Speedsy/modernbert-base-tr-uncased-colbert-6000

PyLate model based on artiwise-ai/modernbert-base-tr-uncased

This is a PyLate model finetuned from artiwise-ai/modernbert-base-tr-uncased on the train 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: artiwise-ai/modernbert-base-tr-uncased
• Document Length: 180 tokens
• Query Length: 32 tokens
• Output Dimensionality: 128 tokens
• Similarity Function: MaxSim
• Training Dataset:
  • train
• Language: en

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': 179, 'do_lower_case': False}) with Transformer model: ModernBertModel 
  (1): Dense({'in_features': 768, 'out_features': 128, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity'})
)

Usage

First install the PyLate library:

pip install -U pylate

Retrieval

PyLate provides a streamlined interface to index and retrieve documents using ColBERT models. The index leverages the Voyager HNSW index to efficiently handle document embeddings and enable fast retrieval.

Indexing documents

First, load the ColBERT model and initialize the Voyager 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=pylate_model_id,
)

# Step 2: Initialize the Voyager index
index = indexes.Voyager(
    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.Voyager(
    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=pylate_model_id,
)

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,
)

Evaluation

Metrics

Py Late Information Retrieval

• Dataset: ['NanoDBPedia', 'NanoFiQA2018', 'NanoHotpotQA', 'NanoMSMARCO', 'NanoNQ', 'NanoSCIDOCS']
• Evaluated with pylate.evaluation.pylate_information_retrieval_evaluator.PyLateInformationRetrievalEvaluator

Metric	NanoDBPedia	NanoFiQA2018	NanoHotpotQA	NanoMSMARCO	NanoNQ	NanoSCIDOCS
MaxSim_accuracy@1	0.8	0.5	0.9	0.48	0.5	0.42
MaxSim_accuracy@3	0.9	0.66	0.98	0.62	0.68	0.6
MaxSim_accuracy@5	0.92	0.68	1.0	0.66	0.78	0.68
MaxSim_accuracy@10	0.98	0.74	1.0	0.86	0.84	0.76
MaxSim_precision@1	0.8	0.5	0.9	0.48	0.5	0.42
MaxSim_precision@3	0.6533	0.2933	0.5	0.2067	0.2333	0.2867
MaxSim_precision@5	0.608	0.216	0.324	0.132	0.16	0.236
MaxSim_precision@10	0.534	0.124	0.168	0.086	0.09	0.152
MaxSim_recall@1	0.0857	0.2759	0.45	0.48	0.49	0.0887
MaxSim_recall@3	0.177	0.4291	0.75	0.62	0.67	0.1777
MaxSim_recall@5	0.2344	0.484	0.81	0.66	0.74	0.2437
MaxSim_recall@10	0.3716	0.5324	0.84	0.86	0.8	0.3117
MaxSim_ndcg@10	0.656	0.482	0.8121	0.6457	0.6587	0.3181
MaxSim_mrr@10	0.8562	0.5794	0.9383	0.5813	0.6164	0.5282
MaxSim_map@100	0.5105	0.4227	0.7456	0.5848	0.612	0.2444

Pylate Custom Nano BEIR

• Dataset: NanoBEIR_mean
• Evaluated with pylate_nano_beir_evaluator.PylateCustomNanoBEIREvaluator

Metric	Value
MaxSim_accuracy@1	0.6
MaxSim_accuracy@3	0.74
MaxSim_accuracy@5	0.7867
MaxSim_accuracy@10	0.8633
MaxSim_precision@1	0.6
MaxSim_precision@3	0.3622
MaxSim_precision@5	0.2793
MaxSim_precision@10	0.1923
MaxSim_recall@1	0.3117
MaxSim_recall@3	0.4706
MaxSim_recall@5	0.5287
MaxSim_recall@10	0.6193
MaxSim_ndcg@10	0.5954
MaxSim_mrr@10	0.6833
MaxSim_map@100	0.52

Training Details

Training Dataset

train

• Dataset: train at bd034f5
• Size: 443,147 training samples
• Columns: query_id, document_ids, and scores
• Approximate statistics based on the first 1000 samples:

  	query_id	document_ids	scores
  type	string	list	list
  details	min: 5 tokens mean: 6.21 tokens max: 8 tokens	size: 32 elements	size: 32 elements

• Samples:

  query_id	document_ids	scores
  817836	['2716076', '6741935', '2681109', '5562684', '3507339', ...]	[1.0, 0.7059561610221863, 0.21702419221401215, 0.38270196318626404, 0.20812414586544037, ...]
  1045170	['5088671', '2953295', '8783471', '4268439', '6339935', ...]	[1.0, 0.6493034362792969, 0.0692221149802208, 0.17963139712810516, 0.6697239875793457, ...]
  1069432	['3724008', '314949', '8657336', '7420456', '879004', ...]	[1.0, 0.3706032931804657, 0.3508036434650421, 0.2823200523853302, 0.17563475668430328, ...]

