Update README.md

README.md

@@ -1,42 +1,106 @@
----
+v---
 tags:
 - sentence-transformers
+- sentence-similarity
 - cross-encoder
 - reranker
-- generated_from_trainer
-- dataset_size:1792739
-- loss:CachedMultipleNegativesRankingLoss
-base_model: tomaarsen/Qwen3-Reranker-0.6B-seq-cls
+- feature-extraction
+- telepix
 pipeline_tag: text-ranking
 library_name: sentence-transformers
+license: apache-2.0
 ---
+<p align="center">
+    <img src="https://cdn-uploads.huggingface.co/production/uploads/61d6f4a4d49065ee28a1ee7e/V8n2En7BlMNHoi1YXVv8Q.png" width="400"/>
+<p>
 
-# CrossEncoder based on tomaarsen/Qwen3-Reranker-0.6B-seq-cls
-
-This is a [Cross Encoder](https://www.sbert.net/docs/cross_encoder/usage/usage.html) model finetuned from [tomaarsen/Qwen3-Reranker-0.6B-seq-cls](https://huggingface.co/tomaarsen/Qwen3-Reranker-0.6B-seq-cls) on the json dataset using the [sentence-transformers](https://www.SBERT.net) library. It computes scores for pairs of texts, which can be used for text reranking and semantic search.
+# PIXIE-Spell-Reranker-Preview-0.6B
+**PIXIE-Spell-Reranker-Preview-0.6B** is a decoder-based reranker trained on Korean and English dataset, 
+developed by [TelePIX Co., Ltd](https://telepix.net/).
+**PIXIE** stands for Tele**PIX** **I**ntelligent **E**mbedding, representing TelePIX’s high-performance embedding technology.
+This model is specifically optimized for semantic reranking tasks in Korean and English, and demonstrates strong performance in aerospace domain applications. Through extensive fine-tuning and domain-specific evaluation, PIXIE shows robust reranking quality for real-world use cases such as document understanding, technical QA, and semantic search in aerospace and related high-precision fields.
+It also performs competitively across a wide range of open-domain Korean and English retrieval benchmarks, making it a versatile foundation for multilingual reranking systems.
 
-## Model Details
 
-### Model Description
+## Model Description
 - **Model Type:** Cross Encoder
-- **Base model:** [tomaarsen/Qwen3-Reranker-0.6B-seq-cls](https://huggingface.co/tomaarsen/Qwen3-Reranker-0.6B-seq-cls) <!-- at revision 6a5829f5079c66e78d911e06fe21931cc00232f7 -->
+<!-- - **Base model:** [Unknown](https://huggingface.co/unknown) -->
 - **Maximum Sequence Length:** 40960 tokens
-- **Number of Output Labels:** 1 label
-- **Training Dataset:**
-    - json
-<!-- - **Language:** Unknown -->
-<!-- - **License:** Unknown -->
-
-### Model Sources
-
-- **Documentation:** [Sentence Transformers Documentation](https://sbert.net)
-- **Documentation:** [Cross Encoder Documentation](https://www.sbert.net/docs/cross_encoder/usage/usage.html)
-- **Repository:** [Sentence Transformers on GitHub](https://github.com/UKPLab/sentence-transformers)
-- **Hugging Face:** [Cross Encoders on Hugging Face](https://huggingface.co/models?library=sentence-transformers&other=cross-encoder)
-
-## Usage
-
-### Direct Usage (Sentence Transformers)
+- **Language:** Multilingual — optimized for high performance in Korean and English
+- **Domain Specialization:** Aerospace
+- **License:** apache-2.0 
+
+
+## Quality Benchmarks
+**PIXIE-Spell-Reranker-Preview-0.6B** is a multilingual reranker specialized for Korean and English reranking tasks. 
+It delivers consistently strong performance across a diverse set of domain-specific and open-domain benchmarks in both languages, demonstrating its effectiveness in real-world reranking applications.
+The table below presents the reranking performance of several rerankers evaluated on a variety of Korean and English benchmarks.
+We report **Normalized Discounted Cumulative Gain (NDCG)** scores, which measure how well a ranked list of documents aligns with ground truth relevance. Higher values indicate better reranking quality.  
+- **Avg. NDCG**: Average of NDCG@1, @3, @5, and @10 across all benchmark datasets.  
+- **NDCG@k**: Relevance quality of the top-*k* retrieved results.
+  
+All evaluations were conducted using the open-source **[Korean-MTEB-Retrieval-Evaluators](https://github.com/BM-K/Korean-MTEB-Retrieval-Evaluators)** codebase to ensure consistent dataset handling, indexing, retrieval, and NDCG@k computation across models.
+
+#### 6 Datasets of MTEB (Korean)
+Our model, **telepix/PIXIE-Spell-Reranker-Preview-0.6B**, achieves strong performance across most metrics and benchmarks, demonstrating strong generalization across domains such as multi-hop QA, long-document retrieval, public health, and e-commerce.
