You can deploy this model and start earning money today!

Views

No views yet

ColBERT MUVERA Small

Name: colbert-muvera-small
Rating: 2.5 (306 reviews)

This is a PyLate model finetuned from bclavie/mini-base on the msmarco-en-bge-gemma-unnormalized dataset. It maps sentences & paragraphs to sequences of 128-dimensional dense vectors and can be used for semantic textual similarity using the MaxSim operator.

This model is trained with un-normalized scores, making it compatible with MUVERA fixed-dimensional encoding.

Usage (txtai)

This model can be used to build embeddings databases with txtai for semantic search and/or as a knowledge source for retrieval augmented generation (RAG).

Note: txtai 9.0+ is required for late interaction model support

python

import txtai

embeddings = txtai.Embeddings(
  path="neuml/colbert-muvera-small",
  content=True
)
embeddings.index(documents())

# Run a query
embeddings.search("query to run")

Late interaction models excel as reranker pipelines.

python

from txtai.pipeline import Reranker, Similarity

similarity = Similarity(path="neuml/colbert-muvera-small", lateencode=True)
ranker = Reranker(embeddings, similarity)
ranker("query to run")

Usage (PyLate)

Alternatively, the model can be loaded with PyLate.

python

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="neuml/colbert-muvera-small",
)

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,
)

Full Model Architecture

ColBERT(
  (0): Transformer({'max_seq_length': 299, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Dense({'in_features': 384, 'out_features': 128, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity'})
)

Evaluation

BEIR Subset

The following table shows a subset of BEIR scored with the txtai benchmarks script.

Scores reported are ndcg@10 and grouped into the following three categories.

FULL multi-vector maxsim

Model	Parameters	ArguAna	NFCorpus	SciFact	Average
AnswerAI ColBERT Small v1	33M	0.4440	0.3649	0.7423	0.5171
ColBERT v2	110M	0.4595	0.3165	0.6456	0.4739
ColBERT MUVERA Micro	4M	0.3947	0.3235	0.6676	0.4619
ColBERT MUVERA Small	33M	0.4455	0.3502	0.7145	0.5034
GTE ModernColBERT v1	149M	0.4946	0.3717	0.7529	0.5397

MUVERA encoding + maxsim re-ranking of the top 100 results per MUVERA paper

Model	Parameters	ArguAna	NFCorpus	SciFact	Average
AnswerAI ColBERT Small v1	33M	0.0317	0.1135	0.0836	0.0763
ColBERT v2	110M	0.4562	0.3025	0.6278	0.4622
ColBERT MUVERA Micro	4M	0.3849	0.3095	0.6464	0.4469
ColBERT MUVERA Small	33M	0.4451	0.3537	0.7148	0.5045
GTE ModernColBERT v1	149M	0.0265	0.1052	0.0556	0.0624

MUVERA encoding only

Model	Parameters	ArguAna	NFCorpus	SciFact	Average
AnswerAI ColBERT Small v1	33M	0.0024	0.0201	0.0047	0.0091
ColBERT v2	110M	0.3463	0.2356	0.5002	0.3607
ColBERT MUVERA Micro	4M	0.2795	0.2348	0.4875	0.3339
ColBERT MUVERA Small	33M	0.3850	0.2928	0.6357	0.4378
GTE ModernColBERT v1	149M	0.0003	0.0203	0.0013	0.0073

Note: The scores reported don't match scores reported in the respective papers due to different default settings in the txtai benchmark scripts.

As noted earlier, models trained with min-max score normalization don't perform well with MUVERA encoding. See this GitHub Issue for more.

In reviewing the scores, this model is surprisingly and unreasonably competitive with the original ColBERT v2 model at only 3% of the size!

Nano BEIR

Dataset: NanoBEIR_mean
Evaluated with pylate.evaluation.nano_beir_evaluator.NanoBEIREvaluator

Metric	Value
MaxSim_accuracy@1	0.5995
MaxSim_accuracy@3	0.7522
MaxSim_accuracy@5	0.7984
MaxSim_accuracy@10	0.8646
MaxSim_precision@1	0.5995
MaxSim_precision@3	0.3589
MaxSim_precision@5	0.2732
MaxSim_precision@10	0.1885
MaxSim_recall@1	0.3574
MaxSim_recall@3	0.5047
MaxSim_recall@5	0.559
MaxSim_recall@10	0.6365
MaxSim_ndcg@10	0.617
MaxSim_mrr@10	0.6868
MaxSim_map@100	0.5394

Training Details

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
gradient_accumulation_steps: 4
learning_rate: 3e-06
num_train_epochs: 1
warmup_ratio: 0.05
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: 4
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: 0.05
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: 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
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

Framework Versions

Python: 3.10.18
Sentence Transformers: 4.0.2
PyLate: 1.3.0
Transformers: 4.52.3
PyTorch: 2.8.0+cu128
Accelerate: 1.10.1
Datasets: 4.0.0
Tokenizers: 0.21.4

Citation

BibTeX

Sentence Transformers

bibtex

@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

bibtex

@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}
}

NeuML/colbert-muvera-small

ColBERT MUVERA Small

Usage (txtai)

Usage (PyLate)

Full Model Architecture

Evaluation

BEIR Subset

FULL multi-vector maxsim

MUVERA encoding + maxsim re-ranking of the top 100 results per MUVERA paper

MUVERA encoding only

Nano BEIR

Training Details

Training Hyperparameters

Non-Default Hyperparameters

All Hyperparameters

Framework Versions

Citation

BibTeX

Sentence Transformers

PyLate

NeuML/
colbert-muvera-small