Tejasw1 / bge-base-case-law-v1

SentenceTransformer based on BAAI/bge-base-en-v1.5

This is a sentence-transformers model finetuned from BAAI/bge-base-en-v1.5 on the json dataset. 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
• Training Dataset:
  • json
• 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("Tejasw1/bge-base-case-law-v1")
# Run inference
sentences = [
    "**1. Key Legal Issues and Holdings:**\n\n* **Construction of a Will:** The main legal issue is the interpretation of the will left by Kothandarama Ayyar, a Hindu inhabitant of the district of Tanjore, to determine the disposition of his properties.\n* **Adoption and Inheritance:** The case revolves around the application of the will's provisions regarding adoption and inheritance, particularly with regards to the properties in dispute.\n* **Construction of Specific Provisions:** The court considered the construction of specific provisions in the will, including Paras 5, 13, and other relevant paragraphs.\n\n**2. Significant Facts of the Case:**\n\n* The testator, Kothandarama Ayyar, died on 25-4-1905, leaving behind his widow, Parbati, and two daughters, Nagammal and Gnanambal.\n* The testator executed his last will on 13-3-1905, giving his widow authority to adopt a son of Gnanambal or a nephew's son of the testator.\n* The will provides for the distribution of the testator's properties among his family members and charities.\n* The dispute revolves around the properties in Kothangudi and Injigudi, which are mentioned in Paras 5 and 13 of the will.\n\n**3. Court's Ruling:**\n\n* The Supreme Court upheld the construction of the will by the High Court, which held that Para 5 of the will was not operative in the present case.\n* The court rejected the argument that Para 5 was meant to be operative only if Gnanambal's son was adopted by the widow.\n* The court held that the testator's main desire was that his widow should adopt the son of his daughter Gnanambal, and that the provisions made for the two daughters, the widow, and the adoptive mother were meant to be applicable under all three contingencies referred to in the will.\n* The court allowed the appeal, setting aside the judgment and decree of the High Court, and restored the judgment and decree of the Subordinate Judge.\n\n**4. Citations:**\n\n* **Venkata Narasimha Appa Row v. Parthasarathy Appa Row**, Privy Council\n* **Edwards, In re, Jones v. Jones**, Romer, L.J.\n* **Venkata Narasimha Appa Row v. Parthasarathy Appa Row**, (1913-14) 41 IA 51\n* **Jones v. Jones**, (1906) 1 Ch 570 (CA)",
    "In cases involving wills, how do courts balance the testator's intentions with the rights of surviving family members?",
    'How does the U.P. Urban Buildings (Regulation of Letting, Rent & Eviction) Act, 1972 determine the applicability of rent control laws to newly constructed buildings?',
]
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

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator

Training Details

Training Dataset

json

• Dataset: json
• Size: 16,465 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 2 tokens mean: 26.38 tokens max: 72 tokens	min: 333 tokens mean: 490.59 tokens max: 512 tokens

• Samples:

