Update README.md

README.md

@@ -1,469 +1,42 @@
 ---
+license: mit
 tags:
 - sentence-transformers
 - sentence-similarity
+- central-bank
+- semantic-textual-similarity
 - feature-extraction
-- generated_from_trainer
-- dataset_size:29026
-- loss:CosineSimilarityLoss
-widget:
-- source_sentence: 'Central Bankers, Bureaucratic Incentives, and Monetary Policy:
-    an Introduction'
-  sentences:
-  - The idea that institutional structures, or property rights, affect individual
-    behavior is not a new one. In fact, this idea has been the focus of much of the
-    public choice/property rights literature of the past two decades. What is relatively
-    new is the application of this principle to central bank behavior. Until recently,
-    economists have tended to analyze central banks in a Utopian framework. The implicit
-    assumption has been that central banks conduct their operations to maximize the
-    public interest. Any deviation from the optimal rate of inflation could be attributed
-    to technical problems associated with unanticipated changes in the money multiplier
-    or money demand. Because of the assumptions employed, the economist’s role is
-    that of a technician, one who identifies the precise relationships between the
-    monetary base, the money supply, and the resulting inflation rate.
-  - This paper examines empirical evidence of the adaptive learning behavior of speculators
-    in the 2005–2008 renminbi appreciation episode, and establishes a theoretical
-    model to explore appreciation policy implications of such a behavior. In our model,
-    speculators form their expectations about the future appreciation premium adaptively
-    by extrapolating past appreciation returns into the future. We find that a rapid
-    appreciation may attract more capital inflows, and the central bank may prefer
-    a slow appreciation to discourage capital inflows. Simulated results can generate
-    hump-shaped paths of the appreciation speed, expected appreciation premium, and
-    capital inflows. In addition, changes in the appreciation speed precede changes
-    in the expected appreciation premium and capital inflows. These results are consistent
-    with empirical evidence in the 2005–2008 renminbi appreciation episode.
-  - This paper makes stochastic projections of the Central Bank of Chile’s (CBCh)
-    balance sheet (stocks and flows) starting from the actual current negative net
-    worth. These projections incorporate the effect on the balance sheet of several
-    macroeconomic variables as well as alternative policy decisions, taking into account
-    the uncertainty and risks inherent in the economy. The article describes and assesses
-    the main causes of the present deficit. In the baseline scenario, the deterministic
-    forecast shows that the Central Bank's net worth will increase slowly to reach
-    positive values after 25 years. However, in the stochastic simulations the Bank's
-    net worth will still be negative 25 years from now, with a 69 percent probability.
-- source_sentence: New Forms of the Management to Loan Size and the Consequences--A
-    Factor Analysis of Money Expanding from 1996 to the Second Quarter of 2004
-  sentences:
-  - When the nominal return on all public liabilities is allowed to adjust to changing
-    market conditions, or the central bank has access to unlimited open market operations,
-    money growth is likely to stimulate output. This is shown in the model used by
-    Lucas in his Nobel Prize Lecture as an example of the non neutral effects of anticipated
-    monetary expansions. A rise in net outside assets increases households' incentives
-    to work through a reallocation of consumption across periods. This result survives
-    with non interest-bearing cash when the latter does not generate relevant distortions.
-  - central bank is insolvent if its plans imply a Ponzi scheme on reserves so the
-    price level becomes infinity. If the central bank enjoys fiscal support, in the
-    form of a dividend rule that pays out net income every period, including when
-    it is negative, it can never become insolvent independently of the fiscal authority.
-    Otherwise, this note distinguishes between intertemporal insolvency, rule insolvency,
-    and period insolvency. While period and rule solvency depend on analyzing dividend
-    rules and sources of risk to net income, evaluating intertemporal solvency requires
-    overcoming the difficult challenge of measuring the present value of seignorage.
-  - This study argues that financializations is not a phenomenon exclusively associated
-    with complex innovation in highly developed financial markets. Financialization
-    also affects countries with 'shallow' financial markets but with a significant
-    presence of transnational financial actors that become a powerful economic and
-    political force able to navigate and shape uneven regulatory and institutional
-    terrains in order to sustain new modes of profit generation. The study distinguishes
-    two stages in financialization of the Romanian economy. The first, central-bank
-    dominated stage saw the systemic transformation of firm-bank-state relations that
-    resulted in the creation of impatient banking and the forestalling of industrial
-    policy options. The second, dependent financialization, is in turn characterised
-    by new modes of profit generation for transnational financial actors, interconnectedness
-    and fragility as the main mechanism of incorporation in European financial structures.
-- source_sentence: Central Bank Participation in Currency Options Markets
-  sentences:
-  - Abstract This paper investigates the impact of the central bank's liquidity operations
-    on the financial constraints of the bank-dependent firms. We use the Reserve Bank
-    of India's liquidity operation called Term Repo Operation (TRO) in the study.
-    The empirical analysis is based on a large scale firm-level data for the period
-    2011–2016 and panel logit estimation method. Our findings indicate that the financial
-    constraints of the bank-dependent firms have reduced than their counterparts since
-    the introduction of the operation. We also show that larger firms reap significant
-    benefit out of TRO than the smaller firms.
