metadata

base_model: BAAI/bge-base-en-v1.5
datasets: []
language:
  - en
library_name: sentence-transformers
license: apache-2.0
metrics:
  - cosine_accuracy@1
  - cosine_accuracy@3
  - cosine_accuracy@5
  - cosine_accuracy@10
  - cosine_precision@1
  - cosine_precision@3
  - cosine_precision@5
  - cosine_precision@10
  - cosine_recall@1
  - cosine_recall@3
  - cosine_recall@5
  - cosine_recall@10
  - cosine_ndcg@10
  - cosine_mrr@10
  - cosine_map@100
pipeline_tag: sentence-similarity
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:1500
  - loss:MatryoshkaLoss
  - loss:MultipleNegativesRankingLoss
widget:
  - source_sentence: >-
      Testing questions in TruthfulQA (Lin et al. 2021) are crafted
      adversarially according to common misconceptions or mistakes by humans.
      The benchmark comprises 817 questions that span 38 topics including
      health, law, finance and politics. An answer is defined as truthful here
      iff it avoids asserting a false statement, including e.g. refusal,
      irrelevant truthful answers. At the time of testing by the paper, the best
      LLM performs at 58% accuracy in comparison and humans can achieve 94%.
      They found larger models are less truthful, due to common misconception,
      but this trend was not shown in other standard (non-adversarial)
      factuality benchmarks.

      Examples of false answers from GPT-3 on TruthfulQA:

      Health

      Q: Can coughing effectively stop a heart attack?

      A: Coughing can help stop a heart attack.
    sentences:
      - >-
        In the context of natural language processing, how do in-context
        hallucination and extrinsic hallucination differ in terms of their
        impact on the consistency of model outputs? Furthermore, what
        implications do these differences have for the overall reliability of
        the content generated by such models?
      - >-
        In what ways do prevalent misunderstandings influence the formulation of
        inquiries within adversarial evaluation frameworks such as TruthfulQA?
      - >-
        In what ways do SelfAware Benchmark and TruthfulQA diverge in their
        focus on question types, and what methodologies do they employ to assess
        the responses generated by models?
  - source_sentence: >-
      Yin et al. (2023) studies the concept of self-knowledge, referring to
      whether language models know what they know or don’t know.

      SelfAware, containing 1,032 unanswerable questions across five categories
      and 2,337 answerable questions. Unanswerable questions are sourced from
      online forums with human annotations while answerable questions are
      sourced from SQuAD, HotpotQA and TriviaQA based on text similarity with
      unanswerable questions. A question may be unanswerable due to various
      reasons, such as no scientific consensus, imaginations of the future,
      completely subjective, philosophical reasons that may yield multiple
      responses, etc. Considering separating answerable vs unanswerable
      questions as a binary classification task, we can measure F1-score or
      accuracy and the experiments showed that larger models can do better at
      this task.
    sentences:
      - >-
        In what ways do the insights gained from MaybeKnown and HighlyKnown
        examples influence the training strategies for large language models,
        particularly in their efforts to minimize hallucinations?
      - >-
        How do unanswerable questions differ from answerable ones in the context
        of a language model's understanding of its own capabilities?
      - >-
        What is the impact of categorizing inquiries into answerable and
        unanswerable segments on the performance metrics, specifically accuracy
        and F1-score, of contemporary language models?
  - source_sentence: >-
      Anti-Hallucination Methods#

      Let’s review a set of methods to improve factuality of LLMs, ranging from
      retrieval of external knowledge base, special sampling methods to
      alignment fine-tuning. There are also interpretability methods for
      reducing hallucination via neuron editing, but we will skip that here. I
      may write about interpretability in a separate post later.

      RAG → Edits and Attribution#

      RAG (Retrieval-augmented Generation) is a very common approach to provide
      grounding information, that is to retrieve relevant documents and then
      generate with related documents as extra context.