+
+| Model Name | # params | Avg. NDCG | NDCG@1 | NDCG@3 | NDCG@5 | NDCG@10 |
+|------|:---:|:---:|:---:|:---:|:---:|:---:|
+| telepix/PIXIE-Spell-Reranker-Preview-0.6B | 0.6B | 0.7896 | 0.7494 | 0.7910 | 0.8022 | 0.8168 |
+|  |  |  |  |  |  |  |
+| BAAI/bge-reranker-v2-m3 | 0.5B | 0.7861 | 0.7448 | 0.7868 | 0.7998 | 0.8133 |
+| dragonkue/bge-reranker-v2-m3-ko | 0.5B | 0.7849 | 0.7505 | 0.7843 | 0.7959 | 0.8089 |
+| Alibaba-NLP/gte-multilingual-reranker-base | 0.3B | 0.7594 | 0.7067 | 0.7610 | 0.7778 | 0.7922 |
+| jinaai/jina-reranker-v2-base-multilingual | 0.3B | 0.6879 | 0.6410 | 0.6888 | 0.7027 | 0.7192 |
+> **Note:** SPLADE shortlist size fixed at **`candidate_k = 100`** for all experiments.
+
+Descriptions of the benchmark datasets used for evaluation are as follows:
+- **Ko-StrategyQA**  
+  A Korean multi-hop open-domain question answering dataset designed for complex reasoning over multiple documents.
+- **AutoRAGRetrieval**  
+  A domain-diverse retrieval dataset covering finance, government, healthcare, legal, and e-commerce sectors.
+- **MIRACLRetrieval**  
+  A document retrieval benchmark built on Korean Wikipedia articles.
+- **PublicHealthQA**  
+  A retrieval dataset focused on medical and public health topics.
+- **BelebeleRetrieval**  
+  A dataset for retrieving relevant content from web and news articles in Korean.
+- **MultiLongDocRetrieval**  
+  A long-document retrieval benchmark based on Korean Wikipedia and mC4 corpus.
+
+> **Note:**
+> While many benchmark datasets are available for evaluation, in this project we chose to use only those that contain clean positive documents for each query. Keep in mind that a benchmark dataset is just that a benchmark. For real-world applications, it is best to construct an evaluation dataset tailored to your specific domain and evaluate embedding models, such as PIXIE, in that environment to determine the most suitable one.
+
+#### 7 Datasets of BEIR (English)
+Our model, **telepix/PIXIE-Spell-Reranker-Preview-0.6B**, achieves strong performance on a wide range of tasks, including fact verification, multi-hop question answering, financial QA, and scientific document retrieval, demonstrating competitive generalization across diverse domains.
+ 
+| Model Name | # params | Avg. NDCG | NDCG@1 | NDCG@3 | NDCG@5 | NDCG@10 |
+|------|:---:|:---:|:---:|:---:|:---:|:---:|
+| telepix/PIXIE-Spell-Reranker-Preview-0.6B | 0.6B | 0.3635 | 0.3692 | 0.3663 | 0.3589 | 0.3594 |
+|  |  |  |  |  |  |  |
+| Alibaba-NLP/gte-multilingual-reranker-base | 0.3B | 0.3284 | 0.3238 | 0.3297 | 0.3282 | 0.3320 |
+| BAAI/bge-reranker-v2-m3 | 0.5B | 0.3143 | 0.3129 | 0.3158 | 0.3124 | 0.3162 |
+| jinaai/jina-reranker-v2-base-multilingual | 0.3B | 0.3118 | 0.3051 | 0.3132 | 0.3104 | 0.3187 |
+| dragonkue/bge-reranker-v2-m3-ko | 0.5B | 0.3042 | 0.3033 | 0.3035 | 0.3016 | 0.3087 |
+> **Note:** BM25 shortlist size fixed at **`candidate_k = 100`** for all experiments.
+
+Descriptions of the benchmark datasets used for evaluation are as follows:
+- **ArguAna**  
+  A dataset for argument retrieval based on claim-counterclaim pairs from online debate forums.
+- **FEVER**  
+  A fact verification dataset using Wikipedia for evidence-based claim validation.
+- **FiQA-2018**  
+  A retrieval benchmark tailored to the finance domain with real-world questions and answers.
+- **HotpotQA**  
+  A multi-hop open-domain QA dataset requiring reasoning across multiple documents.
+- **MSMARCO**  
+  A large-scale benchmark using real Bing search queries and corresponding web documents.
+- **NQ**  
+  A Google QA dataset where user questions are answered using Wikipedia articles.
+- **SCIDOCS**  
+  A citation-based document retrieval dataset focused on scientific papers.
