ModernBERT-base-pylate-pairwise-8e-05-qv1-dv1-embsize128

PyLate model based on answerdotai/ModernBERT-base

This is a PyLate model finetuned from answerdotai/ModernBERT-base on the msmarco-co-condenser-margin-mse-sym-mnrl-mean-v1 dataset. It maps sentences & paragraphs to sequences of 128-dimensional dense vectors and can be used for semantic textual similarity using the MaxSim operator.

Model Details

Model Description

• Model Type: PyLate model
• Base model: answerdotai/ModernBERT-base <!-- at revision 8949b909ec900327062f0ebf497f51aef5e6f0c8 -->
• Document Length: 180 tokens
• Query Length: 32 tokens
• Output Dimensionality: 128 tokens
• Similarity Function: MaxSim
• Training Dataset:
  • msmarco-co-condenser-margin-mse-sym-mnrl-mean-v1
• Language: en

<!-- - **License:** Unknown -->

Model Sources

• Documentation: PyLate Documentation
• Repository: PyLate on GitHub
• Hugging Face: PyLate models on Hugging Face

Full Model Architecture

ColBERT(
  (0): Transformer({'max_seq_length': 8192, 'do_lower_case': False, 'architecture': 'ModernBertModel'})
  (1): Dense({'in_features': 768, 'out_features': 128, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity', 'use_residual': False})
)

Usage

First install the PyLate library:

pip install -U pylate

Retrieval

Use this model with PyLate to index and retrieve documents. The index uses FastPLAID for efficient similarity search.

Indexing documents

Load the ColBERT model and initialize the PLAID index, then encode and index your documents:

from pylate import indexes, models, retrieve

# Step 1: Load the ColBERT model
model = models.ColBERT(
    model_name_or_path="PrasannSinghal/ModernBERT-base-pylate-pairwise-8e-05-qv1-dv1-embsize128",
)

# Step 2: Initialize the PLAID index
index = indexes.PLAID(
    index_folder="pylate-index",
    index_name="index",
    override=True,  # This overwrites the existing index if any
)

# Step 3: Encode the documents
documents_ids = ["1", "2", "3"]
documents = ["document 1 text", "document 2 text", "document 3 text"]

documents_embeddings = model.encode(
    documents,
    batch_size=32,
    is_query=False,  # Ensure that it is set to False to indicate that these are documents, not queries
    show_progress_bar=True,
)

# Step 4: Add document embeddings to the index by providing embeddings and corresponding ids
index.add_documents(
    documents_ids=documents_ids,
    documents_embeddings=documents_embeddings,
)

Note that you do not have to recreate the index and encode the documents every time. Once you have created an index and added the documents, you can re-use the index later by loading it:

# To load an index, simply instantiate it with the correct folder/name and without overriding it
index = indexes.PLAID(
    index_folder="pylate-index",
    index_name="index",
)

Retrieving top-k documents for queries

Once the documents are indexed, you can retrieve the top-k most relevant documents for a given set of queries. To do so, initialize the ColBERT retriever with the index you want to search in, encode the queries and then retrieve the top-k documents to get the top matches ids and relevance scores:

# Step 1: Initialize the ColBERT retriever
retriever = retrieve.ColBERT(index=index)

# Step 2: Encode the queries
queries_embeddings = model.encode(
    ["query for document 3", "query for document 1"],
    batch_size=32,
    is_query=True,  #  # Ensure that it is set to False to indicate that these are queries
    show_progress_bar=True,
)

# Step 3: Retrieve top-k documents
scores = retriever.retrieve(
    queries_embeddings=queries_embeddings,
    k=10,  # Retrieve the top 10 matches for each query
)

Reranking

If you only want to use the ColBERT model to perform reranking on top of your first-stage retrieval pipeline without building an index, you can simply use rank function and pass the queries and documents to rerank:

from pylate import rank, models

queries = [
    "query A",
    "query B",
]

documents = [
    ["document A", "document B"],
    ["document 1", "document C", "document B"],
]

documents_ids = [
    [1, 2],
    [1, 3, 2],
]

model = models.ColBERT(
    model_name_or_path="PrasannSinghal/ModernBERT-base-pylate-pairwise-8e-05-qv1-dv1-embsize128",
)

queries_embeddings = model.encode(
    queries,
    is_query=True,
)

documents_embeddings = model.encode(
    documents,
    is_query=False,
)

reranked_documents = rank.rerank(
    documents_ids=documents_ids,
    queries_embeddings=queries_embeddings,
    documents_embeddings=documents_embeddings,
)