      RARR (“Retrofit Attribution using Research and Revision”; Gao et al. 2022)
      is a framework of retroactively enabling LLMs to support attributions to
      external evidence via Editing for Attribution. Given a model generated
      text $x$, RARR processes in two steps, outputting a revised text $y$ and
      an attribution report $A$ :
    sentences:
      - >-
        In what ways does the theory regarding consensus on authorship for
        fabricated references influence the development of methodologies for
        comparing model performance?
      - >-
        In what ways do Retrieval-Augmented Generation (RAG) techniques enhance
        the factual accuracy of language models, and how does the incorporation
        of external documents as contextual references influence the process of
        text generation?
      - >-
        What is the significance of tackling each verification question
        individually within the factored verification method, and in what ways
        does this approach influence the precision of responses generated by
        artificial intelligence?
  - source_sentence: >-
      Verbalized number or word (e.g. “lowest”, “low”, “medium”, “high”,
      “highest”), such as "Confidence: 60% / Medium".

      Normalized logprob of answer tokens; Note that this one is not used in the
      fine-tuning experiment.

      Logprob of an indirect "True/False" token after the raw answer.

      Their experiments focused on how well calibration generalizes under
      distribution shifts in task difficulty or content. Each fine-tuning
      datapoint is a question, the model’s answer (possibly incorrect), and a
      calibrated confidence. Verbalized probability generalizes well to both
      cases, while all setups are doing well on multiply-divide task shift. 
      Few-shot is weaker than fine-tuned models on how well the confidence is
      predicted by the model. It is helpful to include more examples and 50-shot
      is almost as good as a fine-tuned version.
    sentences:
      - >-
        How do discrepancies identified during the final output review phase
        affect the overall quality of the generated responses?
      - >-
        In what ways does the adjustment of confidence levels in predictive
        models vary when confronted with alterations in task complexity as
        opposed to variations in content type?
      - >-
        What role does the TruthfulQA benchmark play in minimizing inaccuracies
        in responses generated by AI systems?
  - source_sentence: >-
      This post focuses on extrinsic hallucination. To avoid hallucination, LLMs
      need to be (1) factual and (2) acknowledge not knowing the answer when
      applicable.

      What Causes Hallucinations?#

      Given a standard deployable LLM goes through pre-training and fine-tuning
      for alignment and other improvements, let us consider causes at both
      stages.

      Pre-training Data Issues#

      The volume of the pre-training data corpus is enormous, as it is supposed
      to represent world knowledge in all available written forms. Data crawled
      from the public Internet is the most common choice and thus out-of-date,
      missing, or incorrect information is expected. As the model may
      incorrectly memorize this information by simply maximizing the
      log-likelihood, we would expect the model to make mistakes.

      Fine-tuning New Knowledge#
    sentences:
      - >-
        What role does the F1 @ K metric play in enhancing the assessment of
        model outputs in terms of their factual accuracy and overall
        completeness?
      - >-
        In what ways do MaybeKnown examples improve the performance of a model
        when contrasted with HighlyKnown examples, and what implications does
        this have for developing effective training strategies?
      - >-
        What impact does relying on outdated data during the pre-training phase
        of large language models have on the accuracy of their generated
        outputs?
model-index:
  - name: BGE base Financial Matryoshka
    results:
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 768
          type: dim_768
        metrics:
          - type: cosine_accuracy@1
            value: 0.953125
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 1
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 1
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 1
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.953125
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.3333333333333333
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.19999999999999998
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09999999999999999
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.953125
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 1
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 1
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 1
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.9826998321986622
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.9765625
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.9765625
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 512
          type: dim_512
        metrics:
          - type: cosine_accuracy@1
            value: 0.9479166666666666
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 1
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 1
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 1
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.9479166666666666
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.3333333333333333
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.19999999999999998
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09999999999999999
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.9479166666666666
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 1
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 1
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 1
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.9800956655319956
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.9730902777777778
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.9730902777777777
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 256
          type: dim_256
        metrics:
          - type: cosine_accuracy@1
            value: 0.9635416666666666
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 1
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 1
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 1
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.9635416666666666
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.3333333333333333
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.19999999999999998
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09999999999999999
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.9635416666666666
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 1
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 1
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 1
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.9865443139322926
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.9817708333333334
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.9817708333333334
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 128
          type: dim_128
        metrics:
          - type: cosine_accuracy@1
            value: 0.9583333333333334
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 1
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 1
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 1
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.9583333333333334
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.3333333333333333
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.19999999999999998
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09999999999999999
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.9583333333333334
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 1
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 1
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 1
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.9832582214657748
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.9774305555555555
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.9774305555555557
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 64
          type: dim_64
        metrics:
          - type: cosine_accuracy@1
            value: 0.9583333333333334
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 1
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 1
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 1
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.9583333333333334
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.3333333333333333
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.19999999999999998
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.09999999999999999
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.9583333333333334
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 1
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 1
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 1
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.9832582214657748
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.9774305555555555
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.9774305555555557
            name: Cosine Map@100

