SentenceTransformer based on Model Distillation

In this experiment with knowledge distillation for embedding models, i retained 8 layers from the teacher model. This is an attempt to create a lighter, faster version.

the top left graph shows how well your model's predictions match reality. Spearman correlation = 0.887
the top right compares the correlation performance of this model vs the reference(mxbai-embed-large-v1) model - both bars around 0.8-0.9
bottom left shows, this model processes about 45 samples/s and mxbai-embed-large-v1 processes about 30 samples/s.
the bottom right shows a small accuracy drop for this model.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: mixedbread-ai/mxbai-embed-large-v1
Maximum Sequence Length: 512 tokens
Output Dimensionality: 1024 dimensions
Similarity Function: Cosine Similarity

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 1024, '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})
)

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("sentence_transformers_model_id")
# Run inference
sentences = [
    'A black dog is drinking next to a brown and white dog that is looking at an orange ball in the lake, whilst a horse and rider passes behind.',
    'Some men with jerseys are in a bar, watching a soccer match.',
    'There are two people running around a track in lane three and the one wearing a blue shirt with a green thing over the eyes is just barely ahead of the guy wearing an orange shirt and sunglasses.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

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

Evaluation

Metrics

Semantic Similarity

Datasets: sts-dev and sts-test
Evaluated with EmbeddingSimilarityEvaluator

Metric	sts-dev	sts-test
pearson_cosine	0.8654	0.834
spearman_cosine	0.8873	0.8724

Knowledge Distillation

Evaluated with MSEEvaluator

Metric	Value
negative_mse	-3.3795

Training Details

Training Dataset

Unnamed Dataset

Size: 1,195,425 training samples
Columns: sentence and label
Approximate statistics based on the first 1000 samples:
sentence label
type string list
details
min: 4 tokens
mean: 12.24 tokens
max: 52 tokens

size: 1024 elements

	sentence	label
type	string	list
details	min: 4 tokens mean: 12.24 tokens max: 52 tokens	size: 1024 elements

Samples:

sentence	label
`A person on a horse jumps over a broken down airplane.`	`[-0.012967385351657867, 0.3716000020503998, 0.252520889043808, 0.7052643299102783, -0.15118499100208282, ...]`
`Children smiling and waving at camera`	`[0.15414997935295105, 0.6666896939277649, -0.3150098919868469, 1.0102407932281494, 0.4113735556602478, ...]`
`A boy is jumping on skateboard in the middle of a red bridge.`	`[-0.2989530563354492, 0.8571284413337708, -0.48532426357269287, 0.8935043215751648, 0.28524795174598694, ...]`

Loss: MSELoss

Evaluation Dataset

Unnamed Dataset

Size: 10,000 evaluation samples
Columns: sentence and label
Approximate statistics based on the first 1000 samples:
sentence label
type string list
details
min: 5 tokens
mean: 13.23 tokens
max: 57 tokens

size: 1024 elements

	sentence	label
type	string	list
details	min: 5 tokens mean: 13.23 tokens max: 57 tokens	size: 1024 elements

Samples:

sentence	label
`Two women are embracing while holding to go packages.`	`[-0.35094621777534485, 0.4337681233882904, 0.22905530035495758, 0.9438946843147278, -1.0199058055877686, ...]`
`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.`	`[-0.37593328952789307, 0.6690596342086792, -0.14921458065509796, 0.7559019923210144, -0.4093412756919861, ...]`
`A man selling donuts to a customer during a world exhibition event held in the city of Angeles`	`[0.21969863772392273, 0.5065202713012695, -0.25664886832237244, 0.2569092810153961, -0.05940837413072586, ...]`

Loss: MSELoss

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 64
per_device_eval_batch_size: 64
learning_rate: 0.0001
num_train_epochs: 1
warmup_ratio: 0.1
fp16: 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: 0.0001
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: 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: 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: False
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

Framework Versions

Python: 3.10.14
Sentence Transformers: 3.3.1
Transformers: 4.46.3
PyTorch: 2.4.0
Accelerate: 1.1.1
Datasets: 3.1.0
Tokenizers: 0.20.3

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",
}

MSELoss

@inproceedings{reimers-2020-multilingual-sentence-bert,
    title = "Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2020",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/2004.09813",
}