-  - Since 1990, Poland has adopted nearly all possible exchange rate regimes, moving
-    smoothly from fixed peg to pure floating. At the beginning of the transition period
-    exchange rate policy was aimed mainly at stabilising the economy. It supported
-    monetary policy in its fight against inflation, playing the role of a nominal
-    anchor. Under the stabilisation plan, the Polish authorities decided to fix the
-    zloty to the US dollar at the level of 9,500 zlotys to one US$ in January 1990.
-    In that way the government and the central bank delivered a strong nominal anchor
-    for an economy facing dramatic and abrupt adjustments in the real sector and in
-    relative prices. At the same time, the zloty became an internally convertible
-    currency.
-  - This paper models unemployment as a general equilibrium solution in labor and
-    capital markets, while the natural rate hypothesis explains unemployment simply
-    as a partial equilibrium in the labor market. It is shown that monetary policy
-    can have long-run effects by affecting required returns on capital and investment.
-    If monetary policy is primarily concerned with maintaining price stability, the
-    interaction between wage bargaining and the central bank’s credibility as an inflation
-    fighter becomes a crucial factor in determining employment. Different labor market
-    institutions condition different monetary policy reactions. With centralized wage
-    bargaining, a central bank mandate focusing primarily on price stability is sufficient.
-    With an atomistic labor market, the central bank must also consider output as
-    a policy objective.
-- source_sentence: 'The politics of the European Central Bank: principal-agent theory
-    and the democratic deficit'
-  sentences:
-  - ABSTRACTThe stress test of the European Central Bank has become one of the primary
-    regulatory tools for the European banking system. In order to make such a regulatory
-    indicator, different national...
-  - This paper uses the VAR methodology to analyse the effects of European Central
-    Bank monetary policy shocks and euro area output and inflation shocks on the European
-    Union member states from Central and Eastern Europe. First, we look at the strength
-    of effects of identified euro area monetary policy shocks and compare the influence
-    with the one of the domestic policy shocks. Next, we turn to analysis of output
-    and inflation responses to euro area output and inflation shocks relative to aggregate
-    euro area reaction. We provide implications for each country monetary policy decision
-    process and draw conclusions concerning readiness for euro adoption, both from
-    monetary and real economy point of view.
-  - The "conservative central banker" has come under attack recently. On the basis
-    of models in which there is explicit interaction between trade union behavior
-    and monetary policy, it has been argued that if 'trade unions' are averse to inflation,
-    welfare will be lower with a conservative than with a liberal central bank. We
-    reframe this discussion in a standard trade union model. We show that the case
-    against the conservative central banker rests exclusively on the assumption of
-    a strictly nominal outside option (for instance, unemployment benefits) for the
-    union. There is no welfare gain associated with making the central bank less conservative
-    than society, however, if the outside option is in real terms. As the nominal
-    components of the trade union's outside option are mainly public transfers, we
-    also show that the conservative central banker is always optimal if the government
-    can choose the level of nominal unemployment benefits as well as the degree of
-    central bank conservatism.
-- source_sentence: Global Liquidity and Monetary Policy Autonomy
-  sentences:
-  - In the 1970s monetarism was accepted as a principle of policy of central banks.
-    However monetary growth targets currently play no official role in the monetary
-    policy of industrial countries. The majority of monetary authorities excluded
-    money from their analysis. Inflation, GDP and unemployment are not determined
-    by money supply in their models. This article provides a historical perspective
-    on the development and apparent failure of monetarism as a policy guide.
-  - Deep financial crisis which started in 2007 proved to be extremely contagious
-    affecting the financial and banking EU system. Achieving an integrated banking
-    market is the main component of the European policy in the financial-banking services
-    area. The latest developments underlined that the difficulties faced by the banks
-    can negatively impact on the entire financial stability of the member states.
-    That's why, the European Central Bank will be entitled to supervise any bank of
-    the euro area, especially the ones that benefit of public support. Reforming the
-    financial and banking system must be shaped in the frame of insuring some durable
-    national finances, of an urgent recapitalizing of the banks that need that and
-    of elaborating some common fiscal and financial and banking regulations effective
-    in the eurozone
-  - We have confirmed that the cause of Japan’s recession has been the sharp reduction
-    in credit creation that began in 1992 and was triggered by the bad debts in the
-    banking system. We have also found that this was due to excessive loan growth
-    quotas imposed on the banks by the Bank of Japan during the 1980s. Finally, we
-    found that the problem of lack of credit during the 1990s could easily have been
-    solved through monetary policy. Bad debts could have been taken off the banks’
-    balance sheets without costs by the central bank. Even without bank lending, the
-    central bank could have created a recovery a decade ago, by significantly increasing
-    its own credit creation. In other words, Japan’s recession of the 1990s has been
-    the result of the Bank of Japan’s policies.2
-pipeline_tag: sentence-similarity
+- dataset:sentence-transformers/s2orc
+- dataset:allenai/peS2o
 library_name: sentence-transformers
+pipeline_tag: sentence-similarity
+base_model: BAAI/bge-large-en-v1.5
+language: en
+datasets:
+- allenai/peS2o
+- sentence-transformers/s2orc
 ---
 