<!-- ### Direct Usage (Transformers) <details><summary>Click to see the direct usage in Transformers</summary> </details> --> <!-- ### Downstream Usage (Sentence Transformers) You can finetune this model on your own dataset. <details><summary>Click to expand</summary> </details> --> <!-- ### Out-of-Scope Use *List how the model may foreseeably be misused and address what users ought not to do with the model.* -->

Evaluation

Metrics

Col BERTTriplet

• Dataset: msmarco-co-condenser-dev
• Evaluated with <code>pylate.evaluation.colbert_triplet.ColBERTTripletEvaluator</code>

<!-- ## Bias, Risks and Limitations *What are the known or foreseeable issues stemming from this model? You could also flag here known failure cases or weaknesses of the model.* --> <!-- ### Recommendations *What are recommendations with respect to the foreseeable issues? For example, filtering explicit content.* -->

Training Details

Training Dataset

msmarco-co-condenser-margin-mse-sym-mnrl-mean-v1

• Dataset: msmarco-co-condenser-margin-mse-sym-mnrl-mean-v1 at 84ed2d3
• Size: 1,250,000 training samples
• Columns: <code>query</code>, <code>positive</code>, and <code>negative</code>
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative
  type	string	string	string
  details	<ul><li>min: 7 tokens</li><li>mean: 12.26 tokens</li><li>max: 37 tokens</li></ul>	<ul><li>min: 20 tokens</li><li>mean: 82.14 tokens</li><li>max: 225 tokens</li></ul>	<ul><li>min: 27 tokens</li><li>mean: 83.09 tokens</li><li>max: 439 tokens</li></ul>

• Samples:

  query	positive	negative
  <code>what is the meaning of menu planning</code>	<code>Menu planning is the selection of a menu for an event. Such as picking out the dinner for your wedding or even a meal at a Birthday Party. Menu planning is when you are preparing a calendar of meals and you have to sit down and decide what meat and veggies you want to serve on each certain day.</code>	<code>Menu Costs. In economics, a menu cost is the cost to a firm resulting from changing its prices. The name stems from the cost of restaurants literally printing new menus, but economists use it to refer to the costs of changing nominal prices in general.</code>
  <code>how old is brett butler</code>	<code>Brett Butler is 59 years old. To be more precise (and nerdy), the current age as of right now is 21564 days or (even more geeky) 517536 hours. That's a lot of hours!</code>	<code>Passed in: St. John's, Newfoundland and Labrador, Canada. Passed on: 16/07/2016. Published in the St. John's Telegram. Passed away suddenly at the Health Sciences Centre surrounded by his loving family, on July 16, 2016 Robert (Bobby) Joseph Butler, age 52 years. Predeceased by his special aunt Geri Murrin and uncle Mike Mchugh; grandparents Joe and Margaret Murrin and Jack and Theresa Butler.</code>
  <code>when was the last navajo treaty sign?</code>	<code>In Executive Session, Senate of the United States, July 25, 1868. Resolved, (two-thirds of the senators present concurring,) That the Senate advise and consent to the ratification of the treaty between the United States and the Navajo Indians, concluded at Fort Sumner, New Mexico, on the first day of June, 1868.</code>	<code>Share Treaty of Greenville. The Treaty of Greenville was signed August 3, 1795, between the United States, represented by Gen. Anthony Wayne, and chiefs of the Indian tribes located in the Northwest Territory, including the Wyandots, Delawares, Shawnees, Ottawas, Miamis, and others.</code>

• Loss: <code>pylate.losses.cached_contrastive.CachedContrastive</code>

Evaluation Dataset

msmarco-co-condenser-margin-mse-sym-mnrl-mean-v1

• Dataset: msmarco-co-condenser-margin-mse-sym-mnrl-mean-v1 at 84ed2d3
• Size: 1,000 evaluation samples
• Columns: <code>query</code>, <code>positive</code>, and <code>negative</code>
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative
  type	string	string	string
  details	<ul><li>min: 7 tokens</li><li>mean: 12.2 tokens</li><li>max: 30 tokens</li></ul>	<ul><li>min: 24 tokens</li><li>mean: 83.44 tokens</li><li>max: 244 tokens</li></ul>	<ul><li>min: 26 tokens</li><li>mean: 83.38 tokens</li><li>max: 242 tokens</li></ul>