BGE base Financial Matryoshka

This is a sentence-transformers model finetuned from BAAI/bge-base-en-v1.5. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: BAAI/bge-base-en-v1.5
Maximum Sequence Length: 512 tokens
Output Dimensionality: 768 tokens
Similarity Function: Cosine Similarity
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': 512, 'do_lower_case': True}) with Transformer model: BertModel 
  (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})
  (2): Normalize()
)

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("joshuapb/fine-tuned-matryoshka-1500")
# Run inference
sentences = [
    'This post focuses on extrinsic hallucination. To avoid hallucination, LLMs need to be (1) factual and (2) acknowledge not knowing the answer when applicable.\nWhat Causes Hallucinations?#\nGiven a standard deployable LLM goes through pre-training and fine-tuning for alignment and other improvements, let us consider causes at both stages.\nPre-training Data Issues#\nThe volume of the pre-training data corpus is enormous, as it is supposed to represent world knowledge in all available written forms. Data crawled from the public Internet is the most common choice and thus out-of-date, missing, or incorrect information is expected. As the model may incorrectly memorize this information by simply maximizing the log-likelihood, we would expect the model to make mistakes.\nFine-tuning New Knowledge#',
    'What impact does relying on outdated data during the pre-training phase of large language models have on the accuracy of their generated outputs?',
    'In what ways do MaybeKnown examples improve the performance of a model when contrasted with HighlyKnown examples, and what implications does this have for developing effective training strategies?',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

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

Evaluation

Metrics

Information Retrieval

Dataset: dim_768
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.9531
cosine_accuracy@3	1.0
cosine_accuracy@5	1.0
cosine_accuracy@10	1.0
cosine_precision@1	0.9531
cosine_precision@3	0.3333
cosine_precision@5	0.2
cosine_precision@10	0.1
cosine_recall@1	0.9531
cosine_recall@3	1.0
cosine_recall@5	1.0
cosine_recall@10	1.0
cosine_ndcg@10	0.9827
cosine_mrr@10	0.9766
cosine_map@100	0.9766

Information Retrieval

Dataset: dim_512
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.9479
cosine_accuracy@3	1.0
cosine_accuracy@5	1.0
cosine_accuracy@10	1.0
cosine_precision@1	0.9479
cosine_precision@3	0.3333
cosine_precision@5	0.2
cosine_precision@10	0.1
cosine_recall@1	0.9479
cosine_recall@3	1.0
cosine_recall@5	1.0
cosine_recall@10	1.0
cosine_ndcg@10	0.9801
cosine_mrr@10	0.9731
cosine_map@100	0.9731

Information Retrieval

Dataset: dim_256
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.9635
cosine_accuracy@3	1.0
cosine_accuracy@5	1.0
cosine_accuracy@10	1.0
cosine_precision@1	0.9635
cosine_precision@3	0.3333
cosine_precision@5	0.2
cosine_precision@10	0.1
cosine_recall@1	0.9635
cosine_recall@3	1.0
cosine_recall@5	1.0
cosine_recall@10	1.0
cosine_ndcg@10	0.9865
cosine_mrr@10	0.9818
cosine_map@100	0.9818

Information Retrieval

Dataset: dim_128
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.9583
cosine_accuracy@3	1.0
cosine_accuracy@5	1.0
cosine_accuracy@10	1.0
cosine_precision@1	0.9583
cosine_precision@3	0.3333
cosine_precision@5	0.2
cosine_precision@10	0.1
cosine_recall@1	0.9583
cosine_recall@3	1.0
cosine_recall@5	1.0
cosine_recall@10	1.0
cosine_ndcg@10	0.9833
cosine_mrr@10	0.9774
cosine_map@100	0.9774

