This is a sentence-transformers model finetuned from Mihaiii/gte-micro-v4. It maps sentences & paragraphs to a 384-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: Mihaiii/gte-micro-v4
Maximum Sequence Length: 256 tokens
Output Dimensionality: 384 dimensions
Similarity Function: Cosine Similarity
Supported Modality: Text

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'BertModel'})
  (1): Pooling({'embedding_dimension': 384, 'pooling_mode': 'mean', 'include_prompt': True})
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("philipp-zettl/gte-micro-v4-mtg-v2")
# Run inference
sentences = [
    'creature — drake',
    'Title: Fighting Drake\nCost: {2}{U}{U}\nColors: U\nType: Creature — Drake\nDesc: Flying',
    "Title: Big Game Hunter\nCost: {1}{B}{B}\nColors: B\nType: Creature — Human Rebel Assassin\nDesc: When this creature enters, destroy target creature with power 4 or greater. It can't be regenerated.\nMadness {B} (If you discard this card, discard it into exile. When you do, cast it for its madness cost or put it into your graveyard.)",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[ 1.0000,  0.6089,  0.0016],
#         [ 0.6089,  1.0000, -0.1129],
#         [ 0.0016, -0.1129,  1.0000]])

Training Details

Training Dataset

Unnamed Dataset

Size: 2,132,917 training samples
Columns: sentence_0 and sentence_1
Approximate statistics based on the first 100 samples:
sentence_0 sentence_1
type string string
modality text text
details
min: 3 tokens
mean: 10.1 tokens
max: 41 tokens
min: 24 tokens
mean: 62.76 tokens
max: 146 tokens

Samples:

sentence_0	sentence_1
`magnifying glass`	`Title: Magnifying Glass[object Object]Cost: {3}[object Object]Type: Artifact[object Object]Desc: {T}: Add {C}.[object Object]{4}, {T}: Investigate. (Create a Clue token. It's an artifact with "{2}, Sacrifice this token: Draw a card.")`
`sorcery`	`Title: Graveyard Shift[object Object]Cost: {4}{B}[object Object]Colors: B[object Object]Type: Sorcery[object Object]Desc: This spell has flash as long as there are five or more mana values among cards in your graveyard.[object Object]Return target creature card from your graveyard to the battlefield.`
`beacon of unrest`	`Title: Beacon of Unrest[object Object]Cost: {3}{B}{B}[object Object]Colors: B[object Object]Type: Sorcery[object Object]Desc: Put target artifact or creature card from a graveyard onto the battlefield under your control. Shuffle Beacon of Unrest into its owner's library.`

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim", 4 "gather_across_devices": false, 5 "directions": [ 6 "query_to_doc" 7 ], 8 "partition_mode": "joint", 9 "hardness_mode": null, 10 "hardness_strength": 0.0 11}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 96
fp16: True
per_device_eval_batch_size: 96
multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

per_device_train_batch_size: 96
num_train_epochs: 3
max_steps: -1
learning_rate: 5e-05
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_steps: 0
optim: adamw_torch_fused
optim_args: None
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
optim_target_modules: None
gradient_accumulation_steps: 1
average_tokens_across_devices: True
max_grad_norm: 1
label_smoothing_factor: 0.0
bf16: False
fp16: True
bf16_full_eval: False
fp16_full_eval: False
tf32: None
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
use_liger_kernel: False
liger_kernel_config: None
use_cache: False
neftune_noise_alpha: None
torch_empty_cache_steps: None
auto_find_batch_size: False
log_on_each_node: True
logging_nan_inf_filter: True
include_num_input_tokens_seen: no
log_level: passive
log_level_replica: warning
disable_tqdm: False
project: huggingface
trackio_space_id: None
trackio_bucket_id: None
trackio_static_space_id: None
per_device_eval_batch_size: 96
prediction_loss_only: True
eval_on_start: False
eval_do_concat_batches: True
eval_use_gather_object: False
eval_accumulation_steps: None
include_for_metrics: []
batch_eval_metrics: False
save_only_model: False
save_on_each_node: False
enable_jit_checkpoint: False
push_to_hub: False
hub_private_repo: None
hub_model_id: None
hub_strategy: every_save
hub_always_push: False
hub_revision: None
load_best_model_at_end: False
ignore_data_skip: False
restore_callback_states_from_checkpoint: False
full_determinism: False
seed: 42
data_seed: None
use_cpu: False
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
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_pin_memory: True
dataloader_persistent_workers: False
dataloader_prefetch_factor: None
remove_unused_columns: True
label_names: None
train_sampling_strategy: random
length_column_name: length
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
ddp_static_graph: None
ddp_backend: None
ddp_timeout: 1800
fsdp: None
fsdp_config: None
deepspeed: None
debug: []
skip_memory_metrics: True
do_predict: False
resume_from_checkpoint: None
warmup_ratio: None
local_rank: -1
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: round_robin
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Click to expand

