified 4 months ago

preview code

raw

history blame contribute delete

25 kB

metadata

language:
  - en
tags:
  - sentence-transformers
  - cross-encoder
  - reranker
  - generated_from_trainer
  - dataset_size:10000
  - loss:MSELoss
datasets:
  - sentence-transformers/msmarco
pipeline_tag: text-ranking
library_name: sentence-transformers
metrics:
  - map
  - mrr@10
  - ndcg@10
model-index:
  - name: CrossEncoder
    results:
      - task:
          type: cross-encoder-reranking
          name: Cross Encoder Reranking
        dataset:
          name: NanoMSMARCO R100
          type: NanoMSMARCO_R100
        metrics:
          - type: map
            value: 0.0579
            name: Map
          - type: mrr@10
            value: 0.0329
            name: Mrr@10
          - type: ndcg@10
            value: 0.0479
            name: Ndcg@10
      - task:
          type: cross-encoder-reranking
          name: Cross Encoder Reranking
        dataset:
          name: NanoNFCorpus R100
          type: NanoNFCorpus_R100
        metrics:
          - type: map
            value: 0.2867
            name: Map
          - type: mrr@10
            value: 0.4222
            name: Mrr@10
          - type: ndcg@10
            value: 0.2546
            name: Ndcg@10
      - task:
          type: cross-encoder-reranking
          name: Cross Encoder Reranking
        dataset:
          name: NanoNQ R100
          type: NanoNQ_R100
        metrics:
          - type: map
            value: 0.0326
            name: Map
          - type: mrr@10
            value: 0.01
            name: Mrr@10
          - type: ndcg@10
            value: 0.0229
            name: Ndcg@10
      - task:
          type: cross-encoder-nano-beir
          name: Cross Encoder Nano BEIR
        dataset:
          name: NanoBEIR R100 mean
          type: NanoBEIR_R100_mean
        metrics:
          - type: map
            value: 0.1257
            name: Map
          - type: mrr@10
            value: 0.155
            name: Mrr@10
          - type: ndcg@10
            value: 0.1084
            name: Ndcg@10

CrossEncoder

This is a Cross Encoder model trained on the msmarco dataset 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
Maximum Sequence Length: 512 tokens
Number of Output Labels: 1 label
Training Dataset:
- msmarco
Language: en