-# SentenceTransformer
-
-This is a [sentence-transformers](https://www.SBERT.net) model trained. 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:** [Unknown](https://huggingface.co/unknown) -->
-- **Maximum Sequence Length:** 300 tokens
-- **Output Dimensionality:** 1024 dimensions
-- **Similarity Function:** Cosine Similarity
-<!-- - **Training Dataset:** Unknown -->
-<!-- - **Language:** Unknown -->
-<!-- - **License:** Unknown -->
-
-### Model Sources
-
-- **Documentation:** [Sentence Transformers Documentation](https://sbert.net)
-- **Repository:** [Sentence Transformers on GitHub](https://github.com/UKPLab/sentence-transformers)
-- **Hugging Face:** [Sentence Transformers on Hugging Face](https://huggingface.co/models?library=sentence-transformers)
-
-### Full Model Architecture
-
-```
-SentenceTransformer(
-  (0): Transformer({'max_seq_length': 300, 'do_lower_case': True}) with Transformer model: BertModel 
-  (1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': True, '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': False, 'include_prompt': True})
-  (2): Normalize()
-)
-```
-
-## Usage
-
-### Direct Usage (Sentence Transformers)
-
-First install the Sentence Transformers library:
-
-```bash
-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("sentence_transformers_model_id")
-# Run inference
-sentences = [
-    'Global Liquidity and Monetary Policy Autonomy',
-    'In the 1970s monetarism was accepted as a principle of policy of central banks. However monetary growth targets currently play no official role in the monetary policy of industrial countries. The majority of monetary authorities excluded money from their analysis. Inflation, GDP and unemployment are not determined by money supply in their models. This article provides a historical perspective on the development and apparent failure of monetarism as a policy guide.',
-    "Deep financial crisis which started in 2007 proved to be extremely contagious affecting the financial and banking EU system. Achieving an integrated banking market is the main component of the European policy in the financial-banking services area. The latest developments underlined that the difficulties faced by the banks can negatively impact on the entire financial stability of the member states. That's why, the European Central Bank will be entitled to supervise any bank of the euro area, especially the ones that benefit of public support. Reforming the financial and banking system must be shaped in the frame of insuring some durable national finances, of an urgent recapitalizing of the banks that need that and of elaborating some common fiscal and financial and banking regulations effective in the eurozone",
-]
-embeddings = model.encode(sentences)
-print(embeddings.shape)
-# [3, 1024]
-
-# Get the similarity scores for the embeddings
-similarities = model.similarity(embeddings, embeddings)
-print(similarities.shape)
-# [3, 3]
-```
-
-<!--
-### Direct Usage (Transformers)
-
-<details><summary>Click to see the direct usage in Transformers</summary>
-
-</details>
--->
-
-<!--
-### Downstream Usage (Sentence Transformers)
-
-You can finetune this model on your own dataset.
-
-<details><summary>Click to expand</summary>
-
-</details>
--->
-
-<!--
-### Out-of-Scope Use
-
-*List how the model may foreseeably be misused and address what users ought not to do with the model.*
--->
-
-<!--
-## Bias, Risks and Limitations
-
-*What are the known or foreseeable issues stemming from this model? You could also flag here known failure cases or weaknesses of the model.*
--->
-
-<!--
-### Recommendations
-
-*What are recommendations with respect to the foreseeable issues? For example, filtering explicit content.*
--->
-
-## Training Details
-
-### Training Dataset
-
-#### Unnamed Dataset
-
-* Size: 29,026 training samples
-* Columns: <code>sentence_0</code>, <code>sentence_1</code>, and <code>label</code>
-* Approximate statistics based on the first 1000 samples:
-  |         | sentence_0                                                                        | sentence_1                                                                           | label                                                          |
-  |:--------|:----------------------------------------------------------------------------------|:-------------------------------------------------------------------------------------|:---------------------------------------------------------------|
-  | type    | string                                                                            | string                                                                               | float                                                          |
-  | details | <ul><li>min: 6 tokens</li><li>mean: 15.09 tokens</li><li>max: 48 tokens</li></ul> | <ul><li>min: 18 tokens</li><li>mean: 134.34 tokens</li><li>max: 288 tokens</li></ul> | <ul><li>min: 0.0</li><li>mean: 0.48</li><li>max: 1.0</li></ul> |