  anchor	positive
  What factors do courts consider when evaluating the reliability of eyewitness testimonies in murder trials?	1. Key Legal Issues and Holdings: * Culpable Homicide: The court considered the application of Section 302 of the Indian Penal Code, 1860, which deals with punishment for culpable homicide not amounting to murder. * Section 302 IPC: The court upheld the conviction of the accused under Section 302 IPC for the death of Ishwardeen. * Section 34 IPC: The court also upheld the conviction of the accused under Section 34 IPC for the death of Ishwardeen, considering the common intention of the accused. 2. Significant Facts of the Case: * The deceased, Ishwardeen, was killed in a alleged incident involving the accused, Bhagwan Das, Sheo Mohan, and Shanker @ Sheo Shanker. * The incident occurred on August 18, 1983, at the house of Bhagwan Das, where Ishwardeen and his family were residing. * The accused had allegedly demanded rent from Ishwardeen and had threatened to evict him from the house. * Ishwardeen was killed by the accused, who allegedly gave him 8-9 knife blows. * The case was registered under Section 307 IPC, but it was later converted to Section 302 IPC after Ishwardeen's death. 3. Court's Ruling: * The High Court upheld the conviction of Bhagwan Das, Sheo Mohan, and Shanker @ Sheo Shanker under Section 302 IPC and Section 34 IPC. * The court rejected the defense argument that the testimony of the eyewitnesses, Kamla Devi and Subhash, could not be relied upon. * The court relied on the testimony of the eyewitnesses, which was corroborated by the medical evidence. * The court also relied on the post-mortem report, which showed that the antemortem injuries were possible to be caused by a knife at around 9:00 p.m. on August 18, 1983. * The court held that the accused had a common intention to commit the murder of Ishwardeen, and therefore, the conviction under Section 302 IPC with the aid of Section 34 IPC was upheld. * The court also rejected the defense argument that the accused had no motive to commit the murder, and held that the presence of semi-digested food in the stomach of the deceased did not contradict the prosecution's case. 4. Citations: * Dalip Singh v. State of Punjab * Section 302 IPC * Section 34 IPC * Thaman Kumar v. State of Union Territory of Chandigarh * State of H.P. v. Jeet Singh * Appa Bhat v. State of Gujarat * Krishna Mochi v. State of Bihar * Israr v. State of U.P. * Gali Venkataiah v. State of A.P. * Masalti v. State of U.P. * Vadivelu Thevar v. State of Madras * Galivenkataiah v. State of A.P.
  What principles guide the court's decisions on wage fixation in cases involving government undertakings?	1. Key Legal Issues and Holdings: * Wage Structure: The main legal issue is whether the wage structure of a government undertaking in the public sector should be different from that of an undertaking in the private sector. * Section 10(1)(d) of the Industrial Disputes Act, 1947: The court considered the applicability of this provision in the context of wage fixation. * Article 39 and 43 of the Directive Principles of State Policy: The court examined the constitutional implications of making a distinction between laborers in the public and private sectors. * Region-cum-Industry Principle: The court upheld the principle of region-cum-industry in wage fixation. * Gratuity Scheme: The court considered the validity of the gratuity scheme introduced by the Tribunal, which provided for payment of gratuity to workmen upon retirement or termination of service. * Fair Wages: The court examined the issue of fair wages and whether the Tribunal had violated the accepted norms in fixing the wage structure and dearness allowance. * Retirement Age: The court held that the retirement age of the employees should be raised to 60 years, and the discretion to continue employees beyond 58 years should be removed. 2. Significant Facts of the Case: * The case involves a government undertaking, Hindustan Antibiotics Limited, and its workmen. * The company is a limited one, but it is financed entirely by the Central Government. * The workmen presented a charter of fifteen demands to the company, including demands related to wage fixation. * The Industrial Tribunal, Bombay, made an award dated October 8, 1963, which was challenged by the company and the workmen. * The company argued that the wage structure in the public sector should be different from that in the private sector. * The workmen argued that the character of the employer is irrelevant in wage fixation. 