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("kselight/123BERT")
# Get scores for pairs of texts
pairs = [
    ['what is ivana trump', 'The need for an independent investigation. As it stands, all three men in charge of the investigations into the Trump campaign are Republicans, and two of the three are vociferous Trump allies. Burr, the third, also tied himself to Trump during his close 2016 reelection campaign.'],
    ["hogan's goat meaning", 'hoganâ\x80\x99s goat. The phrase like Hoganâ\x80\x99s goat refers to something that is faulty, messed up, or stinks like a goat. The phrase is a reference to R.F. Outcaultâ\x80\x99s seminal newspaper comic Hoganâ\x80\x99s Alley, which debuted in 1895. The title of the strip changed to The Yellow Kid the following year.'],
    ['who made tokyo ghoul', "Tokyo Ghoul (Japanese: æ\x9d±äº¬å\x96°ç¨®ï¼\x88ã\x83\x88ã\x83¼ã\x82\xadã\x83§ã\x83¼ã\x82°ã\x83¼ã\x83«ï¼\x89, Hepburn: TÅ\x8dkyÅ\x8d GÅ«ru) is a Japanese manga series by Sui Ishida. It was serialized in Shueisha's seinen manga magazine Weekly Young Jump between September 2011 and September 2014 and has been collected in fourteen tankÅ\x8dbon volumes as of August 2014."],
    ['neck of the scottie dog', 'Classical guitars. The classical guitar neck blank is relatively small compared to what is needed for construction. This is because a classical neck is constructed differently than most other neck designs. The heel of the neck is built up by stacking blocks of wood to achieve the necessary height.'],
    ['what does bicameral mean in government', 'Top 10 amazing movie makeup transformations. In government, bicameralism is the practice of having two legislative or parliamentary chambers. The relationship between the two chambers of a bicameral legislature can vary. In some cases, they have equal power, and in others, one chamber is clearly superior to the other. It is commonplace in most federal systems to have a bicameral legislature.'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'what is ivana trump',
    [
        'The need for an independent investigation. As it stands, all three men in charge of the investigations into the Trump campaign are Republicans, and two of the three are vociferous Trump allies. Burr, the third, also tied himself to Trump during his close 2016 reelection campaign.',
        'hoganâ\x80\x99s goat. The phrase like Hoganâ\x80\x99s goat refers to something that is faulty, messed up, or stinks like a goat. The phrase is a reference to R.F. Outcaultâ\x80\x99s seminal newspaper comic Hoganâ\x80\x99s Alley, which debuted in 1895. The title of the strip changed to The Yellow Kid the following year.',
        "Tokyo Ghoul (Japanese: æ\x9d±äº¬å\x96°ç¨®ï¼\x88ã\x83\x88ã\x83¼ã\x82\xadã\x83§ã\x83¼ã\x82°ã\x83¼ã\x83«ï¼\x89, Hepburn: TÅ\x8dkyÅ\x8d GÅ«ru) is a Japanese manga series by Sui Ishida. It was serialized in Shueisha's seinen manga magazine Weekly Young Jump between September 2011 and September 2014 and has been collected in fourteen tankÅ\x8dbon volumes as of August 2014.",
        'Classical guitars. The classical guitar neck blank is relatively small compared to what is needed for construction. This is because a classical neck is constructed differently than most other neck designs. The heel of the neck is built up by stacking blocks of wood to achieve the necessary height.',
        'Top 10 amazing movie makeup transformations. In government, bicameralism is the practice of having two legislative or parliamentary chambers. The relationship between the two chambers of a bicameral legislature can vary. In some cases, they have equal power, and in others, one chamber is clearly superior to the other. It is commonplace in most federal systems to have a bicameral legislature.',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Reranking

Datasets: NanoMSMARCO_R100, NanoNFCorpus_R100 and NanoNQ_R100

Evaluated with CrossEncoderRerankingEvaluator with these parameters:

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

Metric	NanoMSMARCO_R100	NanoNFCorpus_R100	NanoNQ_R100
map	0.0579 (-0.4317)	0.2867 (+0.0257)	0.0326 (-0.3870)
mrr@10	0.0329 (-0.4446)	0.4222 (-0.0777)	0.0100 (-0.4167)
ndcg@10	0.0479 (-0.4925)	0.2546 (-0.0705)	0.0229 (-0.4778)

Cross Encoder Nano BEIR

Dataset: NanoBEIR_R100_mean

Evaluated with CrossEncoderNanoBEIREvaluator with these parameters:

{
    "dataset_names": [
        "msmarco",
        "nfcorpus",
        "nq"
    ],
    "rerank_k": 100,
    "at_k": 10,
    "always_rerank_positives": true
}

Metric	Value
map	0.1257 (-0.2643)
mrr@10	0.1550 (-0.3130)
ndcg@10	0.1084 (-0.3469)

Training Details

Training Dataset

msmarco

Dataset: msmarco at 9e329ed
Size: 10,000 training samples
Columns: score, query, and passage

Approximate statistics based on the first 1000 samples:

	score	query	passage
type	float	string	string
details	min: -11.79 mean: 0.58 max: 11.1	min: 9 characters mean: 34.21 characters max: 140 characters	min: 70 characters mean: 342.2 characters max: 894 characters