+  
+## Direct Use (Semantic Search)
 
 First install the Sentence Transformers library:
 
@@ -46,566 +110,65 @@ pip install -U sentence-transformers
 
 Then you can load this model and run inference.
 ```python
+# Requires transformers>=4.51.0
 from sentence_transformers import CrossEncoder
 
-# Download from the 🤗 Hub
-model = CrossEncoder("cross_encoder_model_id")
-# Get scores for pairs of texts
-pairs = [
-    ['<|im_start|>system\nJudge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be "yes" or "no".<|im_end|>\n<|im_start|>user\n<Instruct>: Given a web search query, retrieve relevant passages that answer the query\n<Query>: ATP란?\n', '<Document>: 아데노신 삼인산 아데노신 삼인산(, ATP)은 생명체의 주된 에너지원이다.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n'],
-    ['<|im_start|>system\nJudge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be "yes" or "no".<|im_end|>\n<|im_start|>user\n<Instruct>: Given a web search query, retrieve relevant passages that answer the query\n<Query>: 난촨구와 둥촨구는 어느 나라에 위치해 있습니까?\n', '<Document>: 난촨구(南川区)는 중국 충칭의 구이자 이전의 현이다.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n'],
-    ['<|im_start|>system\nJudge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be "yes" or "no".<|im_end|>\n<|im_start|>user\n<Instruct>: Given a web search query, retrieve relevant passages that answer the query\n<Query>: 그저우와 헤이룽장성 동닝은 어떤 나라와 접경하고 있습니까?\n', '<Document>: 허주(贺州)는 중화인민공화국 광시 좡족 자치구 북동부에 위치한 지급시이다.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n'],
-    ['<|im_start|>system\nJudge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be "yes" or "no".<|im_end|>\n<|im_start|>user\n<Instruct>: Given a web search query, retrieve relevant passages that answer the query\n<Query>: 가짜대나무(Pseudosasa)와 별꽃(Cerastium)은 모두 자생 식물과 관련이 있습니까?\n', '<Document>: 가짜사사(Pseudosasa)는 풀과에 속하는 동아시아 대나무의 속입니다.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n'],
-    ['<|im_start|>system\nJudge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be "yes" or "no".<|im_end|>\n<|im_start|>user\n<Instruct>: Given a web search query, retrieve relevant passages that answer the query\n<Query>: 샤허(Shahhe), 허베이(河北)와 조청(邹城)은 모두 현급 도시인가요?\n', '<Document>: 샤허(Shahe)는 중국 허베이성의 남부에 위치한 싱타이(Xingtai) 지구의 군급 도시입니다.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n'],
-]
-scores = model.predict(pairs)
-print(scores.shape)
-# (5,)
-
-# Or rank different texts based on similarity to a single text
-ranks = model.rank(
-    '<|im_start|>system\nJudge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be "yes" or "no".<|im_end|>\n<|im_start|>user\n<Instruct>: Given a web search query, retrieve relevant passages that answer the query\n<Query>: ATP란?\n',
-    [
-        '<Document>: 아데노신 삼인산 아데노신 삼인산(, ATP)은 생명체의 주된 에너지원이다.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n',
-        '<Document>: 난촨구(南川区)는 중국 충칭의 구이자 이전의 현이다.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n',
-        '<Document>: 허주(贺州)는 중화인민공화국 광시 좡족 자치구 북동부에 위치한 지급시이다.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n',
-        '<Document>: 가짜사사(Pseudosasa)는 풀과에 속하는 동아시아 대나무의 속입니다.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n',
-        '<Document>: 샤허(Shahe)는 중국 허베이성의 남부에 위치한 싱타이(Xingtai) 지구의 군급 도시입니다.<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n',
-    ]
-)
-# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]
-```
-
-<!--
-### 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.*
--->
+def format_queries(query, instruction=None):
+    prefix = '<|im_start|>system\nJudge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be "yes" or "no".<|im_end|>\n<|im_start|>user\n'
+    if instruction is None:
+        instruction = (
+            "Given a web search query, retrieve relevant passages that answer the query"
+        )
+    return f"{prefix}<Instruct>: {instruction}\n<Query>: {query}\n"
 
-<!--
-## 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.*
--->
+def format_document(document):
+    suffix = "<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n"
