SentenceTransformer based on sentence-transformers/all-MiniLM-L6-v2

This is a sentence-transformers model finetuned from sentence-transformers/all-MiniLM-L6-v2. It maps sentences & paragraphs to a 384-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: sentence-transformers/all-MiniLM-L6-v2
Maximum Sequence Length: 256 tokens
Output Dimensionality: 384 dimensions
Similarity Function: Cosine Similarity

Model Sources

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

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 256, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Normalize()
)

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.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("tjohn327/scion-minilm-v2")
# Run inference
sentences = [
    "What role does the function 'target' play in modeling network topology?",
    'The document introduces a formal framework for verifying secure forwarding protocols within the SCION Internet architecture, specifically focusing on the ICING data plane protocol. It employs event systems as labeled transition systems, defining an event system E as a tuple comprising a set of states S, an initial state s0, a set of events E, and a transition relation e−→. The framework formalizes the concepts of reachability and invariants, establishing that a state property P is an invariant if the reachable states from the initial state are contained within P. The refinement of abstract event systems to concrete systems is articulated through mappings that preserve invariants. The document emphasizes parametrization, allowing models to incorporate assumptions on parameters, which are highlighted in gray. An abstract model is defined for a path-aware network architecture, excluding cryptographic elements, and is proven to satisfy path authorization and detectability. The network topology is modeled as a multigraph, with nodes representing Autonomous Systems (ASes) and edges representing links, characterized by a partial bijective function target that facilitates multiple links and forwarding policies. Paths in the network are defined as finite sequences of hop fields, encapsulating local routing information.',
    "The document chunk presents a testbed architecture for evaluating Secure In-Band Network Telemetry (ID-INT) within the SCION Internet Architecture, utilizing a Tofino 2 switch as the ID-INT enabled border routers for two SCION Autonomous Systems (ASes). The Dynamic Multi-Path Transport Protocol (DMTP) is adapted to send probe packets and retrieve telemetry data, focusing on instantaneous queue length at the egress of the border router. The experiment assesses DMTP's ability to adjust sending rates on Path 2 based on ID-INT telemetry, with initial path capacities set at 100 Mbps, later reduced to 75 Mbps. Results indicate that DMTP with ID-INT-enabled congestion control aligns sending rates more closely with available link capacity, achieving a 2% gain in goodput despite ID-INT overhead. The adaptation speed of DMTP using ID-INT telemetry is 35% faster than traditional congestion window-based control, demonstrating improved bandwidth utilization and congestion prevention. Related work includes a software implementation of the SCION reference border router in Go, and a hardware implementation on a NetFPGA SUME card capable of 10 Gbps throughput, highlighting the need for efficient high-bandwidth traffic handling in SCION.",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Information Retrieval

Dataset: val-ir-eval
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.3775
cosine_accuracy@3	0.5858
cosine_accuracy@5	0.681
cosine_accuracy@10	0.7754
cosine_precision@1	0.3775
cosine_precision@3	0.1955
cosine_precision@5	0.1363
cosine_precision@10	0.0777
cosine_recall@1	0.3768
cosine_recall@3	0.5854
cosine_recall@5	0.6802
cosine_recall@10	0.7739
cosine_ndcg@10	0.5692
cosine_mrr@10	0.5045
cosine_map@100	0.5122

Training Details

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 192
per_device_eval_batch_size: 192
fp16: True
multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 192
per_device_eval_batch_size: 192
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: 5e-05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1
num_train_epochs: 3
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: False
fp16: True
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: round_robin

Training Logs

Epoch	Step	val-ir-eval_cosine_ndcg@10
1.0	50	0.5375
2.0	100	0.5626
3.0	150	0.5692

Framework Versions

Python: 3.12.3
Sentence Transformers: 3.4.1
Transformers: 4.49.0
PyTorch: 2.6.0+cu124
Accelerate: 1.4.0
Datasets: 3.3.2
Tokenizers: 0.21.0

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",
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

tjohn327
/

scion-minilm-v2