Samples:

score	query	passage
`6.720487356185913`	`modern definition of democracy`	Links. A Short Definition of Democracy U.S. president Abraham Lincoln (1809-1865) defined democracy as: Â«Government of the people, by the people, for the peopleÂ» Democracy is by far the most challenging form of government-both for politicians and for the people.The term democracy comes from the Greek language and means rule by the (simple) people. The so-called democracies in classical antiquity (Athens and Rome) represent precursors of modern democracies.Like modern democracy, they were created as a reaction to a concentration and abuse of power by the rulers.he term democracy comes from the Greek language and means rule by the (simple) people. The so-called democracies in classical antiquity (Athens and Rome) represent precursors of modern democracies.
`1.6529417037963867`	`is celexa and fluoxetine same`	`Celexa (citalopram hydrobromide) is a type of antidepressant called a selective serotonin reuptake inhibitor (SSRI) indicated for the treatment of depression. Celexa is available in generic form. Common side effects of Celexa include. constipation, nausea, diarrhea, upset stomach, decreased sexual desire,`
`-9.121654828389486`	`what are 2 examples of nonpoint pollution`	`Concept of pollution tax. All such measures are compensatory in nature and it is not called pollution tax. The concept of pollution tax is something different. It entails that instead of doing offsetting work by yourself wherever you hurt environment either willfully or without any intention you have to pay for it.`

Loss: MSELoss with these parameters:

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

Evaluation Dataset

msmarco

Dataset: msmarco at 9e329ed
Size: 1,000 evaluation samples
Columns: score, query, and passage

Approximate statistics based on the first 1000 samples:

	score	query	passage
type	float	string	string
details	min: -11.85 mean: 1.11 max: 11.15	min: 7 characters mean: 34.0 characters max: 186 characters	min: 76 characters mean: 343.66 characters max: 944 characters

Samples:

score	query	passage
`-11.078993638356527`	`what is ivana trump`	`The need for an independent investigation. As it stands, all three men in charge of the investigations into the Trump campaign are Republicans, and two of the three are vociferous Trump allies. Burr, the third, also tied himself to Trump during his close 2016 reelection campaign.`
`8.86651055018107`	`hogan's goat meaning`	`hoganâs goat. The phrase like Hoganâs goat refers to something that is faulty, messed up, or stinks like a goat. The phrase is a reference to R.F. Outcaultâs seminal newspaper comic Hoganâs Alley, which debuted in 1895. The title of the strip changed to The Yellow Kid the following year.`
`8.381712992986044`	`who made tokyo ghoul`	`Tokyo Ghoul (Japanese: æ±äº¬å°ç¨®ï¼ãã¼ãã§ã¼ã°ã¼ã«ï¼, Hepburn: TÅkyÅ GÅ«ru) is a Japanese manga series by Sui Ishida. It was serialized in Shueisha's seinen manga magazine Weekly Young Jump between September 2011 and September 2014 and has been collected in fourteen tankÅbon volumes as of August 2014.`

Loss: MSELoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 16
per_device_eval_batch_size: 16
learning_rate: 8e-06
num_train_epochs: 1
warmup_ratio: 0.1
seed: 12
dataloader_num_workers: 4
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: 16
per_device_eval_batch_size: 16
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: 8e-06
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 1
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: 12
data_seed: None
jit_mode_eval: False
use_ipex: False
bf16: False
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: False
dataloader_num_workers: 4
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
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
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
use_liger_kernel: False
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	NanoMSMARCO_R100_ndcg@10	NanoNFCorpus_R100_ndcg@10	NanoNQ_R100_ndcg@10	NanoBEIR_R100_mean_ndcg@10
-1	-1	-	0.0479 (-0.4925)	0.2546 (-0.0705)	0.0229 (-0.4778)	0.1084 (-0.3469)
0.0064	1	53.6175	-	-	-	-

Framework Versions

Python: 3.11.6
Sentence Transformers: 5.1.1
Transformers: 4.47.1
PyTorch: 2.4.0+cu124
Accelerate: 1.5.1
Datasets: 3.3.2
Tokenizers: 0.21.1

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",
}