metadata

tags:
  - ColBERT
  - PyLate
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:9461702
  - loss:Contrastive
base_model: answerdotai/ModernBERT-base
pipeline_tag: sentence-similarity
library_name: PyLate
metrics:
  - accuracy
model-index:
  - name: PyLate model based on answerdotai/ModernBERT-base
    results:
      - task:
          type: col-berttriplet
          name: Col BERTTriplet
        dataset:
          name: Unknown
          type: unknown
        metrics:
          - type: accuracy
            value: 0.5281999707221985
            name: Accuracy

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-5-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-5-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	0.5282

Training Details

Training Dataset

Unnamed Dataset

Size: 9,461,702 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: 13.05 tokens max: 19 tokens	min: 25 tokens mean: 31.88 tokens max: 32 tokens	min: 16 tokens mean: 31.67 tokens max: 32 tokens

Samples:

question	answer	negative
`what is the maximum income you can make while collecting social security?`	`The Social Security earnings limit is $1,470 per month or $17,640 per year in 2019 for someone age 65 or younger. If you earn more than this amount, you can expect to have $1 withheld from your Social Security benefit for every $2 earned above the limit.`	`Once you reach FRA, there is no cap on how much you can earn and still receive your full Social Security benefit. The earnings limits are adjusted annually for national wage trends. In 2020, you lose $1 in benefits for every $2 earned over $18,240.`
`what is the maximum income you can make while collecting social security?`	`The Social Security earnings limit is $1,470 per month or $17,640 per year in 2019 for someone age 65 or younger. If you earn more than this amount, you can expect to have $1 withheld from your Social Security benefit for every $2 earned above the limit.`	`You can get Social Security retirement or survivors benefits and work at the same time. However, there is a limit to how much you can earn and still receive full benefits. If you are younger than full retirement age and earn more than the yearly earnings limit, we may reduce your benefit amount.`
`what is the maximum income you can make while collecting social security?`	`The Social Security earnings limit is $1,470 per month or $17,640 per year in 2019 for someone age 65 or younger. If you earn more than this amount, you can expect to have $1 withheld from your Social Security benefit for every $2 earned above the limit.`	`If you haven't yet reached full retirement age, you can earn up to $17,640 in income each year without any reduction in benefits. But for each $2 you earn above this limit, the Social Security Administration deducts $1 from your benefit payments. Under full retirement age for part of a year.`

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: 12.93 tokens max: 22 tokens	min: 17 tokens mean: 31.7 tokens max: 32 tokens	min: 14 tokens mean: 31.4 tokens max: 32 tokens

Samples:

question	answer	negative_1
`are bird scooters in nyc?`	`New York State is on the verge of embracing electric scooters and bicycles in a victory for tech leaders and delivery workers who have fought for months to make the speedy devices legal. ... There is just one catch — scooter rental companies like Bird and Lime cannot operate in Manhattan.`	`New York State is on the verge of embracing electric scooters and bicycles in a victory for tech leaders and delivery workers who have fought for months to make the speedy devices legal. ... There is just one catch — scooter rental companies like Bird and Lime cannot operate in Manhattan.`
`can you go into a bar if you're 18?`	`You can enter a bar at 18 but you cannot consume alcoholic beverages until you are 21. ... Some states will make some exceptions for a parent allowing you to drink from their alcoholic beverage, but it is best to not do that in public places if you are under the age of 21 in the USA.`	`1. Re: How old do you have to be to enter a club, bar, pub? Generally 18 is fine, though some upscale bars may extend that to 21. Pubs don't have an age limit to enter, but you may get carded if ordering alcohol.`
`how are blood pressure numbers written and recorded?`	`Blood pressure is recorded as two numbers and written as a ratio: the top number, called the systolic pressure, is the pressure as the heart beats. The bottom number, called the diastolic pressure, is the measurement as the heart relaxes between beats.`	`Blood pressure is recorded with 2 numbers. The systolic pressure (higher number) is the force at which your heart pumps blood around your body. The diastolic pressure (lower number) is the resistance to the blood flow in the blood vessels.`

