MARK-Embedding / README.md

summerstars

Update README.md

f2ab02d verified 5 months ago

preview code

raw

history blame contribute delete

12.7 kB

metadata

tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - dense
  - generated_from_trainer
  - dataset_size:5749
  - loss:CosineSimilarityLoss
widget:
  - source_sentence: >-
      Nterprise Linux Services is expected to be available before then end of
      this year.
    sentences:
      - >-
        Beta versions of Nterprise Linux Services are expected to be available
        on certain HP ProLiant servers in July.
      - Spain turning back the clock on siestas
      - I don't like many flavored drinks.
  - source_sentence: Iran hopes nuclear talks will yield 'roadmap'
    sentences:
      - Iran Nuclear Talks in Geneva Spur High Hopes
      - A black pet dog runs around in the garden of a house.
      - >-
        The witness was a 27-year-old Kosovan parking attendant, who was paid by
        the News of the World, the court heard.
  - source_sentence: Hamas Urges Hizbullah to Pull Fighters Out of Syria
    sentences:
      - >-
        "This was a persistent problem which has not been solved, mechanically
        and physically," said board member Steven Wallace.
      - A small dog jumps over a yellow beam.
      - Hamas calls on Hezbollah to pull forces out of Syria
  - source_sentence: Licensing revenue slid 21 percent, however, to $107.6 million.
    sentences:
      - Britain loses bid to deport radical cleric Abu Qatada
      - A man sits on a bed very close to a small television.
      - >-
        License sales, a key measure of demand, fell 21 percent to $107.6
        million.
  - source_sentence: >-
      Comcast Class A shares were up 8 cents at $30.50 in morning trading on the
      Nasdaq Stock Market.
    sentences:
      - The stock rose 48 cents to $30 yesterday in Nasdaq Stock Market trading.
      - 'Malaysia: Chinese satellite found object in ocean'
      - A boy in a robe sits in a chair.
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - pearson_cosine
  - spearman_cosine
model-index:
  - name: SentenceTransformer
    results:
      - task:
          type: semantic-similarity
          name: 意味的類似性 (Semantic Similarity)
        metrics:
          - type: pearson_cosine
            value: 0.4639747212598005
            name: ピアソン相関係数 (コサイン類似度)
          - type: spearman_cosine
            value: 0.4595105448711385
            name: スピアマン相関係数 (コサイン類似度)
license: gemma

SentenceTransformer

これは、訓練済みのsentence-transformersモデルです。このモデルは、文と段落を256次元の密なベクトル空間にマッピングし、意味的テキスト類似性、意味検索、言い換えマイニング、テキスト分類、クラスタリングなどに使用できます。

モデル詳細

モデルの説明

モデルタイプ: Sentence Transformer
最大シーケンス長: 2048トークン
出力次元数: 256次元
類似度関数: コサイン類似度

モデルのソース

ドキュメント: Sentence Transformers Documentation
リポジトリ: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

完全なモデルアーキテクチャ

SentenceTransformer(
  (0): Transformer({'max_seq_length': 2048, 'do_lower_case': False, 'architecture': 'Gemma3TextModel'})
  (1): Pooling({'word_embedding_dimension': 256, '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})
)

使用方法

直接使用 (Sentence Transformers)

まず、Sentence Transformersライブラリをインストールします:

pip install -U sentence-transformers

次に、このモデルをロードして推論を実行できます。

from sentence_transformers import SentenceTransformer

# 🤗 Hubからダウンロード
model = SentenceTransformer("sentence_transformers_model_id")
# 推論を実行
sentences = [
    'Comcast Class A shares were up 8 cents at $30.50 in morning trading on the Nasdaq Stock Market.',
    'The stock rose 48 cents to $30 yesterday in Nasdaq Stock Market trading.',
    'Malaysia: Chinese satellite found object in ocean',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 256]

# 埋め込みベクトルの類似度スコアを取得
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.5752, 0.2980],
#         [0.5752, 1.0000, 0.2161],
#         [0.2980, 0.2161, 1.0000]])

評価

メトリクス

意味的類似性

EmbeddingSimilarityEvaluatorで評価

メトリクス	値
pearson_cosine	0.464
spearman_cosine	0.4595

訓練詳細

訓練データセット

名称未設定のデータセット

サイズ: 5,749 訓練サンプル
カラム: sentence_0, sentence_1, label

最初の1000サンプルに基づくおおよその統計:

	sentence_0	sentence_1	label
型	string	string	float
詳細	最小: 6 トークン平均: 14.76 トークン最大: 55 トークン	最小: 6 トークン平均: 14.73 トークン最大: 57 トークン	最小: 0.0 平均: 0.55 最大: 1.0

サンプル:

sentence_0	sentence_1	label
`Forecasters said warnings might go up for Cuba later Thursday.`	`Watches or warnings could be issued for eastern Cuba later on Thursday.`	`0.8`
`Death toll in Lebanon bombings rises to 47`	`1 suspect arrested after Lebanon car bombings kill 45`	`0.5599999904632569`
`Three dogs running on a racetrack.`	`Three dogs round a bend at a racetrack.`	`0.9600000381469727`

損失関数: CosineSimilarityLoss 以下のパラメータを使用:
```
{
    "loss_fct": "torch.nn.modules.loss.MSELoss"
}
```

訓練ハイパーパラメータ

デフォルト以外のハイパーパラメータ

eval_strategy: steps
per_device_train_batch_size: 16
per_device_eval_batch_size: 16
multi_dataset_batch_sampler: round_robin

すべてのハイパーパラメータ

クリックして展開

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 16
per_device_eval_batch_size: 16
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: 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}
parallelism_config: None
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch_fused
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
hub_revision: None
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
liger_kernel_config: None
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: round_robin
router_mapping: {}
learning_rate_mapping: {}

訓練ログ

エポック	ステップ	訓練損失	spearman_cosine
1.0	360	-	0.2967
1.3889	500	0.11	0.3338
2.0	720	-	0.3665
2.7778	1000	0.0857	0.4101
3.0	1080	-	0.4595

フレームワークのバージョン

Python: 3.12.11
Sentence Transformers: 5.1.0
Transformers: 4.56.1
PyTorch: 2.8.0+cu126
Accelerate: 1.10.1
Datasets: 4.0.0
Tokenizers: 0.22.0

引用

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",
}