Mercity/memory-retrieval-minilm-l6-v2

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 semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

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

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': 256, 'do_lower_case': False, 'architecture': 'BertModel'})
  (1): Pooling({'word_embedding_dimension': 384, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, '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("Mercity/memory-retrieval-minilm-l6-v2")
# Run inference
sentences = [
    'Provide the amortization schedule difference for CC1 under both prioritization schemas.',
    "Mark's primary long-term goal is to start his own small business within the next four years, requiring significant capital savings to begin operations.",
    "Last spring, Dr. Vasquez received high praise from the Dean for successfully transitioning her entire curriculum to align with a newly adopted, highly structured 'Systems Thinking Framework' mandated by the university board.",
]
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,  0.9104, -0.3007],
#         [ 0.9104,  1.0000, -0.2945],
#         [-0.3007, -0.2945,  1.0000]])

Training Details

Training Dataset

Unnamed Dataset

• Size: 370,023 training samples
• Columns: sentence_0, sentence_1, and sentence_2
• Approximate statistics based on the first 1000 samples:

  	sentence_0	sentence_1	sentence_2
  type	string	string	string
  details	min: 11 tokens mean: 31.57 tokens max: 76 tokens	min: 17 tokens mean: 36.18 tokens max: 64 tokens	min: 13 tokens mean: 32.21 tokens max: 154 tokens

• Samples:

  sentence_0	sentence_1	sentence_2
  Yoga by the sea for Nina's soul? Add scenic drives that whisper sweet nothings.	Nina developed a strong, vocal aversion to driving manual transmission cars after a stressful rental experience in Tuscany in 2018.	Sofia recently shared that her grandmother, who is a major influencer in the family, is recovering from a minor surgery and requires quiet, low-stress environments for the next few months.
  Targeting analysis of 200 sites to sway city green space policies; realistic under these limits?	Dr. Patel has a strong, established working relationship with the City Planning Department head, who is personally invested in seeing the pollinator data incorporated into the 2026 municipal budget review.	Alex reviewed the seed funding agreement which stipulated that 10% ($50,000) of the capital must be reserved solely for executive bonuses upon Series A closing.
  Gluten-free options avoiding cross-contamination in our small kitchen? I truly value this.	Lisa has a very close relationship with her neighbor, Sarah, who is a professional chef specializing in South American cuisine and often offers to help prep on Sundays.	Alex's agent advised varying sentence structure with occasional future tense projections during character betrayals to heighten emotional stakes in the plot.

• Loss: TripletLoss with these parameters:

  {
      "distance_metric": "TripletDistanceMetric.COSINE",
      "triplet_margin": 0.5
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_steps: None
• torch_empty_cache_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: 3
• max_steps: -1
• 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
• 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
• 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: 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, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• parallelism_config: 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
• project: huggingface
• trackio_space_id: trackio
• 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: 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: no
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• liger_kernel_config: None
• eval_use_gather_object: False
• average_tokens_across_devices: True
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss
0.0432	500	0.3372
0.0865	1000	0.1651
0.1297	1500	0.1505
0.1730	2000	0.1367
0.2162	2500	0.128
0.2594	3000	0.1213
0.3027	3500	0.1167
0.3459	4000	0.1087
0.3891	4500	0.1017
0.4324	5000	0.1019
0.4756	5500	0.096
0.5189	6000	0.0963
0.5621	6500	0.0883
0.6053	7000	0.0879
0.6486	7500	0.0865
0.6918	8000	0.0822
0.7350	8500	0.0815
0.7783	9000	0.0784
0.8215	9500	0.0754
0.8648	10000	0.0767
0.9080	10500	0.0732
0.9512	11000	0.0713
0.9945	11500	0.0677
1.0	11564	-
1.0377	12000	0.0611
1.0809	12500	0.0572
1.1242	13000	0.0596
1.1674	13500	0.0576
1.2107	14000	0.0562
1.2539	14500	0.0544
1.2971	15000	0.0543
1.3404	15500	0.0544
1.3836	16000	0.0533
1.4268	16500	0.0515
1.4701	17000	0.0501
1.5133	17500	0.0515
1.5566	18000	0.0494
1.5998	18500	0.0495
1.6430	19000	0.0477
1.6863	19500	0.0471
1.7295	20000	0.0478
1.7727	20500	0.0453
1.8160	21000	0.0447
1.8592	21500	0.0463
1.9025	22000	0.0462
1.9457	22500	0.0445
1.9889	23000	0.0436
2.0	23128	-
2.0322	23500	0.0376
2.0754	24000	0.0377
2.1186	24500	0.0364
2.1619	25000	0.0381
2.2051	25500	0.0382
2.2484	26000	0.0359
2.2916	26500	0.036
2.3348	27000	0.0365
2.3781	27500	0.0362
2.4213	28000	0.036
2.4645	28500	0.0359
2.5078	29000	0.0364
2.5510	29500	0.0355
2.5943	30000	0.0354
2.6375	30500	0.0348
2.6807	31000	0.0362
2.7240	31500	0.0345
2.7672	32000	0.0351
2.8104	32500	0.0345
2.8537	33000	0.0351
2.8969	33500	0.0339
2.9402	34000	0.0354
2.9834	34500	0.0356
3.0	34692	-

Framework Versions

• Python: 3.12.3
• Sentence Transformers: 5.1.2
• Transformers: 4.57.1
• PyTorch: 2.8.0+cu128
• Accelerate: 1.11.0
• Datasets: 4.4.0
• Tokenizers: 0.22.1

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",
}

TripletLoss

@misc{hermans2017defense,
    title={In Defense of the Triplet Loss for Person Re-Identification},
    author={Alexander Hermans and Lucas Beyer and Bastian Leibe},
    year={2017},
    eprint={1703.07737},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}