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	-	-	0.4558
0.0000	1	17.2657	-	-
0.0027	200	17.0631	-	-
0.0054	400	11.2015	-	-
0.0081	600	7.8228	-	-
0.0108	800	6.0774	-	-
0.0135	1000	5.3122	-	-
0.0162	1200	4.3348	-	-
0.0189	1400	2.6982	-	-
0.0216	1600	1.7959	-	-
0.0244	1800	1.3555	-	-
0.0271	2000	1.1443	-	-
0.0298	2200	1.0092	-	-
0.0325	2400	0.9274	-	-
0.0352	2600	0.8485	-	-
0.0379	2800	0.7953	-	-
0.0406	3000	0.7541	-	-
0.0433	3200	0.7302	-	-
0.0460	3400	0.6836	-	-
0.0487	3600	0.6546	-	-
0.0514	3800	0.6219	-	-
0.0541	4000	0.6116	-	-
0.0568	4200	0.5813	-	-
0.0595	4400	0.5499	-	-
0.0622	4600	0.5334	-	-
0.0649	4800	0.5276	-	-
0.0676	5000	0.4969	-	-
0.0703	5200	0.4789	-	-
0.0731	5400	0.4709	-	-
0.0758	5600	0.4598	-	-
0.0785	5800	0.4465	-	-
0.0812	6000	0.4333	-	-
0.0839	6200	0.4258	-	-
0.0866	6400	0.4056	-	-
0.0893	6600	0.3855	-	-
0.0920	6800	0.3855	-	-
0.0947	7000	0.3761	-	-
0.0974	7200	0.369	-	-
0.1001	7400	0.3531	-	-
0.1028	7600	0.3549	-	-
0.1055	7800	0.3342	-	-
0.1082	8000	0.3289	-	-
0.1109	8200	0.3231	-	-
0.1136	8400	0.3197	-	-
0.1163	8600	0.3066	-	-
0.1190	8800	0.309	-	-
0.1218	9000	0.2953	-	-
0.1245	9200	0.284	-	-
0.1272	9400	0.2841	-	-
0.1299	9600	0.2842	-	-
0.1326	9800	0.2764	-	-
0.1353	10000	0.2737	-	-
0.1380	10200	0.2673	-	-
0.1407	10400	0.2556	-	-
0.1434	10600	0.2613	-	-
0.1461	10800	0.2559	-	-
0.1488	11000	0.2557	-	-
0.1515	11200	0.2496	-	-
0.1542	11400	0.2411	-	-
0.1569	11600	0.2446	-	-
0.1596	11800	0.2384	-	-
0.1623	12000	0.2267	-	-
0.1650	12200	0.2401	-	-
0.1677	12400	0.2338	-	-
0.1705	12600	0.2306	-	-
0.1732	12800	0.2259	-	-
0.1759	13000	0.2278	-	-
0.1786	13200	0.2172	-	-
0.1813	13400	0.2254	-	-
0.1840	13600	0.2232	-	-
0.1867	13800	0.2106	-	-
0.1894	14000	0.2187	-	-
0.1921	14200	0.2147	-	-
0.1948	14400	0.2043	-	-
0.1975	14600	0.2017	-	-
0.2002	14800	0.2071	-	-
0.2029	15000	0.2016	-	-
0.2056	15200	0.1994	-	-
0.2083	15400	0.2018	-	-
0.2110	15600	0.1946	-	-
0.2137	15800	0.1911	-	-
0.2165	16000	0.1828	-	-
0.2192	16200	0.1878	-	-
0.2219	16400	0.1839	-	-
0.2246	16600	0.1939	-	-
0.2273	16800	0.1842	-	-
0.2300	17000	0.1912	-	-
0.2327	17200	0.1851	-	-
0.2354	17400	0.1863	-	-
0.2381	17600	0.1829	-	-
0.2408	17800	0.1829	-	-
0.2435	18000	0.177	-	-
0.2462	18200	0.1768	-	-
0.2489	18400	0.1819	-	-
0.2516	18600	0.1778	-	-