3. Court's Ruling: * The court upheld the principle of region-cum-industry in wage fixation. * The court rejected the argument that the character of the employer is relevant in determining wage structures. * The court held that the distinction between public and private sectors is not justified in wage fixation. * The court also noted that the constitutional directive principles of equal pay for equal work and a living wage for all workers should be followed. * The court validated the gratuity scheme introduced by the Tribunal. * The court rejected the contention that the Tribunal had violated the accepted norms in fixing the wage structure and dearness allowance. * The court raised the retirement age of employees to 60 years and removed the discretion to continue employees beyond 58 years. * The court dismissed the appeals by the company and the workmen, except for modifying the award regarding the age of retirement. 4. Citations: * Crown Aluminium Works v. Workmen * Express Newspapers (Private) Ltd. v. Union of India * French Motor Car Co. Ltd. v. Workmen * Hindustan Times Ltd., New Delhi v. Workmen * Other relevant cases and reports cited in the judgment.
  - What role does the recommendation of a State Government play in the land exemption process under the Punjab Towns Improvement Act, 1922?	1. Key Legal Issues and Holdings: * Acquisition of Land: The main legal issue is the acquisition of land by the Improvement Trust under the Punjab Towns Improvement Act, 1922. * Exemption of Orchards: The court considered whether the Improvement Trust was bound to exempt orchards from acquisition under Section 56 of the Act. * Article 14 of the Constitution: The court held that the Improvement Trust did not violate Article 14 of the Constitution by exempting some orchards while acquiring others. * Quasi-Judicial Capacity: The court held that the Improvement Trust acts in a quasi-judicial capacity when determining claims under Section 56 of the Act, but is not bound to give reasons for its decisions. * Locus Standi: The court observed that the appellants had no locus standi to invoke Section 56 of the Act, as the acquisition of their land had not been discovered to be unnecessary for the execution of the scheme. * Power to Exempt Lands: The court held that the Improvement Trust did not possess the power to exempt lands from the scheme under Section 56 of the Act. 2. Significant Facts of the Case: * The Improvement Trust framed a development scheme in 1961 under Section 24 read with Section 28(2) of the Act. * The scheme covered an area of approximately 128 acres, and the Trust acquired the land, including the appellants' land, in 1964. * The appellants applied to the State Government for exempting their land from acquisition on the ground that it contained a fully developed orchard. * The State Government recommended their case to the Chairman of the Improvement Trust, but the Trust refused to exempt their land. * The appellants claimed that the Trust had exempted similar orchards of other persons, and that this was a violation of Article 14 of the Constitution. 3. Court's Ruling: * The High Court initially allowed the appellants' writ petition, directing the Trust to allow them a full opportunity of hearing regarding their case for exemption. * The Trust refused to exempt their land, and the appellants appealed to the High Court. * The High Court dismissed the appeal, holding that the Trust had given reasons for its decision and that the appellants had not shown how their land was unnecessary for the execution of the scheme. * The Supreme Court dismissed the appeal, holding that the Improvement Trust did not violate Article 14 of the Constitution by exempting some orchards while acquiring others. 4. Citations: * Punjab Towns Improvement Act, 1922 * Article 14 of the Constitution of India * Section 56 of the Punjab Towns Improvement Act, 1922 * Section 24 read with Section 28(2) of the Punjab Towns Improvement Act, 1922 * Section 43 of the Punjab Towns Improvement Act, 1922

