README.md

metadata

language: []
library_name: sentence-transformers
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:7005
  - loss:MultipleNegativesRankingLoss_with_logging
base_model: Alibaba-NLP/gte-large-en-v1.5
datasets: []
metrics:
  - cosine_accuracy@1
  - cosine_accuracy@3
  - cosine_accuracy@5
  - cosine_accuracy@10
  - cosine_accuracy@30
  - cosine_accuracy@50
  - cosine_accuracy@100
  - cosine_precision@1
  - cosine_precision@3
  - cosine_precision@5
  - cosine_precision@10
  - cosine_precision@30
  - cosine_precision@50
  - cosine_precision@100
  - cosine_recall@1
  - cosine_recall@3
  - cosine_recall@5
  - cosine_recall@10
  - cosine_recall@30
  - cosine_recall@50
  - cosine_recall@100
  - cosine_ndcg@10
  - cosine_mrr@10
  - cosine_map@100
  - dot_accuracy@1
  - dot_accuracy@3
  - dot_accuracy@5
  - dot_accuracy@10
  - dot_accuracy@30
  - dot_accuracy@50
  - dot_accuracy@100
  - dot_precision@1
  - dot_precision@3
  - dot_precision@5
  - dot_precision@10
  - dot_precision@30
  - dot_precision@50
  - dot_precision@100
  - dot_recall@1
  - dot_recall@3
  - dot_recall@5
  - dot_recall@10
  - dot_recall@30
  - dot_recall@50
  - dot_recall@100
  - dot_ndcg@10
  - dot_mrr@10
  - dot_map@100
widget:
  - source_sentence: What are the client's target industries?
    sentences:
      - |-
        Right.
        And also, you know, heavy equipment.
        Okay, I understand.
      - >-
        And there's a full spectrum.

        It's all about your order offering.

        Right.

        If you're offering, like, a full design platform where now we have way
        more engagement in terms of employee being able to get it from one
        place, and that could be.

        That could take away again, like, my pitch would be basically being on
        the show.
      - >-
        Our competitors are billion dollar corporations.

        So Experian Epsilon, which is owned by IPG or publicis, big french
        company, Axiom, which is owned by IPG.

        Inter public group, huge agency.

        So it's nice competing against multibillion dollar corporations because
        they move at the speed of the Statue of Liberty.
  - source_sentence: What is the strategy for heating products?
    sentences:
      - |-
        Then when you go in to take a look, you say, okay, I've got this.
        Now I need to record my test results so that we do down here.
        And we say, okay, this is me, so I'll pick myself.
        And here we go.
        So once you're in here, you say, okay, it's inspector me.
      - >-
        I don't think we make any margin on these products.

        I'm going to put it on here, though, because I want to add different
        ones.

        So three in one and then.

        Underfloor heating?
      - |-
        How are others using it?
        Use cases like.
        Yeah, for example, we have one, one partner, there's climbo.
  - source_sentence: What feature did Aseel request regarding budget information display?
    sentences:
      - |-
        So you want to do your west coast.
        Do you want to do 10:00 a.m.
        on the morning of 13th?
      - >-
        But the only thing that I just was thinking about is, for example, if I
        was a head teacher and I'm about to approve an order and obviously I go
        and click on the three dots and it tells me my geo budget department by
        GL budget and obviously tells you what your total budget is, your spend
        and what's remaining.

        Is there a way in which I can see what actually went under proof
        expenditure?

        So it should be.

        So to see how much has been committed against the budget?
      - |-
        Awesome.
        And speaking of releases, is there any way I'm not getting the.
        And I'm sure Chris probably is.
  - source_sentence: Does the customer have any other EAP-like resources available?
    sentences:
      - >-
        Every time I make a post, I get.

        I get just a ton of inquiries, you know?

        And we're just.

        We're doing a bunch of cool operational stuff right now, so we're just
        trying to get that all figured out, you know?

        Yeah.

        Well, hey, let me give you a rundown of a couple things I'm doing with,
        like, people in your kind of peripheral.

        Just so you know what we're trying to do to boost the voices of you and
        agencies like you.
      - |-
        So we need Kim and Manju.
        We need to account that for production downtime for on 16th.
        No cutover plan.
      - >-
        They're thinking, well, there we have them already, and they offer all
        these things.

        This is pretty great, you know, because we also use, so we have Voya
        life insurance, and through Voya, they offer a couple eap type of
        resources, too.

        Right.

        So we have additional assistance with another program.

        Right.