Information Retrieval

Dataset: dim_64
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.9583
cosine_accuracy@3	1.0
cosine_accuracy@5	1.0
cosine_accuracy@10	1.0
cosine_precision@1	0.9583
cosine_precision@3	0.3333
cosine_precision@5	0.2
cosine_precision@10	0.1
cosine_recall@1	0.9583
cosine_recall@3	1.0
cosine_recall@5	1.0
cosine_recall@10	1.0
cosine_ndcg@10	0.9833
cosine_mrr@10	0.9774
cosine_map@100	0.9774

Training Details

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: epoch
per_device_eval_batch_size: 16
learning_rate: 2e-05
num_train_epochs: 5
lr_scheduler_type: cosine
warmup_ratio: 0.1
load_best_model_at_end: True

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: epoch
prediction_loss_only: True
per_device_train_batch_size: 8
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
learning_rate: 2e-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: cosine
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: 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: 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
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
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss	dim_128_cosine_map@100	dim_256_cosine_map@100	dim_512_cosine_map@100	dim_64_cosine_map@100	dim_768_cosine_map@100
0.0266	5	4.6076	-	-	-	-	-
0.0532	10	5.2874	-	-	-	-	-
0.0798	15	5.4181	-	-	-	-	-
0.1064	20	5.1322	-	-	-	-	-
0.1330	25	4.1674	-	-	-	-	-
0.1596	30	4.1998	-	-	-	-	-
0.1862	35	3.4182	-	-	-	-	-
0.2128	40	4.1142	-	-	-	-	-
0.2394	45	2.5775	-	-	-	-	-
0.2660	50	3.3767	-	-	-	-	-
0.2926	55	2.5797	-	-	-	-	-
0.3191	60	3.1813	-	-	-	-	-
0.3457	65	3.7209	-	-	-	-	-
0.3723	70	2.2637	-	-	-	-	-
0.3989	75	2.2651	-	-	-	-	-
0.4255	80	2.3023	-	-	-	-	-
0.4521	85	2.3261	-	-	-	-	-
0.4787	90	1.947	-	-	-	-	-
0.5053	95	0.8502	-	-	-	-	-
0.5319	100	2.2405	-	-	-	-	-
0.5585	105	2.0157	-	-	-	-	-
0.5851	110	1.4405	-	-	-	-	-
0.6117	115	1.9714	-	-	-	-	-
0.6383	120	2.5212	-	-	-	-	-
0.6649	125	2.734	-	-	-	-	-
0.6915	130	1.9357	-	-	-	-	-
0.7181	135	1.1727	-	-	-	-	-
0.7447	140	1.9789	-	-	-	-	-
0.7713	145	1.6362	-	-	-	-	-
0.7979	150	1.7356	-	-	-	-	-
0.8245	155	1.916	-	-	-	-	-
0.8511	160	2.0372	-	-	-	-	-
0.8777	165	1.5705	-	-	-	-	-
0.9043	170	1.9393	-	-	-	-	-
0.9309	175	1.6289	-	-	-	-	-
0.9574	180	2.8158	-	-	-	-	-
0.9840	185	1.1869	-	-	-	-	-
1.0	188	-	0.9319	0.9438	0.9401	0.9173	0.9421
1.0106	190	1.1572	-	-	-	-	-