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          768,
          512
      ],
      "matryoshka_weights": [
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy : epoch
• per_device_train_batch_size : 16
• gradient_accumulation_steps : 8
• learning_rate : 2e-05
• num_train_epochs : 4
• lr_scheduler_type : cosine
• warmup_ratio : 0.1
• bf16 : True
• tf32 : True
• load_best_model_at_end : True
• optim : adamw_torch_fused
• batch_sampler : no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir : False
• do_predict : False
• eval_strategy : epoch
• prediction_loss_only : True
• per_device_train_batch_size : 16
• per_device_eval_batch_size : 8
• per_gpu_train_batch_size : None
• per_gpu_eval_batch_size : None
• gradient_accumulation_steps : 8
• eval_accumulation_steps : None
• torch_empty_cache_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 : 4
• 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 : True
• fp16 : False
• fp16_opt_level : O1
• half_precision_backend : auto
• bf16_full_eval : False
• fp16_full_eval : False
• tf32 : True
• 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_fused
• 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
• use_liger_kernel : False
• eval_use_gather_object : False
• batch_sampler : no_duplicates
• multi_dataset_batch_sampler : proportional

Training Logs

Click to expand

Epoch	Step	Training Loss	dim_512_cosine_map@100	dim_768_cosine_map@100
0.0777	10	1.58	-	-
0.1553	20	1.0799	-	-
0.2330	30	0.6653	-	-
0.3107	40	0.4524	-	-
0.3883	50	0.3962	-	-
0.4660	60	0.3472	-	-
0.5437	70	0.3481	-	-
0.6214	80	0.3034	-	-
0.6990	90	0.3612	-	-
0.7767	100	0.2497	-	-
0.8544	110	0.2424	-	-
0.9320	120	0.3037	-	-
0.9942	128	-	0.2359	0.2435
1.0097	130	0.2795	-	-
1.0874	140	0.2519	-	-
1.1650	150	0.2414	-	-
1.2427	160	0.1837	-	-
1.3204	170	0.1734	-	-
1.3981	180	0.1462	-	-
1.4757	190	0.1593	-	-
1.5534	200	0.1648	-	-
1.6311	210	0.1593	-	-
1.7087	220	0.1737	-	-
1.7864	230	0.1237	-	-
1.8641	240	0.1205	-	-
1.9417	250	0.1611	-	-
1.9961	257	-	0.2376	0.2424
2.0194	260	0.1674	-	-
2.0971	270	0.135	-	-
2.1748	280	0.1464	-	-
2.2524	290	0.1119	-	-
2.3301	300	0.089	-	-
2.4078	310	0.0774	-	-
2.4854	320	0.1039	-	-
2.5631	330	0.1218	-	-
2.6408	340	0.1001	-	-
2.7184	350	0.1072	-	-
2.7961	360	0.0774	-	-
2.8738	370	0.0855	-	-
2.9515	380	0.1096	-	-
2.9981	386	-	0.2402	0.2381
3.0291	390	0.1076	-	-
3.1068	400	0.1019	-	-
3.1845	410	0.1139	-	-
3.2621	420	0.0732	-	-
3.3398	430	0.0831	-	-
3.4175	440	0.0613	-	-
3.4951	450	0.092	-	-
3.5728	460	0.0891	-	-
3.6505	470	0.0896	-	-
3.7282	480	0.0861	-	-
3.8058	490	0.0743	-	-
3.8835	500	0.077	-	-
3.9612	510	0.1056	-	-
3.9767	512	-	0.2393	0.2393
0.0777	10	0.3691	-	-
0.1553	20	0.3126	-	-
0.2330	30	0.279	-	-
0.3107	40	0.2477	-	-
0.3883	50	0.2436	-	-
0.4660	60	0.2307	-	-
0.5437	70	0.2487	-	-
0.6214	80	0.2463	-	-
0.6990	90	0.2965	-	-
0.7767	100	0.2101	-	-
0.8544	110	0.1999	-	-
0.9320	120	0.2561	-	-
0.9942	128	-	0.2399	0.242
1.0097	130	0.2504	-	-
1.0874	140	0.246	-	-
1.1650	150	0.2043	-	-
1.2427	160	0.171	-	-
1.3204	170	0.1499	-	-
1.3981	180	0.1402	-	-
1.4757	190	0.1379	-	-
1.5534	200	0.156	-	-
1.6311	210	0.1669	-	-
1.7087	220	0.1578	-	-
1.7864	230	0.1157	-	-
1.8641	240	0.1279	-	-
1.9417	250	0.1766	-	-
1.9961	257	-	0.2386	0.2410
2.0194	260	0.1693	-	-
2.0971	270	0.1424	-	-
2.1748	280	0.1517	-	-
2.2524	290	0.1151	-	-
2.3301	300	0.0974	-	-
2.4078	310	0.083	-	-
2.4854	320	0.1021	-	-
2.5631	330	0.1305	-	-
2.6408	340	0.1102	-	-
2.7184	350	0.1118	-	-
2.7961	360	0.089	-	-
2.8738	370	0.1111	-	-
2.9515	380	0.145	-	-
2.9981	386	-	0.2372	0.2400
3.0291	390	0.1115	-	-
3.1068	400	0.1036	-	-
3.1845	410	0.1164	-	-
3.2621	420	0.0728	-	-
3.3398	430	0.0879	-	-
3.4175	440	0.0657	-	-
3.4951	450	0.0825	-	-
3.5728	460	0.0986	-	-
3.6505	470	0.1074	-	-
3.7282	480	0.0923	-	-
3.8058	490	0.078	-	-
3.8835	500	0.0962	-	-
3.9612	510	0.1078	-	-
3.9767	512	-	0.2378	0.2398

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.5
• Sentence Transformers: 3.1.1
• Transformers: 4.45.2
• PyTorch: 2.5.1+cu124
• Accelerate: 0.34.2
• Datasets: 3.0.0
• Tokenizers: 0.20.1

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}
}