PhilipCisco/qwen3-base-financial

Qwen3 base Financial

This is a sentence-transformers model finetuned from Qwen/Qwen3-Embedding-0.6B on the json dataset. It maps sentences & paragraphs to a 1024-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: Qwen/Qwen3-Embedding-0.6B
• Maximum Sequence Length: 32768 tokens
• Output Dimensionality: 1024 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • json
• 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): Transformer({'max_seq_length': 32768, 'do_lower_case': False, 'architecture': 'Qwen3Model'})
  (1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': True, '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("PhilipCisco/qwen3-base-financial")
# Run inference
queries = [
    "Which section of the financial document addresses Financial Statements and Supplementary Data?",
]
documents = [
    'Financial Statements and Supplementary Data are addressed in Item 8 of the financial document.',
    'The 7% Notes due 2029 are scheduled to mature on February 15, 2029.',
    'The gift card liability was $145,014 in 2022 and increased to $164,930 in 2023.',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 1024] [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[ 0.6634,  0.0292, -0.0534]])

Evaluation

Metrics

Information Retrieval

• Dataset: dim_1024
• Evaluated with InformationRetrievalEvaluator with these parameters:

  {
      "truncate_dim": 1024
  }
  

Metric	Value
cosine_accuracy@1	0.7179
cosine_accuracy@3	0.8536
cosine_accuracy@5	0.8771
cosine_accuracy@10	0.9207
cosine_precision@1	0.7179
cosine_precision@3	0.2845
cosine_precision@5	0.1754
cosine_precision@10	0.0921
cosine_recall@1	0.7179
cosine_recall@3	0.8536
cosine_recall@5	0.8771
cosine_recall@10	0.9207
cosine_ndcg@10	0.8231
cosine_mrr@10	0.7916
cosine_map@100	0.7948

Training Details

Training Dataset

json

• Dataset: json
• Size: 5,600 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 7 tokens mean: 20.73 tokens max: 50 tokens	min: 10 tokens mean: 47.95 tokens max: 431 tokens

• Samples:

  anchor	positive
  What was the increase in sales and marketing expenses for the year ended December 31, 2023 compared to 2022?	Sales and marketing expenses increased by $42.5 million, or 6%, for the year ended December 31, 2023 compared to 2022.
  What method is used to provide information about legal proceedings in the Annual Report on Form 10-K?	Information about legal proceedings in the Annual Report on Form 10-K is incorporated by reference under several notes and sections.
  How did selling, distribution, and administration expenses change in 2023 compared to previous years?	In 2023, the decline in Selling, distribution and administration expense was driven by lower compensation expense associated with workforce reductions, lower costs for professional services and lower freight and warehousing expenses as a result of lower shipments during 2023. Additionally, Selling, distribution and administration expense in 2023 included $116.0 million of intangible asset impairment charges as compared to $281.0 million of intangible asset impairment charges in 2022.

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          1024
      ],
      "matryoshka_weights": [
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 4
• per_device_eval_batch_size: 4
• gradient_accumulation_steps: 8
• learning_rate: 2e-05
• num_train_epochs: 4
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• bf16: True
• tf32: True
• load_best_model_at_end: True
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: epoch
• prediction_loss_only: True
• per_device_train_batch_size: 4
• per_device_eval_batch_size: 4
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 8
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 2e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 4
• max_steps: -1
• lr_scheduler_type: cosine
• 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: 42
• 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: True
• 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: 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}
• 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: no_duplicates
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	dim_1024_cosine_ndcg@10
-1	-1	-	0.7762
0.0571	10	0.0102	-
0.1143	20	0.01	-
0.1714	30	0.002	-
0.2286	40	0.0162	-
0.2857	50	0.0015	-
0.3429	60	0.0064	-
0.4	70	0.0052	-
0.4571	80	0.0026	-
0.5143	90	0.0098	-
0.5714	100	0.0137	-
0.6286	110	0.0159	-
0.6857	120	0.0062	-
0.7429	130	0.0076	-
0.8	140	0.0046	-
0.8571	150	0.0341	-
0.9143	160	0.0032	-
0.9714	170	0.0051	-
1.0	175	-	0.7873
1.0286	180	0.0106	-
1.0857	190	0.0003	-
1.1429	200	0.0011	-
1.2	210	0.0017	-
1.2571	220	0.0081	-
1.3143	230	0.0005	-
1.3714	240	0.0185	-
1.4286	250	0.0008	-
1.4857	260	0.0034	-
1.5429	270	0.0042	-
1.6	280	0.0088	-
1.6571	290	0.0026	-
1.7143	300	0.0038	-
1.7714	310	0.0032	-
1.8286	320	0.0012	-
1.8857	330	0.0027	-
1.9429	340	0.0073	-
2.0	350	0.0033	0.8056
2.0571	360	0.0013	-
2.1143	370	0.0023	-
2.1714	380	0.0094	-
2.2286	390	0.0132	-
2.2857	400	0.0026	-
2.3429	410	0.0054	-
2.4	420	0.0035	-
2.4571	430	0.0019	-
2.5143	440	0.0003	-
2.5714	450	0.0059	-
2.6286	460	0.0006	-
2.6857	470	0.0004	-
2.7429	480	0.0102	-
2.8	490	0.0011	-
2.8571	500	0.0075	-
2.9143	510	0.013	-
2.9714	520	0.0022	-
3.0	525	-	0.8238
3.0286	530	0.0019	-
3.0857	540	0.0057	-
3.1429	550	0.0042	-
3.2	560	0.0008	-
3.2571	570	0.0001	-
3.3143	580	0.0015	-
3.3714	590	0.0175	-
3.4286	600	0.0006	-
3.4857	610	0.0003	-
3.5429	620	0.0056	-
3.6	630	0.002	-
3.6571	640	0.0091	-
3.7143	650	0.0009	-
3.7714	660	0.0011	-
3.8286	670	0.0001	-
3.8857	680	0.0014	-
3.9429	690	0.0019	-
4.0	700	0.0001	0.8231

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.14
• Sentence Transformers: 5.1.0
• Transformers: 4.56.1
• PyTorch: 2.8.0+cu128
• Accelerate: 1.10.1
• Datasets: 2.19.1
• Tokenizers: 0.22.0

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",
}

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{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}
}