• Samples:

  query	positive	negative
  <code>what county is holly springs nc in</code>	<code>Holly Springs, North Carolina. Holly Springs is a town in Wake County, North Carolina, United States. As of the 2010 census, the town population was 24,661, over 2½ times its population in 2000. Contents.</code>	<code>The Mt. Holly Springs Park & Resort. One of the numerous trolley routes that carried people around the county at the turn of the century was the Carlisle & Mt. Holly Railway Company. The “Holly Trolley” as it came to be known was put into service by Patricio Russo and made its first run on May 14, 1901.</code>
  <code>how long does nyquil stay in your system</code>	<code>In order to understand exactly how long Nyquil lasts, it is absolutely vital to learn about the various ingredients in the drug. One of the ingredients found in Nyquil is Doxylamine, which is an antihistamine. This specific medication has a biological half-life or 6 to 12 hours. With this in mind, it is possible for the drug to remain in the system for a period of 12 to 24 hours. It should be known that the specifics will depend on a wide variety of different factors, including your age and metabolism.</code>	<code>I confirmed that NyQuil is about 10% alcohol, a higher content than most domestic beers. When I asked about the relatively high proof, I was told that the alcohol dilutes the active ingredients. The alcohol free version is there for customers with addiction issues.. also found that in that version there is twice the amount of DXM. When I asked if I could speak to a chemist or scientist, I was told they didn't have anyone who fit that description there. It’s been eight years since I kicked NyQuil. I've been sober from alcohol for four years.</code>
  <code>what are mineral water</code>	<code>1 Mineral water – water from a mineral spring that contains various minerals, such as salts and sulfur compounds. 2 It comes from a source tapped at one or more bore holes or spring, and originates from a geologically and physically protected underground water source. Mineral water – water from a mineral spring that contains various minerals, such as salts and sulfur compounds. 2 It comes from a source tapped at one or more bore holes or spring, and originates from a geologically and physically protected underground water source.</code>	<code>Minerals for Your Body. Drinking mineral water is beneficial to health and well-being. But it is not only the amount of water you drink that is important-what the water contains is even more essential.inerals for Your Body. Drinking mineral water is beneficial to health and well-being. But it is not only the amount of water you drink that is important-what the water contains is even more essential.</code>

• Loss: <code>pylate.losses.cached_contrastive.CachedContrastive</code>

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 512
• per_device_eval_batch_size: 512
• learning_rate: 8e-05
• num_train_epochs: 1
• warmup_ratio: 0.05
• bf16: True
• dataloader_num_workers: 8
• dataloader_pin_memory: False
• batch_sampler: no_duplicates

All Hyperparameters

<details><summary>Click to expand</summary>

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: no
• prediction_loss_only: True
• per_device_train_batch_size: 512
• per_device_eval_batch_size: 512
• 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: 8e-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: 1
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.05
• 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: None
• local_rank: 1
• ddp_backend: None
• tpu_num_cores: None
• tpu_metrics_debug: False
• debug: []
• dataloader_drop_last: True
• dataloader_num_workers: 8
• 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
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• dataloader_pin_memory: False
• 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: False
• 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: no_duplicates
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

</details>

Training Logs

<details><summary>Click to expand</summary> </details>

Framework Versions

• Python: 3.10.0
• Sentence Transformers: 5.2.2
• PyLate: 1.3.4
• Transformers: 4.56.2
• PyTorch: 2.9.1+cu128
• Accelerate: 1.12.0
• Datasets: 4.5.0
• 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"
}

PyLate

@misc{PyLate,
title={PyLate: Flexible Training and Retrieval for Late Interaction Models},
author={Chaffin, Antoine and Sourty, Raphaël},
url={https://github.com/lightonai/pylate},
year={2024}
}

CachedContrastive

@misc{gao2021scaling,
    title={Scaling Deep Contrastive Learning Batch Size under Memory Limited Setup},
    author={Luyu Gao and Yunyi Zhang and Jiawei Han and Jamie Callan},
    year={2021},
    eprint={2101.06983},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

<!-- ## Glossary *Clearly define terms in order to be accessible across audiences.* --> <!-- ## Model Card Authors *Lists the people who create the model card, providing recognition and accountability for the detailed work that goes into its construction.* --> <!-- ## Model Card Contact *Provides a way for people who have updates to the Model Card, suggestions, or questions, to contact the Model Card authors.* -->