ayushexel/colbert-ModernBERT-base-2-neg-1-epoch-gooaq-1995000

PyLate model based on answerdotai/ModernBERT-base

This is a PyLate model finetuned from answerdotai/ModernBERT-base. 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
• Document Length: 180 tokens
• Query Length: 32 tokens
• Output Dimensionality: 128 tokens
• Similarity Function: MaxSim

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=ayushexel/colbert-ModernBERT-base-2-neg-1-epoch-gooaq-1995000,
)

# 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=ayushexel/colbert-ModernBERT-base-2-neg-1-epoch-gooaq-1995000,
)

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

Col BERTTriplet

• Evaluated with pylate.evaluation.colbert_triplet.ColBERTTripletEvaluator

Metric	Value
accuracy	0.4942

Training Details

Training Dataset

Unnamed Dataset

• Size: 3,787,119 training samples
• Columns: question, answer, and negative
• Approximate statistics based on the first 1000 samples:

  	question	answer	negative
  type	string	string	string
  details	min: 9 tokens mean: 13.19 tokens max: 21 tokens	min: 18 tokens mean: 31.78 tokens max: 32 tokens	min: 13 tokens mean: 31.68 tokens max: 32 tokens

• Samples:

  question	answer	negative
  how is sickle cell anemia beneficial?	Having two copies of the mutated genes cause sickle cell anemia, but having just one copy does not, and can actually protect against malaria - an example of how mutations are sometimes beneficial.	Description. Sickle cell disease is a group of disorders that affects hemoglobin, the molecule in red blood cells that delivers oxygen to cells throughout the body. People with this disorder have atypical hemoglobin molecules called hemoglobin S, which can distort red blood cells into a sickle, or crescent, shape.
  how is sickle cell anemia beneficial?	Having two copies of the mutated genes cause sickle cell anemia, but having just one copy does not, and can actually protect against malaria - an example of how mutations are sometimes beneficial.	Sickle cell anemia is caused by a mutation in the gene that tells your body to make the iron-rich compound that makes blood red and enables red blood cells to carry oxygen from your lungs throughout your body (hemoglobin).
  can you get pregnant naturally if you have pcos?	Polycystic ovarian syndrome (PCOS) is one of the most common causes of female infertility, affecting an estimated 5 million women. 1 But you can get pregnant with PCOS. There are a number of effective fertility treatments available, from Clomid to gonadotropins to IVF.	One in every 10 women in India has polycystic ovary syndrome (PCOS), a common endocrinal system disorder among women of reproductive age, according to a study by PCOS Society. And out of every 10 women diagnosed with PCOS, six are teenage girls. PCOS was described as early as 1935.

• Loss: pylate.losses.contrastive.Contrastive

Evaluation Dataset

Unnamed Dataset

• Size: 5,000 evaluation samples
• Columns: question, answer, and negative_1
• Approximate statistics based on the first 1000 samples:

  	question	answer	negative_1
  type	string	string	string
  details	min: 9 tokens mean: 13.11 tokens max: 22 tokens	min: 16 tokens mean: 31.72 tokens max: 32 tokens	min: 14 tokens mean: 31.37 tokens max: 32 tokens

• Samples:

  question	answer	negative_1
  what are gharial related to?	What is a gharial? Gharials, sometimes called gavials, are a type of Asian crocodilian distinguished by their long, thin snouts. Crocodilians are a group of reptiles that includes crocodiles, alligators, caimans, and more.	false Gharials live in South America. true Gharials are an endangered species.
  what is electron transport chain and chemiosmosis?	The electron transport chain consists of a series of electron carriers that eventually transfer electrons from NADH and FADH2 to oxygen. The chemiosmotic theory states that the transfer of electrons down an electron transport system through a series of oxidation-reduction reactions releases energy.	The electron transport chain is a series of proteins and organic molecules found in the inner membrane of the mitochondria. ... Together, the electron transport chain and chemiosmosis make up oxidative phosphorylation.
  how to transfer pictures from icloud to usb?	["Manually download all the files from Apple's iCloud website to a folder on your PC and then copy/paste or move them to your USB drive.", "Download iCloud for Windows and find the iCloud folder in your File Explorer. Then, copy the photos from your PCs' iCloud folder and paste them to your USB Drive."]	["Manually download all the files from Apple's iCloud website to a folder on your PC and then copy/paste or move them to your USB drive.", "Download iCloud for Windows and find the iCloud folder in your File Explorer. Then, copy the photos from your PCs' iCloud folder and paste them to your USB Drive."]