0.2543	18800	0.1803	-	-
0.2570	19000	0.1758	-	-
0.2597	19200	0.1736	-	-
0.2624	19400	0.1759	-	-
0.2652	19600	0.1751	-	-
0.2679	19800	0.1739	-	-
0.2706	20000	0.1677	-	-
0	0	-	-	0.5018
0.2706	20000	-	1.0521	-
0.2733	20200	0.1681	-	-
0.2760	20400	0.1672	-	-
0.2787	20600	0.1695	-	-
0.2814	20800	0.1696	-	-
0.2841	21000	0.1662	-	-
0.2868	21200	0.1612	-	-
0.2895	21400	0.1678	-	-
0.2922	21600	0.1617	-	-
0.2949	21800	0.1635	-	-
0.2976	22000	0.1622	-	-
0.3003	22200	0.1647	-	-
0.3030	22400	0.1634	-	-
0.3057	22600	0.1597	-	-
0.3084	22800	0.1616	-	-
0.3111	23000	0.1538	-	-
0.3139	23200	0.1601	-	-
0.3166	23400	0.1583	-	-
0.3193	23600	0.161	-	-
0.3220	23800	0.1539	-	-
0.3247	24000	0.1602	-	-
0.3274	24200	0.1493	-	-
0.3301	24400	0.1536	-	-
0.3328	24600	0.1572	-	-
0.3355	24800	0.1577	-	-
0.3382	25000	0.1508	-	-
0.3409	25200	0.1514	-	-
0.3436	25400	0.1506	-	-
0.3463	25600	0.1544	-	-
0.3490	25800	0.1574	-	-
0.3517	26000	0.1507	-	-
0.3544	26200	0.1462	-	-
0.3571	26400	0.1527	-	-
0.3598	26600	0.1474	-	-
0.3626	26800	0.1516	-	-
0.3653	27000	0.1447	-	-
0.3680	27200	0.1484	-	-
0.3707	27400	0.1454	-	-
0.3734	27600	0.1467	-	-
0.3761	27800	0.1517	-	-
0.3788	28000	0.1505	-	-
0.3815	28200	0.1395	-	-
0.3842	28400	0.145	-	-
0.3869	28600	0.143	-	-
0.3896	28800	0.1417	-	-
0.3923	29000	0.142	-	-
0.3950	29200	0.1401	-	-
0.3977	29400	0.1399	-	-
0.4004	29600	0.1437	-	-
0.4031	29800	0.1399	-	-
0.4058	30000	0.1394	-	-
0.4085	30200	0.1373	-	-
0.4113	30400	0.1388	-	-
0.4140	30600	0.1384	-	-
0.4167	30800	0.1434	-	-
0.4194	31000	0.1398	-	-
0.4221	31200	0.1476	-	-
0.4248	31400	0.1387	-	-
0.4275	31600	0.1346	-	-
0.4302	31800	0.137	-	-
0.4329	32000	0.135	-	-
0.4356	32200	0.1363	-	-
0.4383	32400	0.1336	-	-
0.4410	32600	0.1323	-	-
0.4437	32800	0.1371	-	-
0.4464	33000	0.1305	-	-
0.4491	33200	0.1315	-	-
0.4518	33400	0.1366	-	-
0.4545	33600	0.1336	-	-
0.4573	33800	0.1349	-	-
0.4600	34000	0.1338	-	-
0.4627	34200	0.1388	-	-
0.4654	34400	0.1312	-	-
0.4681	34600	0.1299	-	-
0.4708	34800	0.1325	-	-
0.4735	35000	0.1277	-	-
0.4762	35200	0.132	-	-
0.4789	35400	0.1322	-	-
0.4816	35600	0.1286	-	-
0.4843	35800	0.1322	-	-
0.4870	36000	0.1342	-	-
0.4897	36200	0.1306	-	-
0.4924	36400	0.1339	-	-
0.4951	36600	0.1327	-	-
0.4978	36800	0.129	-	-
0.5005	37000	0.1301	-	-
0.5032	37200	0.1266	-	-
