langcache-embed-v3 / README.md

radoslavralev

Add new SentenceTransformer model

91abacf verified 2 days ago

preview code

raw

history blame contribute delete

20.2 kB

metadata

language:
  - en
license: apache-2.0
tags:
  - biencoder
  - sentence-transformers
  - text-classification
  - sentence-pair-classification
  - semantic-similarity
  - semantic-search
  - retrieval
  - reranking
  - generated_from_trainer
  - dataset_size:1460771
  - loss:ArcFaceInBatchLoss
base_model: Alibaba-NLP/gte-modernbert-base
widget:
  - source_sentence: >-
      "How much would I need to narrate a ""Let's Play"" video in order to make
      money from it on YouTube?"
    sentences:
      - How much money do people make from YouTube videos with 1 million views?
      - >-
        "How much would I need to narrate a ""Let's Play"" video in order to
        make money from it on YouTube?"
      - '"Does the sentence, ""I expect to be disappointed,"" make sense?"'
  - source_sentence: '"I appreciate that.'
    sentences:
      - >-
        "How is the Mariner rewarded in ""The Rime of the Ancient Mariner"" by
        Samuel Taylor Coleridge?"
      - '"I appreciate that.'
      - I can appreciate that.
  - source_sentence: >-
      """It is very easy to defeat someone, but too hard to win some one"". What
      does the previous sentence mean?"
    sentences:
      - '"How can you use the word ""visceral"" in a sentence?"'
      - >-
        """It is very easy to defeat someone, but too hard to win some one"".
        What does the previous sentence mean?"
      - >-
        "What does ""The loudest one in the room is the weakest one in the
        room."" Mean?"
  - source_sentence: >-
      " We condemn this raid which is in our view illegal and morally and
      politically unjustifiable , " London-based NCRI official Ali Safavi told
      Reuters by telephone .
    sentences:
      - >-
        London-based NCRI official Ali Safavi told Reuters : " We condemn this
        raid , which is in our view illegal and morally and politically
        unjustifiable . "
      - >-
        The social awkwardness is complicated by the fact that Marianne is a
        white girl living with a black family .
      - art's cause, this in my opinion
  - source_sentence: >-
      "If you click ""like"" on an old post that someone made on your wall yet
      you're no longer Facebook friends, will they still receive a
      notification?"
    sentences:
      - >-
        "Is there is any two wheeler having a gear box which has the feature
        ""automatic neutral"" when the engine is off while it is in gear?"
      - >-
        "If you click ""like"" on an old post that someone made on your wall yet
        you're no longer Facebook friends, will they still receive a
        notification?"
      - >-
        "If your teenage son posted ""La commedia e finita"" on his Facebook
        wall, would you be concerned?"
datasets:
  - redis/langcache-sentencepairs-v2
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - cosine_accuracy@1
  - cosine_precision@1
  - cosine_recall@1
  - cosine_ndcg@10
  - cosine_mrr@1
  - cosine_map@100
  - cosine_auc_precision_cache_hit_ratio
  - cosine_auc_similarity_distribution
model-index:
  - name: Redis fine-tuned BiEncoder model for semantic caching on LangCache
    results:
      - task:
          type: custom-information-retrieval
          name: Custom Information Retrieval
        dataset:
          name: test
          type: test
        metrics:
          - type: cosine_accuracy@1
            value: 0.5880558568329718
            name: Cosine Accuracy@1
          - type: cosine_precision@1
            value: 0.5880558568329718
            name: Cosine Precision@1
          - type: cosine_recall@1
            value: 0.5707119922832199
            name: Cosine Recall@1
          - type: cosine_ndcg@10
            value: 0.771771481653434
            name: Cosine Ndcg@10
          - type: cosine_mrr@1
            value: 0.5880558568329718
            name: Cosine Mrr@1
          - type: cosine_map@100
            value: 0.7214095423928245
            name: Cosine Map@100
          - type: cosine_auc_precision_cache_hit_ratio
            value: 0.35287530778716975
            name: Cosine Auc Precision Cache Hit Ratio
          - type: cosine_auc_similarity_distribution
            value: 0.16742922746173
            name: Cosine Auc Similarity Distribution

Redis fine-tuned BiEncoder model for semantic caching on LangCache

This is a sentence-transformers model finetuned from Alibaba-NLP/gte-modernbert-base on the LangCache Sentence Pairs (all) dataset. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for sentence pair similarity.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: Alibaba-NLP/gte-modernbert-base
Maximum Sequence Length: 100 tokens
Output Dimensionality: 768 dimensions
Similarity Function: Cosine Similarity
Training Dataset:
- LangCache Sentence Pairs (all)
Language: en
License: apache-2.0

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 100, 'do_lower_case': False, 'architecture': 'ModernBertModel'})
  (1): Pooling({'word_embedding_dimension': 768, '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})
  (mlp_hidden): Dense({'in_features': 768, 'out_features': 768, 'bias': True, 'activation_function': 'torch.nn.modules.activation.ReLU'})
  (mlp_out): Dense({'in_features': 768, 'out_features': 768, 'bias': True, 'activation_function': 'torch.nn.modules.linear.Identity'})
)