+    return f"<Document>: {document}{suffix}"
 
-<!--
-### Recommendations
 
-*What are recommendations with respect to the foreseeable issues? For example, filtering explicit content.*
--->
+model = CrossEncoder("telepix/PIXIE-Spell-Reranker-Preview-0.6B")
 
-## Training Details
+task = "Given a web search query, retrieve relevant passages that answer the query"
 
-### Training Dataset
-
-#### json
-
-* Dataset: json
-* Size: 1,792,739 training samples
-* Columns: <code>query</code>, <code>positive</code>, <code>negative_1</code>, <code>negative_2</code>, and <code>negative_3</code>
-* Approximate statistics based on the first 1000 samples:
-  |         | query                                                                                             | positive                                                                                         | negative_1                                                                                       | negative_2                                                                                      | negative_3                                                                                       |
-  |:--------|:--------------------------------------------------------------------------------------------------|:-------------------------------------------------------------------------------------------------|:-------------------------------------------------------------------------------------------------|:------------------------------------------------------------------------------------------------|:-------------------------------------------------------------------------------------------------|
-  | type    | string                                                                                            | string                                                                                           | string                                                                                           | string                                                                                          | string                                                                                           |
-  | details | <ul><li>min: 289 characters</li><li>mean: 317.46 characters</li><li>max: 406 characters</li></ul> | <ul><li>min: 90 characters</li><li>mean: 154.19 characters</li><li>max: 184 characters</li></ul> | <ul><li>min: 72 characters</li><li>mean: 149.13 characters</li><li>max: 184 characters</li></ul> | <ul><li>min: 79 characters</li><li>mean: 148.5 characters</li><li>max: 184 characters</li></ul> | <ul><li>min: 70 characters</li><li>mean: 149.09 characters</li><li>max: 184 characters</li></ul> |
-* Samples:
-  | query                                                                                                                                                                                                                                                                                                                                                       | positive                                                                                                                                 | negative_1                                                                                                                                  | negative_2                                                                                                                                       | negative_3                                                                                                                                        |
-  |:------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|:-----------------------------------------------------------------------------------------------------------------------------------------|:--------------------------------------------------------------------------------------------------------------------------------------------|:-------------------------------------------------------------------------------------------------------------------------------------------------|:--------------------------------------------------------------------------------------------------------------------------------------------------|