-* Samples:
-  | sentence_0                                                                                                                                                                     | sentence_1                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                   | label            |
-  |:-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|:-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|:-----------------|
-  | <code>Half-Life Estimation under the Taylor Rule ∗</code>                                                                                                                      | <code>Abstract We aim to investigate the simultaneous and interacted effects of central bank qualitative and quantitative communication on private inflation expectations, measured with survey and market-based measures. The effects of ECB inflation projections and Governing Council members’ speeches are identified through an instrumental-variables estimation using a principal component analysis to generate relevant instruments. We find that ECB projections have a positive effect on current-year forecasts, and that ECB projections and speeches are substitutes at longer horizons. Moreover, ECB speeches and the ECB rate reinforce the effect of ECB projections when they are consistent, and convey the same signal about inflationary pressures.</code>                                                                                                                                                                                            | <code>0.0</code> |
-  | <code>Lasting Legacy on Financial Regulatory reform</code>                                                                                                                     | <code>At their meeting on October 23, the G20 finance ministers and central bank governors prioritized a number of regulatory issues for discussion at the leader’s summit. The familiar items included a ritualistic commitment to implement all reforms endorsed already by the G20 in an internationally consistent and non-discriminatory manner, such as those relating to over-the-counter derivatives, compensation practices, accounting standards and credit rating agencies.</code>                                                                                                                                                                                                                                                                                                                                                                                                                                                                                | <code>1.0</code> |
-  | <code>Tři roky po měnové krizi: rekapitulace událostí a jejich souvislostí [Three years after the exchange rate crisis: recapitulation of events and their connections]</code> | <code>We examine the European Central Bank's ad-hoc communication and explore how it informs future monetary policy decisions. Using the rich dataset of the inter-meeting verbal communication among the members of the European Central Bank's Governing Council between 2008 and 2014, we construct a measure of communication assessing its inclination towards easing, tightening or maintaining the monetary policy stance. We find that this measure provides useful additional information about future monetary policy decisions, even when we control for market-based interest rate expectations and lagged decisions. Our results also suggest that, in particular, communication shortly before monetary policy meetings, related to unconventional measures and/or by the ECB President explain the future ECB rate changes well. Overall, these results point to the importance of transparency in understanding the future course of monetary policy.</code> | <code>0.0</code> |
-* Loss: [<code>CosineSimilarityLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#cosinesimilarityloss) with these parameters:
-  ```json
-  {
-      "loss_fct": "torch.nn.modules.loss.MSELoss"
-  }
-  ```
-
-### Training Hyperparameters
-#### Non-Default Hyperparameters
-
-- `per_device_train_batch_size`: 16
-- `per_device_eval_batch_size`: 16
-- `num_train_epochs`: 1
-- `multi_dataset_batch_sampler`: round_robin
-
-#### All Hyperparameters
-<details><summary>Click to expand</summary>
-
-- `overwrite_output_dir`: False
-- `do_predict`: False
-- `eval_strategy`: no
-- `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`: 1
-- `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}
-- `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
-- `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
-
-</details>
-
-### Training Logs
-| Epoch  | Step | Training Loss |
-|:------:|:----:|:-------------:|
-| 0.2755 | 500  | 0.0844        |
-| 0.5510 | 1000 | 0.0667        |
-| 0.8264 | 1500 | 0.0607        |
+# bge-centralbank
 