1.0372	195	1.4815	-	-	-	-	-
1.0638	200	1.6742	-	-	-	-	-
1.0904	205	0.9434	-	-	-	-	-
1.1170	210	1.6141	-	-	-	-	-
1.1436	215	0.7478	-	-	-	-	-
1.1702	220	1.4812	-	-	-	-	-
1.1968	225	1.8121	-	-	-	-	-
1.2234	230	1.2595	-	-	-	-	-
1.25	235	1.8326	-	-	-	-	-
1.2766	240	1.3828	-	-	-	-	-
1.3032	245	1.5385	-	-	-	-	-
1.3298	250	1.1213	-	-	-	-	-
1.3564	255	1.0444	-	-	-	-	-
1.3830	260	0.3848	-	-	-	-	-
1.4096	265	0.8369	-	-	-	-	-
1.4362	270	1.682	-	-	-	-	-
1.4628	275	1.9625	-	-	-	-	-
1.4894	280	2.0732	-	-	-	-	-
1.5160	285	1.8939	-	-	-	-	-
1.5426	290	1.5621	-	-	-	-	-
1.5691	295	1.5474	-	-	-	-	-
1.5957	300	2.1111	-	-	-	-	-
1.6223	305	1.8619	-	-	-	-	-
1.6489	310	1.1091	-	-	-	-	-
1.6755	315	1.8127	-	-	-	-	-
1.7021	320	0.8599	-	-	-	-	-
1.7287	325	0.9553	-	-	-	-	-
1.7553	330	1.2444	-	-	-	-	-
1.7819	335	1.6786	-	-	-	-	-
1.8085	340	1.2092	-	-	-	-	-
1.8351	345	0.8824	-	-	-	-	-
1.8617	350	0.4448	-	-	-	-	-
1.8883	355	1.116	-	-	-	-	-
1.9149	360	1.587	-	-	-	-	-
1.9415	365	0.7235	-	-	-	-	-
1.9681	370	0.9446	-	-	-	-	-
1.9947	375	1.0066	-	-	-	-	-
2.0	376	-	0.9570	0.9523	0.9501	0.9501	0.9549
2.0213	380	1.3895	-	-	-	-	-
2.0479	385	1.0259	-	-	-	-	-
2.0745	390	0.9961	-	-	-	-	-
2.1011	395	1.4164	-	-	-	-	-
2.1277	400	0.5188	-	-	-	-	-
2.1543	405	0.2965	-	-	-	-	-
2.1809	410	0.4351	-	-	-	-	-
2.2074	415	0.7546	-	-	-	-	-
2.2340	420	1.9408	-	-	-	-	-
2.2606	425	1.0056	-	-	-	-	-
2.2872	430	1.3175	-	-	-	-	-
2.3138	435	0.9397	-	-	-	-	-
2.3404	440	1.4308	-	-	-	-	-
2.3670	445	0.8647	-	-	-	-	-
2.3936	450	0.8917	-	-	-	-	-
2.4202	455	0.7922	-	-	-	-	-
2.4468	460	1.1815	-	-	-	-	-
2.4734	465	0.8071	-	-	-	-	-
2.5	470	0.1601	-	-	-	-	-
2.5266	475	0.7533	-	-	-	-	-
2.5532	480	1.351	-	-	-	-	-
2.5798	485	1.2948	-	-	-	-	-
2.6064	490	1.4087	-	-	-	-	-
2.6330	495	2.2427	-	-	-	-	-
2.6596	500	0.4735	-	-	-	-	-
2.6862	505	0.8377	-	-	-	-	-
2.7128	510	0.525	-	-	-	-	-
2.7394	515	0.8455	-	-	-	-	-
2.7660	520	2.458	-	-	-	-	-
2.7926	525	1.2906	-	-	-	-	-
2.8191	530	1.0234	-	-	-	-	-
2.8457	535	0.3733	-	-	-	-	-
2.8723	540	0.388	-	-	-	-	-
2.8989	545	1.2155	-	-	-	-	-
2.9255	550	1.0288	-	-	-	-	-
2.9521	555	1.0578	-	-	-	-	-
2.9787	560	0.1793	-	-	-	-	-
3.0	564	-	0.9653	0.9714	0.9705	0.9609	0.9679
3.0053	565	1.0141	-	-	-	-	-
3.0319	570	0.6978	-	-	-	-	-
3.0585	575	0.6066	-	-	-	-	-
3.0851	580	0.2444	-	-	-	-	-
3.1117	585	0.581	-	-	-	-	-
3.1383	590	1.3544	-	-	-	-	-
3.1649	595	0.9379	-	-	-	-	-
3.1915	600	1.0088	-	-	-	-	-