• Loss: pylate.losses.distillation.Distillation

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• gradient_accumulation_steps: 2
• learning_rate: 3e-05
• num_train_epochs: 1
• bf16: True

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 8
• per_device_eval_batch_size: 8
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 2
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 3e-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: 0.0
• 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
• 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: 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}
• 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: None
• hub_always_push: False
• 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
• 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
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	NanoDBPedia_MaxSim_ndcg@10	NanoFiQA2018_MaxSim_ndcg@10	NanoHotpotQA_MaxSim_ndcg@10	NanoMSMARCO_MaxSim_ndcg@10	NanoNQ_MaxSim_ndcg@10	NanoSCIDOCS_MaxSim_ndcg@10	NanoBEIR_mean_MaxSim_ndcg@10
0.0036	100	0.0649	-	-	-	-	-	-	-
0.0072	200	0.0559	-	-	-	-	-	-	-
0.0108	300	0.0518	-	-	-	-	-	-	-
0.0144	400	0.051	-	-	-	-	-	-	-
0.0181	500	0.0492	0.6421	0.3808	0.7993	0.5565	0.5826	0.3050	0.5444
0.0217	600	0.0467	-	-	-	-	-	-	-
0.0253	700	0.0451	-	-	-	-	-	-	-
0.0289	800	0.0443	-	-	-	-	-	-	-
0.0325	900	0.0443	-	-	-	-	-	-	-
0.0361	1000	0.0437	0.6449	0.4015	0.8003	0.5437	0.6092	0.3134	0.5522
0.0397	1100	0.0433	-	-	-	-	-	-	-
0.0433	1200	0.0427	-	-	-	-	-	-	-
0.0469	1300	0.0414	-	-	-	-	-	-	-
0.0505	1400	0.0417	-	-	-	-	-	-	-
0.0542	1500	0.0418	0.6412	0.4285	0.8154	0.5866	0.6181	0.3219	0.5686
0.0578	1600	0.0404	-	-	-	-	-	-	-
0.0614	1700	0.0417	-	-	-	-	-	-	-
0.0650	1800	0.0407	-	-	-	-	-	-	-
0.0686	1900	0.0398	-	-	-	-	-	-	-
0.0722	2000	0.0401	0.6499	0.4354	0.8150	0.5610	0.6445	0.3152	0.5702
0.0758	2100	0.0404	-	-	-	-	-	-	-
0.0794	2200	0.0395	-	-	-	-	-	-	-
0.0830	2300	0.0404	-	-	-	-	-	-	-
0.0867	2400	0.0393	-	-	-	-	-	-	-
0.0903	2500	0.0387	0.6571	0.4435	0.8112	0.5786	0.6809	0.3232	0.5824
0.0939	2600	0.0397	-	-	-	-	-	-	-
0.0975	2700	0.0393	-	-	-	-	-	-	-
0.1011	2800	0.0384	-	-	-	-	-	-	-
0.1047	2900	0.0382	-	-	-	-	-	-	-
0.1083	3000	0.0381	0.6437	0.4751	0.8175	0.5711	0.6422	0.3203	0.5783
0.1119	3100	0.0382	-	-	-	-	-	-	-
0.1155	3200	0.0381	-	-	-	-	-	-	-
0.1191	3300	0.0385	-	-	-	-	-	-	-
0.1228	3400	0.0374	-	-	-	-	-	-	-
0.1264	3500	0.0382	0.6437	0.4833	0.8282	0.5955	0.6436	0.3190	0.5856
0.1300	3600	0.0365	-	-	-	-	-	-	-
0.1336	3700	0.0379	-	-	-	-	-	-	-
0.1372	3800	0.0376	-	-	-	-	-	-	-
0.1408	3900	0.0376	-	-	-	-	-	-	-
0.1444	4000	0.0378	0.6511	0.4760	0.8151	0.5806	0.6874	0.3140	0.5874
0.1480	4100	0.0365	-	-	-	-	-	-	-
0.1516	4200	0.0362	-	-	-	-	-	-	-
0.1553	4300	0.0374	-	-	-	-	-	-	-
0.1589	4400	0.0359	-	-	-	-	-	-	-
0.1625	4500	0.0368	0.6530	0.4458	0.8122	0.6101	0.6896	0.3174	0.5880
0.1661	4600	0.0356	-	-	-	-	-	-	-
0.1697	4700	0.0364	-	-	-	-	-	-	-
0.1733	4800	0.0352	-	-	-	-	-	-	-
0.1769	4900	0.0357	-	-	-	-	-	-	-
0.1805	5000	0.0366	0.6611	0.4680	0.8152	0.6260	0.6715	0.3252	0.5945
0.1841	5100	0.0358	-	-	-	-	-	-	-
0.1877	5200	0.0366	-	-	-	-	-	-	-
0.1914	5300	0.0348	-	-	-	-	-	-	-
0.1950	5400	0.036	-	-	-	-	-	-	-
0.1986	5500	0.0337	0.6595	0.4823	0.8162	0.6241	0.6620	0.3216	0.5943
0.2022	5600	0.0347	-	-	-	-	-	-	-
0.2058	5700	0.0361	-	-	-	-	-	-	-
0.2094	5800	0.0356	-	-	-	-	-	-	-
0.2130	5900	0.0359	-	-	-	-	-	-	-
0.2166	6000	0.0359	0.6560	0.4820	0.8121	0.6457	0.6587	0.3181	0.5954

Framework Versions

• Python: 3.11.12
• Sentence Transformers: 4.0.2
• PyLate: 1.2.0
• Transformers: 4.48.2
• PyTorch: 2.6.0+cu124
• Accelerate: 1.6.0
• Datasets: 3.6.0
• Tokenizers: 0.21.1

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"
}

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}
}