• Loss: pylate.losses.contrastive.Contrastive

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 128
• per_device_eval_batch_size: 128
• learning_rate: 3e-06
• num_train_epochs: 1
• warmup_ratio: 0.1
• seed: 12
• bf16: True
• dataloader_num_workers: 12
• load_best_model_at_end: 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: 128
• per_device_eval_batch_size: 128
• 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: 3e-06
• 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: 0.1
• 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: 12
• 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: 12
• 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
• 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

Click to expand

Epoch	Step	Training Loss	Validation Loss	accuracy
0	0	-	-	0.4482
0.0000	1	19.1821	-	-
0.0068	200	14.4411	-	-
0.0135	400	7.9989	-	-
0.0203	600	5.6695	-	-
0.0270	800	4.0985	-	-
0.0338	1000	2.121	-	-
0.0406	1200	1.3478	-	-
0.0473	1400	1.0677	-	-
0.0541	1600	0.9243	-	-
0.0608	1800	0.8415	-	-
0.0676	2000	0.7807	-	-
0.0744	2200	0.7055	-	-
0.0811	2400	0.6782	-	-
0.0879	2600	0.6443	-	-
0.0946	2800	0.6028	-	-
0.1014	3000	0.5886	-	-
0.1082	3200	0.5484	-	-
0.1149	3400	0.5236	-	-
0.1217	3600	0.5049	-	-
0.1284	3800	0.4855	-	-
0.1352	4000	0.4565	-	-
0.1420	4200	0.4545	-	-
0.1487	4400	0.4313	-	-
0.1555	4600	0.4221	-	-
0.1622	4800	0.4102	-	-
0.1690	5000	0.4086	-	-
0.1758	5200	0.3935	-	-
0.1825	5400	0.3808	-	-
0.1893	5600	0.3768	-	-
0.1960	5800	0.3663	-	-
0.2028	6000	0.3625	-	-
0.2096	6200	0.3587	-	-
0.2163	6400	0.3429	-	-
0.2231	6600	0.3414	-	-
0.2298	6800	0.3359	-	-
0.2366	7000	0.322	-	-
0.2434	7200	0.3192	-	-
0.2501	7400	0.317	-	-
0.2569	7600	0.3137	-	-
0.2636	7800	0.3073	-	-
0.2704	8000	0.3103	-	-
0.2771	8200	0.3058	-	-
0.2839	8400	0.2998	-	-
0.2907	8600	0.294	-	-
0.2974	8800	0.2902	-	-
0.3042	9000	0.2907	-	-
0.3109	9200	0.2858	-	-
0.3177	9400	0.2802	-	-
0.3245	9600	0.2788	-	-
0.3312	9800	0.2725	-	-
0.3380	10000	0.2765	-	-
0.3447	10200	0.2722	-	-
0.3515	10400	0.2603	-	-
0.3583	10600	0.2712	-	-
0.3650	10800	0.2609	-	-
0.3718	11000	0.2637	-	-
0.3785	11200	0.2562	-	-
0.3853	11400	0.2511	-	-
0.3921	11600	0.2552	-	-
0.3988	11800	0.2494	-	-
0.4056	12000	0.2557	-	-
0.4123	12200	0.2459	-	-
0.4191	12400	0.2442	-	-
0.4259	12600	0.2471	-	-