-  | <code><|im_start|>system<br>Judge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be "yes" or "no".<|im_end|><br><|im_start|>user<br><Instruct>: Given a web search query, retrieve relevant passages that answer the query<br><Query>: ATP란?<br></code>                            | <code><Document>: 아데노신 삼인산 아데노신 삼인산(, ATP)은 생명체의 주된 에너지원이다.<|im_end|><br><|im_start|>assistant<br><think><br><br></think><br><br></code> | <code><Document>: ATP ATP는 다음 뜻의 약자이다.<|im_end|><br><|im_start|>assistant<br><think><br><br></think><br><br></code>                         | <code><Document>: 해당 실제로 ADP는 ADPMg로, ATP는 ATPMg로 존재한다.<|im_end|><br><|im_start|>assistant<br><think><br><br></think><br><br></code>             | <code><Document>: ATE ATE는 다음을 가리킨다.<|im_end|><br><|im_start|>assistant<br><think><br><br></think><br><br></code>                                 |
-  | <code><|im_start|>system<br>Judge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be "yes" or "no".<|im_end|><br><|im_start|>user<br><Instruct>: Given a web search query, retrieve relevant passages that answer the query<br><Query>: 난촨구와 둥촨구는 어느 나라에 위치해 있습니까?<br></code>       | <code><Document>: 난촨구(南川区)는 중국 충칭의 구이자 이전의 현이다.<|im_end|><br><|im_start|>assistant<br><think><br><br></think><br><br></code>             | <code><Document>: 남풍현(南丰县)은 중국 장시성(江西省) 푸저우(福州)에 위치한 군이다.<|im_end|><br><|im_start|>assistant<br><think><br><br></think><br><br></code>      | <code><Document>: 도교, 광둥 도교(道滘)는 중국 남부 광둥성 동관 시의 관할 하에 있는 도시입니다.<|im_end|><br><|im_start|>assistant<br><think><br><br></think><br><br></code>    | <code><Document>: 동포구 동포구는 중국 쓰촨성의 구역입니다. 이곳은 메이산시의 관할 하에 있습니다.<|im_end|><br><|im_start|>assistant<br><think><br><br></think><br><br></code>      |
-  | <code><|im_start|>system<br>Judge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be "yes" or "no".<|im_end|><br><|im_start|>user<br><Instruct>: Given a web search query, retrieve relevant passages that answer the query<br><Query>: 그저우와 헤이룽장성 동닝은 어떤 나라와 접경하고 있습니까?<br></code> | <code><Document>: 허주(贺州)는 중화인민공화국 광시 좡족 자치구 북동부에 위치한 지급시이다.<|im_end|><br><|im_start|>assistant<br><think><br><br></think><br><br></code> | <code><Document>: 지관구(지관구)는 중국 인민공화국 헤이룽장성 지시시의 구이자 시청 소재지입니다.<|im_end|><br><|im_start|>assistant<br><think><br><br></think><br><br></code> | <code><Document>: 헤동 가도(河东街道)는 중국 광시(广西) 리우저우(柳州) 청중 구(城中区)의 가도입니다.<|im_end|><br><|im_start|>assistant<br><think><br><br></think><br><br></code> | <code><Document>: 화닝현 (华宁县; 병음: Huáníng Xiàn)은 중국 윈난성 유시시에 위치해 있습니다.<|im_end|><br><|im_start|>assistant<br><think><br><br></think><br><br></code> |
-* Loss: [<code>CachedMultipleNegativesRankingLoss</code>](https://sbert.net/docs/package_reference/cross_encoder/losses.html#cachedmultiplenegativesrankingloss) with these parameters:
-  ```json
-  {
-      "scale": 15,
-      "num_negatives": 61,
-      "activation_fn": "torch.nn.modules.activation.Sigmoid",
-      "mini_batch_size": 4
-  }
-  ```
-
-### Training Hyperparameters
-#### Non-Default Hyperparameters
-
-- `per_device_train_batch_size`: 1024
-- `per_device_eval_batch_size`: 32
-- `learning_rate`: 2e-05
-- `num_train_epochs`: 1
-- `warmup_ratio`: 0.05
-- `bf16`: True
-- `ddp_find_unused_parameters`: True
-- `ddp_timeout`: 7200
-- `batch_sampler`: no_duplicates
-
-#### 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`: 1024
-- `per_device_eval_batch_size`: 32
-- `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`: 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`: 1
-- `max_steps`: -1
-- `lr_scheduler_type`: linear
-- `lr_scheduler_kwargs`: {}
-- `warmup_ratio`: 0.05
-- `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`: None
-- `local_rank`: 0
-- `ddp_backend`: None
-- `tpu_num_cores`: None
-- `tpu_metrics_debug`: False
-- `debug`: []
-- `dataloader_drop_last`: True
