oneryalcin/static-embedding-chess

Static chess embedding (512d) -- themes/openings <-> positions

This is a sentence-transformers model trained. It maps sentences & paragraphs to a 512-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

• Model Type: Sentence Transformer
• Maximum Sequence Length: inf tokens
• Output Dimensionality: 512 dimensions
• Similarity Function: Cosine Similarity
• Supported Modality: Text
• Language: en
• License: apache-2.0

Model Sources

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

Full Model Architecture

SentenceTransformer(
  (0): StaticEmbedding({})
)

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("oneryalcin/static-embedding-chess")
# Run inference
queries = [
    '[UNK] deflection discoveredAttack [UNK] queensideAttack short Philidor Defense [UNK] Defense Other variations',
]
documents = [
    'themes crushing deflection discoveredAttack middlegame queensideAttack short opening Philidor Defense Philidor Defense Other variations moves d3c3 d4b3 c1b1 d7d1 d3c3+d4b3 d4b3+c1b1 c1b1+d7d1',
    'themes advantage discoveredAttack middlegame short opening Philidor Defense Philidor Defense Other variations moves e4d4 d3f5 c8b8 d1d4 e4d4+d3f5 d3f5+c8b8 c8b8+d1d4',
    'themes crushing middlegame pin queensideAttack short opening Sicilian Defense Sicilian Defense Najdorf Variation moves c3d5 c5b3 c1b1 b3d2 c3d5+c5b3 c5b3+c1b1 c1b1+b3d2',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 512] [3, 512]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[0.8405, 0.5061, 0.2136]])

Evaluation

Metrics

Information Retrieval

• Datasets: chess-ir and chess-ir-tokens
• Evaluated with InformationRetrievalEvaluator

Metric	chess-ir	chess-ir-tokens
cosine_accuracy@1	0.005	0.0794
cosine_accuracy@10	0.07	0.2593
cosine_precision@1	0.005	0.0794
cosine_precision@10	0.008	0.0603
cosine_recall@1	0.0017	0.0022
cosine_recall@10	0.0267	0.024
cosine_ndcg@10	0.0168	0.0672
cosine_mrr@10	0.0207	0.1233
cosine_map@100	0.0141	0.0332

Training Details

Training Dataset

Unnamed Dataset

• Size: 1,619,946 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 100 samples:

  	anchor	positive
  type	string	string
  modality	text	text
  details	min: 21 characters mean: 75.57 characters max: 122 characters	min: 86 characters mean: 158.13 characters max: 256 characters

• Samples:

  anchor	positive
  kingsideAttack mate mateIn1 middlegame oneMove Horwitz Defense Horwitz Defense [UNK] variations	themes kingsideAttack mate mateIn1 middlegame oneMove opening Horwitz Defense Horwitz Defense Other variations moves f7h8 g6g2 f7h8+g6g2
  backRankMate endgame mate mateIn2 short Kings Knight Opening Kings Knight Opening [UNK] [UNK]	themes backRankMate endgame mate mateIn2 short opening Kings Knight Opening Kings Knight Opening Other variations moves c5d4 c3c8 g5d8 c8d8 c5d4+c3c8 c3c8+g5d8 g5d8+c8d8
  kingsideAttack mate mateIn1 middlegame oneMove Sicilian Defense Sicilian Defense Paulsen-Basman Defense	themes kingsideAttack mate mateIn1 middlegame oneMove opening Sicilian Defense Sicilian Defense Paulsen-Basman Defense moves g3f3 c7h2 g3f3+c7h2

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          512,
          256,
          128,
          64,
          32
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 4096
• num_train_epochs: 20
• learning_rate: 0.01
• warmup_steps: 0.1
• weight_decay: 0.01
• per_device_eval_batch_size: 4096
• push_to_hub: True
• hub_model_id: oneryalcin/static-embedding-chess
• load_best_model_at_end: True
• seed: 12

All Hyperparameters

Click to expand

• per_device_train_batch_size: 4096
• num_train_epochs: 20
• max_steps: -1
• learning_rate: 0.01
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_steps: 0.1
• optim: adamw_torch_fused
• optim_args: None
• weight_decay: 0.01
• 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: 1.0
• label_smoothing_factor: 0.0
• bf16: False
• fp16: False
• 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: 4096
• 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: True
• hub_private_repo: None
• hub_model_id: oneryalcin/static-embedding-chess
• hub_strategy: every_save
• hub_always_push: False
• hub_revision: None
• load_best_model_at_end: True
• ignore_data_skip: False
• restore_callback_states_from_checkpoint: False
• full_determinism: False
• seed: 12
• 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: []
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• 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: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	chess-ir_cosine_ndcg@10	chess-ir-tokens_cosine_ndcg@10
-1	-1	-	0.0123	0.0561
0.0025	1	27.3123	-	-
0.2020	80	26.3304	-	-
0.4040	160	22.2114	-	-
0.6061	240	17.4522	-	-
0.8081	320	12.8864	-	-
1.0	396	-	0.0800	0.1181
1.0101	400	9.1439	-	-
1.2121	480	6.5434	-	-
1.4141	560	4.9138	-	-
1.6162	640	3.9819	-	-
1.8182	720	3.4584	-	-
2.0	792	-	0.0505	0.0938
2.0202	800	3.1303	-	-
2.2222	880	2.9652	-	-
2.4242	960	2.8584	-	-
2.6263	1040	2.7907	-	-
2.8283	1120	2.7475	-	-
3.0	1188	-	0.0251	0.0830
3.0303	1200	2.7031	-	-
3.2323	1280	2.6927	-	-
3.4343	1360	2.6516	-	-
3.6364	1440	2.6441	-	-
3.8384	1520	2.6202	-	-
4.0	1584	-	0.0168	0.0672

Training Time

• Training: 4.1 minutes
• Evaluation: 0.2 seconds
• Total: 4.1 minutes

Framework Versions

• Python: 3.12.10
• Sentence Transformers: 5.5.0
• Transformers: 5.8.0
• PyTorch: 2.11.0
• Accelerate: 1.13.0
• Datasets: 4.8.5
• Tokenizers: 0.22.2

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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning},
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

MultipleNegativesRankingLoss

@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},
}