0.4326	12800	0.2499	-	-
0.4394	13000	0.2417	-	-
0.4461	13200	0.246	-	-
0.4529	13400	0.2423	-	-
0.4597	13600	0.2361	-	-
0.4664	13800	0.2395	-	-
0.4732	14000	0.2347	-	-
0.4799	14200	0.2433	-	-
0.4867	14400	0.2353	-	-
0.4935	14600	0.2316	-	-
0.5002	14800	0.2305	-	-
0.5070	15000	0.2321	-	-
0.5137	15200	0.2322	-	-
0.5205	15400	0.2292	-	-
0.5273	15600	0.2204	-	-
0.5340	15800	0.2248	-	-
0.5408	16000	0.2223	-	-
0.5475	16200	0.2226	-	-
0.5543	16400	0.2245	-	-
0.5611	16600	0.222	-	-
0.5678	16800	0.2133	-	-
0.5746	17000	0.2222	-	-
0.5813	17200	0.2147	-	-
0.5881	17400	0.2192	-	-
0.5949	17600	0.2109	-	-
0.6016	17800	0.2164	-	-
0.6084	18000	0.2145	-	-
0.6151	18200	0.2107	-	-
0.6219	18400	0.2172	-	-
0.6287	18600	0.2132	-	-
0.6354	18800	0.2049	-	-
0.6422	19000	0.2125	-	-
0.6489	19200	0.211	-	-
0.6557	19400	0.2141	-	-
0.6625	19600	0.2073	-	-
0.6692	19800	0.2095	-	-
0.676	20000	0.2082	-	-
0	0	-	-	0.4942
0.676	20000	-	0.9978	-
0.6827	20200	0.2118	-	-
0.6895	20400	0.208	-	-
0.6963	20600	0.2092	-	-
0.7030	20800	0.2138	-	-
0.7098	21000	0.1991	-	-
0.7165	21200	0.2027	-	-
0.7233	21400	0.204	-	-
0.7301	21600	0.2065	-	-
0.7368	21800	0.1986	-	-
0.7436	22000	0.2008	-	-
0.7503	22200	0.201	-	-
0.7571	22400	0.2042	-	-
0.7638	22600	0.2004	-	-
0.7706	22800	0.1988	-	-
0.7774	23000	0.1997	-	-
0.7841	23200	0.2066	-	-
0.7909	23400	0.2012	-	-
0.7976	23600	0.1976	-	-
0.8044	23800	0.1975	-	-
0.8112	24000	0.1969	-	-
0.8179	24200	0.202	-	-
0.8247	24400	0.1978	-	-
0.8314	24600	0.2033	-	-
0.8382	24800	0.1964	-	-
0.8450	25000	0.2046	-	-
0.8517	25200	0.2002	-	-
0.8585	25400	0.1987	-	-
0.8652	25600	0.1945	-	-
0.8720	25800	0.1973	-	-
0.8788	26000	0.197	-	-
0.8855	26200	0.194	-	-
0.8923	26400	0.1965	-	-
0.8990	26600	0.2025	-	-
0.9058	26800	0.1913	-	-
0.9126	27000	0.1964	-	-
0.9193	27200	0.1949	-	-
0.9261	27400	0.194	-	-
0.9328	27600	0.1964	-	-
0.9396	27800	0.1954	-	-
0.9464	28000	0.1924	-	-
0.9531	28200	0.1948	-	-
0.9599	28400	0.1963	-	-
0.9666	28600	0.1929	-	-
0.9734	28800	0.1993	-	-
0.9802	29000	0.1926	-	-
0.9869	29200	0.1874	-	-
0.9937	29400	0.1976	-	-

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.0
• Sentence Transformers: 4.0.1
• PyLate: 1.1.7
• Transformers: 4.48.2
• PyTorch: 2.6.0+cu124
• Accelerate: 1.6.0
• Datasets: 3.5.0
• 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}
}