        But with our eap, which is through Magellan, they would just usually
        would just be better than the other comparisons when it came down to it.
  - source_sentence: What was Nathan's response to the initial proposal from Global Air U?
    sentences:
      - But I was listening to everything that you were talking about.
      - |-
        And hopefully that should update now in your account in a second.
        Yeah.
        If you give that a go now, you should see all the way to August 2025.
      - |-
        I don't see on the proposal.
        I don't see anything class or the class related.
        Um.
        Oh, so for the course.
        No, no.
pipeline_tag: sentence-similarity
model-index:
  - name: SentenceTransformer based on Alibaba-NLP/gte-large-en-v1.5
    results:
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: Unknown
          type: unknown
        metrics:
          - type: cosine_accuracy@1
            value: 0.32793959007551243
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.48975188781014023
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.5663430420711975
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.6612729234088457
            name: Cosine Accuracy@10
          - type: cosine_accuracy@30
            value: 0.7669902912621359
            name: Cosine Accuracy@30
          - type: cosine_accuracy@50
            value: 0.8155339805825242
            name: Cosine Accuracy@50
          - type: cosine_accuracy@100
            value: 0.8597626752966558
            name: Cosine Accuracy@100
          - type: cosine_precision@1
            value: 0.32793959007551243
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.1902193455591514
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.13829557713052856
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.08716289104638619
            name: Cosine Precision@10
          - type: cosine_precision@30
            value: 0.038439410284070476
            name: Cosine Precision@30
          - type: cosine_precision@50
            value: 0.025717367853290186
            name: Cosine Precision@50
          - type: cosine_precision@100
            value: 0.014282632146709814
            name: Cosine Precision@100
          - type: cosine_recall@1
            value: 0.19877399359600004
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.32606462218112703
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.39100529100529097
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.475571479940412
            name: Cosine Recall@10
          - type: cosine_recall@30
            value: 0.6031369325867708
            name: Cosine Recall@30
          - type: cosine_recall@50
            value: 0.660217290799815
            name: Cosine Recall@50
          - type: cosine_recall@100
            value: 0.7195099398982894
            name: Cosine Recall@100
          - type: cosine_ndcg@10
            value: 0.3784769275629581
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.42950420369514186
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.3193224907975288
            name: Cosine Map@100
          - type: dot_accuracy@1
            value: 0.3290183387270766
            name: Dot Accuracy@1
          - type: dot_accuracy@3
            value: 0.4886731391585761
            name: Dot Accuracy@3
          - type: dot_accuracy@5
            value: 0.5717367853290184
            name: Dot Accuracy@5
          - type: dot_accuracy@10
            value: 0.6634304207119741
            name: Dot Accuracy@10
          - type: dot_accuracy@30
            value: 0.7669902912621359
            name: Dot Accuracy@30
          - type: dot_accuracy@50
            value: 0.8133764832793959
            name: Dot Accuracy@50
          - type: dot_accuracy@100
            value: 0.8619201725997843
            name: Dot Accuracy@100
          - type: dot_precision@1
            value: 0.3290183387270766
            name: Dot Precision@1
          - type: dot_precision@3
            value: 0.18985976267529667
            name: Dot Precision@3
          - type: dot_precision@5
            value: 0.1387270765911543
            name: Dot Precision@5
          - type: dot_precision@10
            value: 0.08737864077669903
            name: Dot Precision@10
          - type: dot_precision@30
            value: 0.038511326860841424
            name: Dot Precision@30
          - type: dot_precision@50
            value: 0.025652642934196335
            name: Dot Precision@50
          - type: dot_precision@100
            value: 0.0143042071197411
            name: Dot Precision@100
          - type: dot_recall@1
            value: 0.19940326364274585
            name: Dot Recall@1
          - type: dot_recall@3
            value: 0.32588483073919966
            name: Dot Recall@3
          - type: dot_recall@5
            value: 0.39370216263420144
            name: Dot Recall@5
          - type: dot_recall@10
            value: 0.4770997071967946
            name: Dot Recall@10
          - type: dot_recall@30
            value: 0.6043595143918767
            name: Dot Recall@30
          - type: dot_recall@50
            value: 0.659138542148251
            name: Dot Recall@50
          - type: dot_recall@100
            value: 0.7219987671443983
            name: Dot Recall@100
          - type: dot_ndcg@10
            value: 0.3791495475200093
            name: Dot Ndcg@10
          - type: dot_mrr@10
            value: 0.4305302991387128
            name: Dot Mrr@10
          - type: dot_map@100
            value: 0.31951258454174397
            name: Dot Map@100