0.5060	37400	0.1295	-	-
0.5087	37600	0.1263	-	-
0.5114	37800	0.1321	-	-
0.5141	38000	0.1213	-	-
0.5168	38200	0.1253	-	-
0.5195	38400	0.13	-	-
0.5222	38600	0.1234	-	-
0.5249	38800	0.1259	-	-
0.5276	39000	0.1303	-	-
0.5303	39200	0.1268	-	-
0.5330	39400	0.1229	-	-
0.5357	39600	0.1291	-	-
0.5384	39800	0.1257	-	-
0.5411	40000	0.1249	-	-
0	0	-	-	0.5130
0.5411	40000	-	1.0519	-
0.5438	40200	0.1259	-	-
0.5465	40400	0.1253	-	-
0.5492	40600	0.1229	-	-
0.5519	40800	0.1296	-	-
0.5547	41000	0.1222	-	-
0.5574	41200	0.1216	-	-
0.5601	41400	0.1226	-	-
0.5628	41600	0.1256	-	-
0.5655	41800	0.1198	-	-
0.5682	42000	0.1275	-	-
0.5709	42200	0.1222	-	-
0.5736	42400	0.1229	-	-
0.5763	42600	0.123	-	-
0.5790	42800	0.1162	-	-
0.5817	43000	0.1234	-	-
0.5844	43200	0.1253	-	-
0.5871	43400	0.1221	-	-
0.5898	43600	0.1223	-	-
0.5925	43800	0.1244	-	-
0.5952	44000	0.1254	-	-
0.5979	44200	0.1227	-	-
0.6006	44400	0.1168	-	-
0.6034	44600	0.1184	-	-
0.6061	44800	0.1191	-	-
0.6088	45000	0.1174	-	-
0.6115	45200	0.1103	-	-
0.6142	45400	0.1181	-	-
0.6169	45600	0.1192	-	-
0.6196	45800	0.1206	-	-
0.6223	46000	0.1196	-	-
0.625	46200	0.1199	-	-
0.6277	46400	0.1226	-	-
0.6304	46600	0.1174	-	-
0.6331	46800	0.118	-	-
0.6358	47000	0.1185	-	-
0.6385	47200	0.1193	-	-
0.6412	47400	0.1181	-	-
0.6439	47600	0.1228	-	-
0.6466	47800	0.1235	-	-
0.6494	48000	0.1191	-	-
0.6521	48200	0.1142	-	-
0.6548	48400	0.1166	-	-
0.6575	48600	0.1218	-	-
0.6602	48800	0.1189	-	-
0.6629	49000	0.1196	-	-
0.6656	49200	0.1153	-	-
0.6683	49400	0.1132	-	-
0.6710	49600	0.1191	-	-
0.6737	49800	0.1148	-	-
0.6764	50000	0.1087	-	-
0.6791	50200	0.1145	-	-
0.6818	50400	0.1175	-	-
0.6845	50600	0.1145	-	-
0.6872	50800	0.1175	-	-
0.6899	51000	0.1131	-	-
0.6926	51200	0.112	-	-
0.6953	51400	0.1165	-	-
0.6981	51600	0.124	-	-
0.7008	51800	0.1129	-	-
0.7035	52000	0.1111	-	-
0.7062	52200	0.1143	-	-
0.7089	52400	0.1118	-	-
0.7116	52600	0.116	-	-
0.7143	52800	0.1181	-	-
0.7170	53000	0.1145	-	-
0.7197	53200	0.1161	-	-
0.7224	53400	0.1124	-	-
0.7251	53600	0.1123	-	-
0.7278	53800	0.1115	-	-
0.7305	54000	0.1119	-	-
0.7332	54200	0.114	-	-
0.7359	54400	0.1145	-	-
0.7386	54600	0.1095	-	-
0.7413	54800	0.1199	-	-
0.7440	55000	0.1129	-	-
0.7468	55200	0.1147	-	-
0.7495	55400	0.1091	-	-
0.7522	55600	0.11	-	-
0.7549	55800	0.1061	-	-