-- `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`: True
-- `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`: 7200
-- `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`: {}
-
-</details>
+queries = [
+    "텔레픽스는 어떤 산업 분야에서 위성 데이터를 활용하나요?",
+    "국방 분야에 어떤 위성 서비스가 제공되나요?",
+    "텔레픽스의 기술 수준은 어느 정도인가요?",
+    "국방 분야에 어떤 위성 서비스가 제공되나요?",         # 부분/비관련 예시용
+    "텔레픽스는 어떤 산업 분야에서 위성 데이터를 활용하나요?" # 부분/관련 예시용
+]
 
-### Training Logs
-<details><summary>Click to expand</summary>
+documents = [
+    "텔레픽스는 해양, 자원, 농업 등 다양한 분야에서 위성 데이터를 분석하여 서비스를 제공합니다.",
+    "정찰 및 감시 목적의 위성 영상을 통해 국방 관련 정밀 분석 서비스를 제공합니다.",
+    "TelePIX의 광학 탑재체 및 AI 분석 기술은 Global standard를 상회하는 수준으로 평가받고 있습니다.",
+    "텔레픽스는 우주에서 수집한 정보를 분석하여 '우주 경제(Space Economy)'라는 새로운 가치를 창출하고 있습니다.",
+    "텔레픽스는 위성 영상 획득부터 분석, 서비스 제공까지 전 주기를 아우르는 솔루션을 제공합니다.",
+]
 
-| Epoch  | Step | Training Loss |
-|:------:|:----:|:-------------:|
-| 0.0034 | 1    | 1.2714        |
-| 0.0069 | 2    | 1.3902        |
-| 0.0103 | 3    | 1.3308        |
-| 0.0137 | 4    | 1.2726        |
-| 0.0172 | 5    | 1.2519        |
-| 0.0206 | 6    | 1.1254        |
-| 0.0241 | 7    | 0.9001        |
-| 0.0275 | 8    | 0.7529        |
-| 0.0309 | 9    | 0.9942        |
-| 0.0344 | 10   | 0.8769        |
-| 0.0378 | 11   | 0.6895        |
-| 0.0412 | 12   | 0.6813        |
-| 0.0447 | 13   | 0.6841        |
-| 0.0481 | 14   | 0.6025        |
-| 0.0515 | 15   | 0.619         |
-| 0.0550 | 16   | 0.6005        |
-| 0.0584 | 17   | 0.5917        |
-| 0.0619 | 18   | 0.5658        |
-| 0.0653 | 19   | 0.5571        |
-| 0.0687 | 20   | 0.5411        |
-| 0.0722 | 21   | 0.5374        |
-| 0.0756 | 22   | 0.5304        |
-| 0.0790 | 23   | 0.5103        |
-| 0.0825 | 24   | 0.5184        |
-| 0.0859 | 25   | 0.5036        |
-| 0.0893 | 26   | 0.5213        |
-| 0.0928 | 27   | 0.5399        |
-| 0.0962 | 28   | 0.5414        |
-| 0.0997 | 29   | 0.5177        |
-| 0.1031 | 30   | 0.5248        |
-| 0.1065 | 31   | 0.5196        |
-| 0.1100 | 32   | 0.499         |
-| 0.1134 | 33   | 0.514         |
-| 0.1168 | 34   | 0.5154        |
-| 0.1203 | 35   | 0.5114        |
-| 0.1237 | 36   | 0.508         |
-| 0.1271 | 37   | 0.5117        |
-| 0.1306 | 38   | 0.495         |
-| 0.1340 | 39   | 0.5304        |
-| 0.1375 | 40   | 0.4956        |
-| 0.1409 | 41   | 0.5274        |
-| 0.1443 | 42   | 0.5181        |
-| 0.1478 | 43   | 0.5103        |
-| 0.1512 | 44   | 0.5116        |
-| 0.1546 | 45   | 0.499         |
-| 0.1581 | 46   | 0.5072        |
-| 0.1615 | 47   | 0.5044        |
-| 0.1649 | 48   | 0.5071        |
-| 0.1684 | 49   | 0.5129        |
-| 0.1718 | 50   | 0.5095        |
-| 0.1753 | 51   | 0.5174        |
-| 0.1787 | 52   | 0.4748        |
-| 0.1821 | 53   | 0.4507        |
-| 0.1856 | 54   | 0.4927        |
-| 0.1890 | 55   | 0.452         |
-| 0.1924 | 56   | 0.4999        |
-| 0.1959 | 57   | 0.4744        |
-| 0.1993 | 58   | 0.4486        |
-| 0.2027 | 59   | 0.4725        |
-| 0.2062 | 60   | 0.4723        |
-| 0.2096 | 61   | 0.4747        |
-| 0.2131 | 62   | 0.4317        |
-| 0.2165 | 63   | 0.4668        |
-| 0.2199 | 64   | 0.453         |
-| 0.2234 | 65   | 0.4457        |
-| 0.2268 | 66   | 0.4179        |
-| 0.2302 | 67   | 0.4124        |
-| 0.2337 | 68   | 0.4454        |
-| 0.2371 | 69   | 0.4222        |
-| 0.2405 | 70   | 0.4151        |
-| 0.2440 | 71   | 0.4172        |
-| 0.2474 | 72   | 0.422         |
-| 0.2509 | 73   | 0.4088        |
-| 0.2543 | 74   | 0.4107        |
-| 0.2577 | 75   | 0.3977        |
-| 0.2612 | 76   | 0.4141        |
-| 0.2646 | 77   | 0.3991        |
-| 0.2680 | 78   | 0.3955        |
-| 0.2715 | 79   | 0.3864        |
-| 0.2749 | 80   | 0.4147        |
-| 0.2784 | 81   | 0.4084        |
-| 0.2818 | 82   | 0.4139        |
-| 0.2852 | 83   | 0.3999        |
-| 0.2887 | 84   | 0.4305        |
-| 0.2921 | 85   | 0.4188        |
-| 0.2955 | 86   | 0.4171        |
-| 0.2990 | 87   | 0.407         |
-| 0.3024 | 88   | 0.3871        |
-| 0.3058 | 89   | 0.389         |