Epoch	Step	Training Loss
0.0225	500	1.8730
0.0450	1000	0.5353
0.0675	1500	0.4567
0.0900	2000	0.4285
0.1125	2500	0.4066
0.1350	3000	0.3939
0.1575	3500	0.3833
0.1800	4000	0.3718
0.2025	4500	0.3707
0.2250	5000	0.3703
0.2475	5500	0.3630
0.2701	6000	0.3537
0.2926	6500	0.3575
0.3151	7000	0.3548
0.3376	7500	0.3565
0.3601	8000	0.3476
0.3826	8500	0.3422
0.4051	9000	0.3423
0.4276	9500	0.3414
0.4501	10000	0.3410
0.4726	10500	0.3480
0.4951	11000	0.3406
0.5176	11500	0.3293
0.5401	12000	0.3354
0.5626	12500	0.3337
0.5851	13000	0.3371
0.6076	13500	0.3376
0.6301	14000	0.3347
0.6526	14500	0.3297
0.6751	15000	0.3326
0.6976	15500	0.3297
0.7201	16000	0.3213
0.7426	16500	0.3270
0.7651	17000	0.3291
0.7876	17500	0.3291
0.8102	18000	0.3235
0.8327	18500	0.3293
0.8552	19000	0.3212
0.8777	19500	0.3261
0.9002	20000	0.3191
0.9227	20500	0.3176
0.9452	21000	0.3145
0.9677	21500	0.3267
0.9902	22000	0.3197
1.0127	22500	0.3160
1.0352	23000	0.3230
1.0577	23500	0.3182
1.0802	24000	0.3221
1.1027	24500	0.3169
1.1252	25000	0.3125
1.1477	25500	0.3135
1.1702	26000	0.3149
1.1927	26500	0.3182
1.2152	27000	0.3246
1.2377	27500	0.3161
1.2602	28000	0.3173
1.2827	28500	0.3137
1.3052	29000	0.3147
1.3278	29500	0.3110
1.3503	30000	0.3129
1.3728	30500	0.3111
1.3953	31000	0.3132
1.4178	31500	0.3176
1.4403	32000	0.3092
1.4628	32500	0.3177
1.4853	33000	0.3031
1.5078	33500	0.3126
1.5303	34000	0.3144
1.5528	34500	0.3061
1.5753	35000	0.3120
1.5978	35500	0.3083
1.6203	36000	0.3087
1.6428	36500	0.3131
1.6653	37000	0.3108
1.6878	37500	0.3131
1.7103	38000	0.3092
1.7328	38500	0.3099
1.7553	39000	0.3104
1.7778	39500	0.3049
1.8003	40000	0.3061
1.8228	40500	0.3105
1.8454	41000	0.3031
1.8679	41500	0.3008
1.8904	42000	0.3108
1.9129	42500	0.3071
1.9354	43000	0.3067
1.9579	43500	0.3077
1.9804	44000	0.3094
2.0029	44500	0.3031
2.0254	45000	0.3045
2.0479	45500	0.3056
2.0704	46000	0.3075
2.0929	46500	0.3054
2.1154	47000	0.2982
2.1379	47500	0.3003
2.1604	48000	0.3077
2.1829	48500	0.3012
2.2054	49000	0.3060
2.2279	49500	0.2995
2.2504	50000	0.3060
2.2729	50500	0.3098
2.2954	51000	0.3002
2.3179	51500	0.3004
2.3404	52000	0.3095
2.3629	52500	0.3028
2.3855	53000	0.3040
2.4080	53500	0.3056
2.4305	54000	0.3066
2.4530	54500	0.3013
2.4755	55000	0.3074
2.4980	55500	0.3054
2.5205	56000	0.3053
2.5430	56500	0.3001
2.5655	57000	0.2987
2.5880	57500	0.3104
2.6105	58000	0.3044
2.6330	58500	0.3015
2.6555	59000	0.3076
2.6780	59500	0.3012
2.7005	60000	0.3022
2.7230	60500	0.3084
2.7455	61000	0.3004
2.7680	61500	0.3056
2.7905	62000	0.3057
2.8130	62500	0.2993
2.8355	63000	0.3010
2.8580	63500	0.3019
2.8805	64000	0.2993
2.9031	64500	0.3040
2.9256	65000	0.3003
2.9481	65500	0.3000
2.9706	66000	0.2982
2.9931	66500	0.3003

Training Time

Training: 1.7 hours

Framework Versions

Python: 3.13.11
Sentence Transformers: 5.6.0
Transformers: 5.14.1
PyTorch: 2.13.0+cu130
Accelerate: 1.14.0
Datasets: 5.0.0
Tokenizers: 0.22.2

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",
}

MultipleNegativesRankingLoss

bibtex

@misc{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}

	sentence_0	sentence_1
type	string	string
modality	text	text
details	min: 3 tokens mean: 10.1 tokens max: 41 tokens	min: 24 tokens mean: 62.76 tokens max: 146 tokens