SentenceTransformer based on Alibaba-NLP/gte-large-en-v1.5

This is a sentence-transformers model finetuned from Alibaba-NLP/gte-large-en-v1.5. It maps sentences & paragraphs to a 1024-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: Alibaba-NLP/gte-large-en-v1.5
• Maximum Sequence Length: 8192 tokens
• Output Dimensionality: 1024 tokens
• Similarity Function: Cosine Similarity

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': 8192, 'do_lower_case': False}) with Transformer model: NewModel 
  (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("model_3")
# Run inference
sentences = [
    "What was Nathan's response to the initial proposal from Global Air U?",
    "I don't see on the proposal.\nI don't see anything class or the class related.\nUm.\nOh, so for the course.\nNo, no.",
    'And hopefully that should update now in your account in a second.\nYeah.\nIf you give that a go now, you should see all the way to August 2025.',
]
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

Information Retrieval

• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.3279
cosine_accuracy@3	0.4898
cosine_accuracy@5	0.5663
cosine_accuracy@10	0.6613
cosine_accuracy@30	0.767
cosine_accuracy@50	0.8155
cosine_accuracy@100	0.8598
cosine_precision@1	0.3279
cosine_precision@3	0.1902
cosine_precision@5	0.1383
cosine_precision@10	0.0872
cosine_precision@30	0.0384
cosine_precision@50	0.0257
cosine_precision@100	0.0143
cosine_recall@1	0.1988
cosine_recall@3	0.3261
cosine_recall@5	0.391
cosine_recall@10	0.4756
cosine_recall@30	0.6031
cosine_recall@50	0.6602
cosine_recall@100	0.7195
cosine_ndcg@10	0.3785
cosine_mrr@10	0.4295
cosine_map@100	0.3193
dot_accuracy@1	0.329
dot_accuracy@3	0.4887
dot_accuracy@5	0.5717
dot_accuracy@10	0.6634
dot_accuracy@30	0.767
dot_accuracy@50	0.8134
dot_accuracy@100	0.8619
dot_precision@1	0.329
dot_precision@3	0.1899
dot_precision@5	0.1387
dot_precision@10	0.0874
dot_precision@30	0.0385
dot_precision@50	0.0257
dot_precision@100	0.0143
dot_recall@1	0.1994
dot_recall@3	0.3259
dot_recall@5	0.3937
dot_recall@10	0.4771
dot_recall@30	0.6044
dot_recall@50	0.6591
dot_recall@100	0.722
dot_ndcg@10	0.3791
dot_mrr@10	0.4305
dot_map@100	0.3195

Training Details

Training Dataset

Unnamed Dataset

• Size: 7,005 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 8 tokens mean: 14.59 tokens max: 25 tokens	min: 12 tokens mean: 60.98 tokens max: 170 tokens

• Samples:

  anchor	positive
  What progress has been made with setting up Snowflake share?	He finally got around to giving me the information necessary to set up Snowflake share. I will be submitting the application to get back set up. Once the database is set up, then we just need to figure out how to configure Snowflake share, which it's going to be in the documentation. We should be set on that end. We also are going to have a conversation with someone named Peter Tsanghen, who's, who owns Jira platform. Great.
  Who is Peter Tsanghen and what is the planned interaction with him?	He finally got around to giving me the information necessary to set up Snowflake share. I will be submitting the application to get back set up. Once the database is set up, then we just need to figure out how to configure Snowflake share, which it's going to be in the documentation. We should be set on that end. We also are going to have a conversation with someone named Peter Tsanghen, who's, who owns Jira platform. Great.
  Who is Peter Tsanghen and what is the planned interaction with him?	Uh, and so now we just have to meet with Peter. Peter is someone who I used to work with on, he used to work on, uh, syndicated data products. So I used to work with him on that.

• Loss: main.MultipleNegativesRankingLoss_with_logging

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 4
• per_device_eval_batch_size: 4
• num_train_epochs: 2
• max_steps: 1751
• disable_tqdm: True
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• prediction_loss_only: True
• per_device_train_batch_size: 4
• per_device_eval_batch_size: 4
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_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
• num_train_epochs: 2
• max_steps: 1751
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.0
• 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
• 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: True
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• 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}
• 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
• 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_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin

Training Logs

Epoch	Step	cosine_map@100
0.0114	20	0.2538
0.0228	40	0.2601
0.0342	60	0.2724
0.0457	80	0.2911
0.0571	100	0.2976
0.0685	120	0.3075
0.0799	140	0.3071
0.0913	160	0.3111
0.1027	180	0.3193

Framework Versions

• Python: 3.10.9
• Sentence Transformers: 3.0.1
• Transformers: 4.39.3
• PyTorch: 2.3.1+cu121
• Accelerate: 0.31.0
• Datasets: 2.20.0
• Tokenizers: 0.15.2

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",
}