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("redis/langcache-embed-v3")
# Run inference
sentences = [
    '"If you click ""like"" on an old post that someone made on your wall yet you\'re no longer Facebook friends, will they still receive a notification?"',
    '"If you click ""like"" on an old post that someone made on your wall yet you\'re no longer Facebook friends, will they still receive a notification?"',
    '"If your teenage son posted ""La commedia e finita"" on his Facebook wall, would you be concerned?"',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 1.0000, 0.2617],
#         [1.0000, 1.0000, 0.2617],
#         [0.2617, 0.2617, 1.0000]])

Evaluation

Metrics

Custom Information Retrieval

Dataset: test
Evaluated with ir_evaluator.CustomInformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.5881
cosine_precision@1	0.5881
cosine_recall@1	0.5707
cosine_ndcg@10	0.7718
cosine_mrr@1	0.5881
cosine_map@100	0.7214
cosine_auc_precision_cache_hit_ratio	0.3529
cosine_auc_similarity_distribution	0.1674

Training Details

Training Dataset

LangCache Sentence Pairs (all)

Dataset: LangCache Sentence Pairs (all)
Size: 132,354 training samples
Columns: anchor, positive, and negative

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative
type	string	string	string
details	min: 4 tokens mean: 25.33 tokens max: 100 tokens	min: 4 tokens mean: 24.98 tokens max: 100 tokens	min: 5 tokens mean: 19.06 tokens max: 68 tokens

Samples:

anchor	positive	negative
`What high potential jobs are there other than computer science?`	`What high potential jobs are there other than computer science?`	`Why IT or Computer Science jobs are being over rated than other Engineering jobs?`
`Would India ever be able to develop a missile system like S300 or S400 missile?`	`Would India ever be able to develop a missile system like S300 or S400 missile?`	`Should India buy the Russian S400 air defence missile system?`
`water from the faucet is being drunk by a yellow dog`	`A yellow dog is drinking water from the faucet`	`Childlessness is low in Eastern European countries.`

Loss: losses.ArcFaceInBatchLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim",
    "gather_across_devices": false
}

Evaluation Dataset

LangCache Sentence Pairs (all)

Dataset: LangCache Sentence Pairs (all)
Size: 132,354 evaluation samples
Columns: anchor, positive, and negative

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative
type	string	string	string
details	min: 4 tokens mean: 25.33 tokens max: 100 tokens	min: 4 tokens mean: 24.98 tokens max: 100 tokens	min: 5 tokens mean: 19.06 tokens max: 68 tokens

Samples:

anchor	positive	negative
`What high potential jobs are there other than computer science?`	`What high potential jobs are there other than computer science?`	`Why IT or Computer Science jobs are being over rated than other Engineering jobs?`
`Would India ever be able to develop a missile system like S300 or S400 missile?`	`Would India ever be able to develop a missile system like S300 or S400 missile?`	`Should India buy the Russian S400 air defence missile system?`
`water from the faucet is being drunk by a yellow dog`	`A yellow dog is drinking water from the faucet`	`Childlessness is low in Eastern European countries.`

Loss: losses.ArcFaceInBatchLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim",
    "gather_across_devices": false
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 8192
per_device_eval_batch_size: 8192
gradient_accumulation_steps: 2
weight_decay: 0.001
adam_beta2: 0.98
adam_epsilon: 1e-06
num_train_epochs: 1
warmup_ratio: 0.05
bf16: True
dataloader_num_workers: 4
dataloader_prefetch_factor: 4
load_best_model_at_end: True
optim: stable_adamw
ddp_find_unused_parameters: False
dataloader_persistent_workers: True
push_to_hub: True
hub_model_id: redis/langcache-embed-v3
eval_on_start: True
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 8192
per_device_eval_batch_size: 8192
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 2
eval_accumulation_steps: None
torch_empty_cache_steps: None
learning_rate: 5e-05
weight_decay: 0.001
adam_beta1: 0.9
adam_beta2: 0.98
adam_epsilon: 1e-06
max_grad_norm: 1.0
num_train_epochs: 1
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.05
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: False
dataloader_num_workers: 4
dataloader_prefetch_factor: 4
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}
parallelism_config: None
deepspeed: None
label_smoothing_factor: 0.0
optim: stable_adamw
optim_args: None
adafactor: False
group_by_length: False
length_column_name: length
ddp_find_unused_parameters: False
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
dataloader_pin_memory: True
dataloader_persistent_workers: True
skip_memory_metrics: True
use_legacy_prediction_loop: False
push_to_hub: True
resume_from_checkpoint: None
hub_model_id: redis/langcache-embed-v3
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: True
use_liger_kernel: False
liger_kernel_config: None
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: None
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Validation Loss	test_cosine_ndcg@10
0	0	2.9916	0.7718

Framework Versions

Python: 3.12.3
Sentence Transformers: 5.1.0
Transformers: 4.56.0
PyTorch: 2.8.0+cu128
Accelerate: 1.10.1
Datasets: 4.0.0
Tokenizers: 0.22.0

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",
}