-| 0.3093 | 90   | 0.3813        |
-| 0.3127 | 91   | 0.3814        |
-| 0.3162 | 92   | 0.3732        |
-| 0.3196 | 93   | 0.3899        |
-| 0.3230 | 94   | 0.3655        |
-| 0.3265 | 95   | 0.3638        |
-| 0.3299 | 96   | 0.3784        |
-| 0.3333 | 97   | 0.3729        |
-| 0.3368 | 98   | 0.3665        |
-| 0.3402 | 99   | 0.3579        |
-| 0.3436 | 100  | 0.3414        |
-| 0.3471 | 101  | 0.3304        |
-| 0.3505 | 102  | 0.347         |
-| 0.3540 | 103  | 0.3076        |
-| 0.3574 | 104  | 0.3111        |
-| 0.3608 | 105  | 0.3121        |
-| 0.3643 | 106  | 0.3272        |
-| 0.3677 | 107  | 0.3108        |
-| 0.3711 | 108  | 0.3092        |
-| 0.3746 | 109  | 0.2951        |
-| 0.3780 | 110  | 0.3195        |
-| 0.3814 | 111  | 0.2915        |
-| 0.3849 | 112  | 0.2855        |
-| 0.3883 | 113  | 0.2904        |
-| 0.3918 | 114  | 0.2873        |
-| 0.3952 | 115  | 0.273         |
-| 0.3986 | 116  | 0.2779        |
-| 0.4021 | 117  | 0.2939        |
-| 0.4055 | 118  | 0.276         |
-| 0.4089 | 119  | 0.2535        |
-| 0.4124 | 120  | 0.2774        |
-| 0.4158 | 121  | 0.2597        |
-| 0.4192 | 122  | 0.2541        |
-| 0.4227 | 123  | 0.2587        |
-| 0.4261 | 124  | 0.27          |
-| 0.4296 | 125  | 0.2724        |
-| 0.4330 | 126  | 0.2446        |
-| 0.4364 | 127  | 0.2747        |
-| 0.4399 | 128  | 0.268         |
-| 0.4433 | 129  | 0.2585        |
-| 0.4467 | 130  | 0.2652        |
-| 0.4502 | 131  | 0.2685        |
-| 0.4536 | 132  | 0.2565        |
-| 0.4570 | 133  | 0.2503        |
-| 0.4605 | 134  | 0.2634        |
-| 0.4639 | 135  | 0.2501        |
-| 0.4674 | 136  | 0.2479        |
-| 0.4708 | 137  | 0.2628        |
-| 0.4742 | 138  | 0.2505        |
-| 0.4777 | 139  | 0.2468        |
-| 0.4811 | 140  | 0.2365        |
-| 0.4845 | 141  | 0.2496        |
-| 0.4880 | 142  | 0.248         |
-| 0.4914 | 143  | 0.2604        |
-| 0.4948 | 144  | 0.2477        |
-| 0.4983 | 145  | 0.259         |
-| 0.5017 | 146  | 0.2556        |
-| 0.5052 | 147  | 0.2618        |
-| 0.5086 | 148  | 0.2583        |
-| 0.5120 | 149  | 0.2588        |
-| 0.5155 | 150  | 0.2468        |
-| 0.5189 | 151  | 0.2437        |
-| 0.5223 | 152  | 0.2595        |
-| 0.5258 | 153  | 0.2647        |
-| 0.5292 | 154  | 0.2699        |
-| 0.5326 | 155  | 0.2529        |
-| 0.5361 | 156  | 0.2339        |
-| 0.5395 | 157  | 0.2557        |
-| 0.5430 | 158  | 0.2402        |
-| 0.5464 | 159  | 0.2583        |
-| 0.5498 | 160  | 0.2688        |
-| 0.5533 | 161  | 0.2567        |
-| 0.5567 | 162  | 0.2702        |
-| 0.5601 | 163  | 0.2669        |
-| 0.5636 | 164  | 0.2699        |
-| 0.5670 | 165  | 0.2561        |
-| 0.5704 | 166  | 0.2406        |
-| 0.5739 | 167  | 0.2438        |
-| 0.5773 | 168  | 0.2523        |
-| 0.5808 | 169  | 0.2535        |
-| 0.5842 | 170  | 0.2533        |
-| 0.5876 | 171  | 0.2643        |
-| 0.5911 | 172  | 0.2684        |
-| 0.5945 | 173  | 0.2503        |
-| 0.5979 | 174  | 0.2735        |
-| 0.6014 | 175  | 0.2612        |
-| 0.6048 | 176  | 0.2721        |
-| 0.6082 | 177  | 0.2533        |
-| 0.6117 | 178  | 0.2704        |
-| 0.6151 | 179  | 0.2609        |
-| 0.6186 | 180  | 0.2605        |
-| 0.6220 | 181  | 0.2664        |
-| 0.6254 | 182  | 0.2516        |
-| 0.6289 | 183  | 0.2513        |
-| 0.6323 | 184  | 0.2439        |
-| 0.6357 | 185  | 0.258         |
-| 0.6392 | 186  | 0.2534        |
-| 0.6426 | 187  | 0.2638        |
-| 0.6460 | 188  | 0.2535        |
-| 0.6495 | 189  | 0.2481        |
-| 0.6529 | 190  | 0.264         |
-| 0.6564 | 191  | 0.2418        |
-| 0.6598 | 192  | 0.2326        |
-| 0.6632 | 193  | 0.2476        |
-| 0.6667 | 194  | 0.2271        |
-| 0.6701 | 195  | 0.229         |
-| 0.6735 | 196  | 0.2303        |
-| 0.6770 | 197  | 0.2272        |
-| 0.6804 | 198  | 0.2309        |
-| 0.6838 | 199  | 0.2159        |
-| 0.6873 | 200  | 0.2178        |
-| 0.6907 | 201  | 0.208         |
-| 0.6942 | 202  | 0.2257        |
-| 0.6976 | 203  | 0.2032        |
-| 0.7010 | 204  | 0.2047        |
