SentenceTransformer based on sentence-transformers/all-MiniLM-L6-v2

This is a sentence-transformers model finetuned from sentence-transformers/all-MiniLM-L6-v2. It maps sentences & paragraphs to a 384-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: sentence-transformers/all-MiniLM-L6-v2
• Maximum Sequence Length: 256 tokens
• Output Dimensionality: 384 dimensions
• Similarity Function: Cosine Similarity
• Supported Modality: Text

Model Sources

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

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'BertModel'})
  (1): Pooling({'embedding_dimension': 384, 'pooling_mode': 'mean', '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("nishp24/legalmind-embedder-v2")
# Run inference
sentences = [
    'What is the general principle that allows a court to prevent a party from going back on an underlying assumption in a transaction?',
    'action, estoppel cannot do away with the need to consideration, and so forth. All these can now be seen to merge into one general principle shorn of limitations. When the parties to a transaction proceed on the basis of an underlying assumption - either of fact or of law - whether due to misrepresentation or mistake makes no difference - on which they have conducted the dealings between them - neither of them will be allowed to go back on that assumption when it would be unfair or unjust to allow him to do so. If one of them does seek to go back on it, the courts will give the other side remedy as the equity of the cases demands. That general principle applies to this case. Both the plaintiff and the bank proceeded for years on the underlying assumption that the guarantee of the plaintiffs applied to the $3,250,000 advanced by the bank for the Nassau Building. Their dealings in rearranging the portfolio, in releasing properties and monies, will all conducted on that basis. On that basis the bank apply the surplus of $750,000 (on the English properties) in discharge of the obligations of the plaintiffs under the guarantee. It would be most unfair and unjust to allow the liquidator to depart from that basis and claim back now the $750,000. That was ultimately the paramount reason why the judge rejected the liquidator\'s claim. He summed up his view in this one sentence, ante, p. 108C: "…I am satisfied that Mr. Foster\'s conduct, though of course completely innocent, so influenced Mr. Oldfield\'s conduct, as to render it unconscionable on the part of the plaintiffs now to take advantage of the bank\'s error." Later at p. 108G the judge speaks of it being "unconscionable" for the representor to go back on his representation. In those phrases, the judge is applying the general principle of estoppel which I have stated. I agree with his analysis of the cases and with his conclusion…." Eveleigh LJ dealt with the case "as one of estoppel by convention" (Page 126 A) quoting with apparent approval the passage from Spencer Bower and Turner, Estoppel by Representation: "When the parties have acted in their transaction upon the agreed assumption that a given state of facts is to be accepted between them as true, then as regards that transaction each will be estopped against the other from',
    'where the other party liable was Armour. v) The judge derived what he thought the parties thought the NCC meant from:- a) Mr West\'s letter to IFA 6 months after the Agreement on 10th August 2006 referring to the agreement that key aspects would be undertaken by separate contractors, and saying that the NCC meant that one such contractor, GlasSpace, provided their own warranty: "and we are simply asking you as project manager and "independent arbiter" to confirm when the work has been completed & we should release the 5% retainer to GlasSpace". b) The fact that Mr Finlay had written "Yes Agreed" against the first statement in Mr West\'s letter, and "Fine Understand" against the second. vi) For unfairness to be established under regulation 5(1) of the UTCC Regulations, the NCC would have to be cumulatively (a) contrary to the requirements of good faith, and (b) a clause which causes a significant imbalance in the parties\' rights and obligations arising under the Agreement to the detriment of the Wests. He referred to Lord Bingham\'s speech at page 491 in Director General of Fair Trading v. First National Bank plc [2002] 1 AC 481 (the "First National" case). vii) The judge concluded that Mr Finlay was not guilty of any lack of good faith in the sense described by Lord Bingham. Though the NCC was not recommended by RIBA, it was at least consistent with a form of wording that was. The judge\'s reasoning on interest On this aspect, the judge accepted Mr West\'s evidence that the cost of his borrowing to fund the remedial works was at a rate of approximately 8% per annum, and awarded interest at 7% over base, rejecting IFA\'s submission that 3% over base rate was an appropriate rate on the ground that it "would not begin to compensate the Wests for the cost of their borrowing and being kept out of their money". The judge\'s reasoning on inconvenience, distress and discomfort The judge relied first on the judgment of Akenhead J in AXA Insurance UK plc v. Cunningham Lindsay UK [2007] EWCA 3023 (TCC) (the "AXA Insurance" case) who had said that in 2007 damages for inconvenience, distress and discomfort caused by breaches of contract would generally not exceed £2,500 per person per annum. He reminded himself that he should be astute not to award damages for the stress and vexation of litigation. The',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.5624, 0.1161],
#         [0.5624, 1.0000, 0.0879],
#         [0.1161, 0.0879, 1.0000]])

Training Details

Training Dataset

Unnamed Dataset

• Size: 5,993 training samples
• Columns: sentence_0 and sentence_1
• Approximate statistics based on the first 100 samples:

  	sentence_0	sentence_1
  type	string	string
  modality	text	text
  details	min: 12 tokensmean: 29.88 tokensmax: 85 tokens	min: 81 tokensmean: 250.88 tokensmax: 256 tokens