+**bge-centralbank** is a domain-adapted Sentence Transformer developed to assess semantic similarity in central bank-related texts. It is based on [`BAAI/bge-large-en-v1.5`](https://huggingface.co/BAAI/bge-large-en-v1.5), and adapted through both unsupervised and supervised training.
 
-### Framework Versions
-- Python: 3.11.13
-- Sentence Transformers: 4.1.0
-- Transformers: 4.52.4
-- PyTorch: 2.7.1+cu126
-- Accelerate: 1.7.0
-- Datasets: 2.14.4
-- Tokenizers: 0.21.1
+##  Training Setup
 
-## Citation
+### 1. Pretraining
 
-### BibTeX
+The model was first pretrained using TSDAE on 177,842 title–abstract pairs drawn from the `peS2o` dataset — a lightweight and pre-cleaned subset of the full S2ORC corpus. From this dataset, a domain-specific corpus was constructed by filtering on keywords relevant to macroeconomics, monetary policy, and financial markets. This step enables the model to better capture the language structures and terminology common in central bank literature.
 
-#### 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",
-}
-```
+### 2. Supervised Fine-tuning
 
-<!--
-## Glossary
+Fine-tuning was conducted using a subset of the `sentence-transformers/s2orc` dataset. All **title–abstract pairs** were selected where the abstract contained the term *central bank*, resulting in 15,513 positive examples (label = 1). Each pair represents a real paper, with a matching title and abstract.
 
-*Clearly define terms in order to be accessible across audiences.*
--->
+An equal number of negative examples (label = 0) was generated by randomly mismatching titles and abstracts from unrelated papers.
 
-<!--
-## Model Card Authors
+A separate validation set of 2,000 title–abstract pairs (1,000 positive, 1,000 negative) was held out during training. The remaining 29,026 examples were used for supervised training with `CosineSimilarityLoss` using the `sentence-transformers` framework.
 
-*Lists the people who create the model card, providing recognition and accountability for the detailed work that goes into its construction.*
--->
+##  Evaluation
 
-<!--
-## Model Card Contact
+The model was evaluated on the held-out validation set. Compared to the base model, `bge-centralbank` achieved a substantially lower average similarity score on negative pairs (from **0.6230** to **0.1177**), showing improved ability to distinguish semantically unrelated text. It also achieved a higher **point-biserial correlation** (from **0.7933** to **0.9025**), indicating better alignment with binary STS labels.
 
-*Provides a way for people who have updates to the Model Card, suggestions, or questions, to contact the Model Card authors.*
--->
+---