0.7576	56000	0.1136	-	-
0.7603	56200	0.112	-	-
0.7630	56400	0.1116	-	-
0.7657	56600	0.1132	-	-
0.7684	56800	0.1067	-	-
0.7711	57000	0.1116	-	-
0.7738	57200	0.1119	-	-
0.7765	57400	0.1097	-	-
0.7792	57600	0.1095	-	-
0.7819	57800	0.1101	-	-
0.7846	58000	0.1121	-	-
0.7873	58200	0.1118	-	-
0.7900	58400	0.1152	-	-
0.7927	58600	0.1106	-	-
0.7955	58800	0.1106	-	-
0.7982	59000	0.1117	-	-
0.8009	59200	0.1089	-	-
0.8036	59400	0.1087	-	-
0.8063	59600	0.111	-	-
0.8090	59800	0.1095	-	-
0.8117	60000	0.1144	-	-
0	0	-	-	0.5282
0.8117	60000	-	1.0542	-
0.8144	60200	0.1134	-	-
0.8171	60400	0.1107	-	-
0.8198	60600	0.1102	-	-
0.8225	60800	0.1088	-	-
0.8252	61000	0.1123	-	-
0.8279	61200	0.1081	-	-
0.8306	61400	0.1097	-	-
0.8333	61600	0.1077	-	-
0.8360	61800	0.1069	-	-
0.8387	62000	0.109	-	-
0.8415	62200	0.1086	-	-
0.8442	62400	0.1144	-	-
0.8469	62600	0.107	-	-
0.8496	62800	0.1064	-	-
0.8523	63000	0.1077	-	-
0.8550	63200	0.1044	-	-
0.8577	63400	0.103	-	-
0.8604	63600	0.1106	-	-
0.8631	63800	0.1137	-	-
0.8658	64000	0.1109	-	-
0.8685	64200	0.112	-	-
0.8712	64400	0.1111	-	-
0.8739	64600	0.1073	-	-
0.8766	64800	0.1067	-	-
0.8793	65000	0.1084	-	-
0.8820	65200	0.1081	-	-
0.8847	65400	0.1096	-	-
0.8874	65600	0.1084	-	-
0.8902	65800	0.1014	-	-
0.8929	66000	0.1071	-	-
0.8956	66200	0.1043	-	-
0.8983	66400	0.1112	-	-
0.9010	66600	0.1089	-	-
0.9037	66800	0.1086	-	-
0.9064	67000	0.1025	-	-
0.9091	67200	0.1024	-	-
0.9118	67400	0.1101	-	-
0.9145	67600	0.1075	-	-
0.9172	67800	0.1059	-	-
0.9199	68000	0.1085	-	-
0.9226	68200	0.1036	-	-
0.9253	68400	0.1056	-	-
0.9280	68600	0.1071	-	-
0.9307	68800	0.1065	-	-
0.9334	69000	0.1117	-	-
0.9361	69200	0.1074	-	-
0.9389	69400	0.1021	-	-
0.9416	69600	0.1081	-	-
0.9443	69800	0.1071	-	-
0.9470	70000	0.1056	-	-
0.9497	70200	0.1108	-	-
0.9524	70400	0.1093	-	-
0.9551	70600	0.1065	-	-
0.9578	70800	0.1092	-	-
0.9605	71000	0.1081	-	-
0.9632	71200	0.1031	-	-
0.9659	71400	0.1075	-	-
0.9686	71600	0.1101	-	-
0.9713	71800	0.1063	-	-
0.9740	72000	0.1076	-	-
0.9767	72200	0.1039	-	-
0.9794	72400	0.1102	-	-
0.9821	72600	0.1085	-	-
0.9848	72800	0.1068	-	-
0.9876	73000	0.1062	-	-
0.9903	73200	0.1049	-	-
0.9930	73400	0.1132	-	-
0.9957	73600	0.1095	-	-
0.9984	73800	0.1072	-	-

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}
}