• Samples:

  sentence_0	sentence_1
  What authorities must be parties to an agreement if part of what is to be maintained or transferred is situated in their area?	be a party to such an agreement if part of what is to be maintained or transferred is situated in their area and the remainder in the area or areas of one or more other authorities (whether or not local authorities) who are also parties to the agreement. . 43 In section 111 (access agreements and orders as respects canals other than commercial waterways and cruising waterways), for “the Waterways Board” substitute “Canal & River Trust”. 44 1 Section 112 (power to extinguish statutory rights and obligations in respect of canals not comprised in undertaking of Board) is amended as follows. 2 In the heading, after “Board” insert “or Canal & River Trust”. 3 In subsection (1) — a after “Waterways Board” insert “or Canal & River Trust”; b after “the Minister” insert “, or, in the case of a canal in Scotland, the Scottish Ministers,”. 4 In subsection (3) — a in paragraph (a), for “the Minister” substitute “the authority making the order”; b in paragraph (d), for “the Minister” substitute “the...
  What are the factors that a review panel must consider before adjourning a review under section 13(1) of this legislation?	only make representations with the consent of the arranging authority. 13 1 A review panel may from time to time adjourn a review but, before doing so, must consider the effect of any adjournment on — a each of the parties to the review; b any victim of the incident leading to the exclusion; and c where such person is not the relevant person — i the pupil; or ii any parent of the pupil. 2 If the relevant person has requested the appointment of a SEN expert but the SEN expert is not in attendance, the relevant person may ask the review panel to adjourn the review to a later date or time so that the SEN expert, or an alternative SEN expert, may attend. 3 Where a relevant person asks the review panel to adjourn the review in pursuance of sub-paragraph (2) the review panel must adjourn the review. 4 The review panel must ensure that the relevant person is aware of their right to request an adjournment in pursuance of sub-paragraph (2). 5 A review may be adjourned if the effect of the adjou...
  Is it fair, just and reasonable to impose a duty of care in this case?	fair, just and reasonable to impose this duty of care. When the Court reaches this stage of the test, foreseeability and proximity have been established. The basis for the denial of a remedy in these circumstances on this third ground must be very compelling. I can find no proper ground in this case on which to conclude that it would not be fair, just and reasonable to impose a duty to avoid foreseeable harm to a sufficiently proximate victim. ANALOGY AND INCREMENTALISM At this point the decision in Sharp becomes highly relevant. St John Poulton's Trustee is also material because there the court distinguished Sharp on the basis that unlike Sharp the case before it involved a pure omission. The case before me does not involve an omission: it involves the entry of false information on the Register, which is a positive act of the kind done in Sharp. I do not mean to suggest that there is no distinction between the present case and Sharp's case. If that were so, it would be enough simply t...

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false,
      "directions": [
          "query_to_doc"
      ],
      "partition_mode": "joint",
      "hardness_mode": null,
      "hardness_strength": 0.0
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 16
• num_train_epochs: 1
• per_device_eval_batch_size: 16
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

• per_device_train_batch_size: 16
• num_train_epochs: 1
• max_steps: -1
• learning_rate: 5e-05
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_steps: 0
• optim: adamw_torch_fused
• optim_args: None
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• optim_target_modules: None
• gradient_accumulation_steps: 1
• average_tokens_across_devices: True
• max_grad_norm: 1
• label_smoothing_factor: 0.0
• bf16: False
• fp16: False
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• use_liger_kernel: False
• liger_kernel_config: None
• use_cache: False
• neftune_noise_alpha: None
• torch_empty_cache_steps: None
• auto_find_batch_size: False
• log_on_each_node: True
• logging_nan_inf_filter: True
• include_num_input_tokens_seen: no
• log_level: passive
• log_level_replica: warning
• disable_tqdm: False
• project: huggingface
• trackio_space_id: None
• trackio_bucket_id: None
• trackio_static_space_id: None
• per_device_eval_batch_size: 16
• prediction_loss_only: True
• eval_on_start: False
• eval_do_concat_batches: True
• eval_use_gather_object: False
• eval_accumulation_steps: None
• include_for_metrics: []
• batch_eval_metrics: False
• save_only_model: False
• save_on_each_node: False
• enable_jit_checkpoint: False
• push_to_hub: False
• hub_private_repo: None
• hub_model_id: None
• hub_strategy: every_save
• hub_always_push: False
• hub_revision: None
• load_best_model_at_end: False
• ignore_data_skip: False
• restore_callback_states_from_checkpoint: False
• full_determinism: False
• seed: 42
• data_seed: None
• use_cpu: False
• 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
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• dataloader_prefetch_factor: None
• remove_unused_columns: True
• label_names: None
• train_sampling_strategy: random
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• ddp_static_graph: None
• ddp_backend: None
• ddp_timeout: 1800
• fsdp: None
• fsdp_config: None
• deepspeed: None
• debug: []
• skip_memory_metrics: True
• do_predict: False
• resume_from_checkpoint: None
• warmup_ratio: None
• local_rank: -1
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin
• router_mapping: {}
• learning_rate_mapping: {}

Training Time

• Training: 1.2 minutes

Framework Versions

• Python: 3.12.13
• Sentence Transformers: 5.5.1
• Transformers: 5.12.0
• PyTorch: 2.11.0+cu128
• Accelerate: 1.14.0
• Datasets: 4.0.0
• Tokenizers: 0.22.2

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",
}

MultipleNegativesRankingLoss

@misc{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}