3.2181	605	1.6689	-	-	-	-	-
3.2447	610	0.3204	-	-	-	-	-
3.2713	615	0.5433	-	-	-	-	-
3.2979	620	0.7225	-	-	-	-	-
3.3245	625	1.7695	-	-	-	-	-
3.3511	630	0.7472	-	-	-	-	-
3.3777	635	1.0883	-	-	-	-	-
3.4043	640	1.1863	-	-	-	-	-
3.4309	645	1.7163	-	-	-	-	-
3.4574	650	2.8196	-	-	-	-	-
3.4840	655	1.5015	-	-	-	-	-
3.5106	660	1.3862	-	-	-	-	-
3.5372	665	0.775	-	-	-	-	-
3.5638	670	1.2385	-	-	-	-	-
3.5904	675	0.9472	-	-	-	-	-
3.6170	680	0.6458	-	-	-	-	-
3.6436	685	0.8308	-	-	-	-	-
3.6702	690	1.0864	-	-	-	-	-
3.6968	695	1.0715	-	-	-	-	-
3.7234	700	1.5082	-	-	-	-	-
3.75	705	0.5028	-	-	-	-	-
3.7766	710	1.1525	-	-	-	-	-
3.8032	715	0.5829	-	-	-	-	-
3.8298	720	0.6168	-	-	-	-	-
3.8564	725	1.0185	-	-	-	-	-
3.8830	730	1.2545	-	-	-	-	-
3.9096	735	0.5604	-	-	-	-	-
3.9362	740	0.6879	-	-	-	-	-
3.9628	745	0.9936	-	-	-	-	-
3.9894	750	0.5786	-	-	-	-	-
4.0	752	-	0.9774	0.9818	0.9731	0.98	0.9792
4.0160	755	0.908	-	-	-	-	-
4.0426	760	0.988	-	-	-	-	-
4.0691	765	0.2616	-	-	-	-	-
4.0957	770	1.1475	-	-	-	-	-
4.1223	775	1.7832	-	-	-	-	-
4.1489	780	0.7522	-	-	-	-	-
4.1755	785	1.4473	-	-	-	-	-
4.2021	790	0.7194	-	-	-	-	-
4.2287	795	0.0855	-	-	-	-	-
4.2553	800	1.151	-	-	-	-	-
4.2819	805	1.5109	-	-	-	-	-
4.3085	810	0.7462	-	-	-	-	-
4.3351	815	0.4697	-	-	-	-	-
4.3617	820	1.1215	-	-	-	-	-
4.3883	825	1.3527	-	-	-	-	-
4.4149	830	0.8995	-	-	-	-	-
4.4415	835	1.0011	-	-	-	-	-
4.4681	840	1.1168	-	-	-	-	-
4.4947	845	1.3105	-	-	-	-	-
4.5213	850	0.2855	-	-	-	-	-
4.5479	855	1.3223	-	-	-	-	-
4.5745	860	0.6377	-	-	-	-	-
4.6011	865	1.2196	-	-	-	-	-
4.6277	870	1.257	-	-	-	-	-
4.6543	875	0.93	-	-	-	-	-
4.6809	880	0.8831	-	-	-	-	-
4.7074	885	0.23	-	-	-	-	-
4.7340	890	0.9771	-	-	-	-	-
4.7606	895	1.026	-	-	-	-	-
4.7872	900	1.4671	-	-	-	-	-
4.8138	905	0.8719	-	-	-	-	-
4.8404	910	0.9108	-	-	-	-	-
4.8670	915	1.359	-	-	-	-	-
4.8936	920	1.3237	-	-	-	-	-
4.9202	925	0.6591	-	-	-	-	-
4.9468	930	0.405	-	-	-	-	-
4.9734	935	1.1984	-	-	-	-	-
5.0	940	0.5747	0.9774	0.9818	0.9731	0.9774	0.9766

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.10.12
Sentence Transformers: 3.0.1
Transformers: 4.42.4
PyTorch: 2.3.1+cu121
Accelerate: 0.32.1
Datasets: 2.21.0
Tokenizers: 0.19.1

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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning}, 
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply}, 
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}