-| 0.7045 | 205  | 0.2223        |
-| 0.7079 | 206  | 0.1964        |
-| 0.7113 | 207  | 0.1846        |
-| 0.7148 | 208  | 0.1899        |
-| 0.7182 | 209  | 0.1986        |
-| 0.7216 | 210  | 0.1898        |
-| 0.7251 | 211  | 0.1999        |
-| 0.7285 | 212  | 0.1754        |
-| 0.7320 | 213  | 0.1912        |
-| 0.7354 | 214  | 0.1702        |
-| 0.7388 | 215  | 0.17          |
-| 0.7423 | 216  | 0.1768        |
-| 0.7457 | 217  | 0.1647        |
-| 0.7491 | 218  | 0.1711        |
-| 0.7526 | 219  | 0.1507        |
-| 0.7560 | 220  | 0.1657        |
-| 0.7595 | 221  | 0.1498        |
-| 0.7629 | 222  | 0.1557        |
-| 0.7663 | 223  | 0.1651        |
-| 0.7698 | 224  | 0.1446        |
-| 0.7732 | 225  | 0.1519        |
-| 0.7766 | 226  | 0.1453        |
-| 0.7801 | 227  | 0.1561        |
-| 0.7835 | 228  | 0.1557        |
-| 0.7869 | 229  | 0.1493        |
-| 0.7904 | 230  | 0.1476        |
-| 0.7938 | 231  | 0.1453        |
-| 0.7973 | 232  | 0.1312        |
-| 0.8007 | 233  | 0.1531        |
-| 0.8041 | 234  | 0.1498        |
-| 0.8076 | 235  | 0.134         |
-| 0.8110 | 236  | 0.1361        |
-| 0.8144 | 237  | 0.1461        |
-| 0.8179 | 238  | 0.148         |
-| 0.8213 | 239  | 0.1465        |
-| 0.8247 | 240  | 0.1452        |
-| 0.8282 | 241  | 0.1399        |
-| 0.8316 | 242  | 0.1291        |
-| 0.8351 | 243  | 0.1354        |
-| 0.8385 | 244  | 0.1719        |
-| 0.8419 | 245  | 0.1555        |
-| 0.8454 | 246  | 0.1472        |
-| 0.8488 | 247  | 0.1516        |
-| 0.8522 | 248  | 0.1579        |
-| 0.8557 | 249  | 0.161         |
-| 0.8591 | 250  | 0.1661        |
-| 0.8625 | 251  | 0.155         |
-| 0.8660 | 252  | 0.1706        |
-| 0.8694 | 253  | 0.1527        |
-| 0.8729 | 254  | 0.1695        |
-| 0.8763 | 255  | 0.1904        |
-| 0.8797 | 256  | 0.186         |
-| 0.8832 | 257  | 0.1723        |
-| 0.8866 | 258  | 0.1881        |
-| 0.8900 | 259  | 0.1915        |
-| 0.8935 | 260  | 0.1969        |
-| 0.8969 | 261  | 0.1967        |
-| 0.9003 | 262  | 0.2038        |
-| 0.9038 | 263  | 0.1917        |
-| 0.9072 | 264  | 0.19          |
-| 0.9107 | 265  | 0.2161        |
-| 0.9141 | 266  | 0.222         |
-| 0.9175 | 267  | 0.2361        |
-| 0.9210 | 268  | 0.2538        |
-| 0.9244 | 269  | 0.2408        |
-| 0.9278 | 270  | 0.2372        |
-| 0.9313 | 271  | 0.2292        |
-| 0.9347 | 272  | 0.238         |
-| 0.9381 | 273  | 0.2243        |
-| 0.9416 | 274  | 0.2443        |
-| 0.9450 | 275  | 0.2435        |
-| 0.9485 | 276  | 0.2476        |
-| 0.9519 | 277  | 0.2259        |
-| 0.9553 | 278  | 0.2327        |
-| 0.9588 | 279  | 0.2345        |
-| 0.9622 | 280  | 0.2413        |
+pairs = [
+    [format_queries(query, task), format_document(doc)]
+    for query, doc in zip(queries, documents)
+]
 
-</details>
+scores = model.predict(pairs)
+print(scores.tolist())
+```
 
-### Framework Versions
-- Python: 3.11.12
-- Sentence Transformers: 5.0.0
-- Transformers: 4.53.1
-- PyTorch: 2.8.0+cu128
-- Accelerate: 1.5.2
-- Datasets: 2.21.0
-- Tokenizers: 0.21.1
+## License
+The PIXIE-Spell-Reranker-Preview-0.6B model is licensed under Apache License 2.0.
 
 ## Citation
-
-### BibTeX
-
-#### 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",
+```
+@software{TelePIX-PIXIE-Spell-Reranker-Preview-0.6B,
+  title={PIXIE-Spell-Reranker-Preview-0.6B},
+  author={TelePIX AI Research Team and Bongmin Kim},
+  year={2025},
+  url={https://huggingface.co/telepix/PIXIE-Spell-Reranker-Preview-0.6B}
 }
 ```
 
-<!--
-## Glossary
-
-*Clearly define terms in order to be accessible across audiences.*
--->
-
-<!--
-## Model Card Authors
-
-*Lists the people who create the model card, providing recognition and accountability for the detailed work that goes into its construction.*
--->
-
-<!--
-## Model Card Contact
+## Contact
 
-*Provides a way for people who have updates to the Model Card, suggestions, or questions, to contact the Model Card authors.*
--->
+If you have any suggestions or questions about the PIXIE, please reach out to the authors at bmkim@telepix.net.