Replace auto-generated card with the full model card

README.md

@@ -1,525 +1,155 @@
 ---
+language:
+  - mfe
+  - en
+  - fr
+license: mit
+library_name: sentence-transformers
+pipeline_tag: sentence-similarity
 tags:
-- sentence-transformers
-- sentence-similarity
-- feature-extraction
-- dense
-- generated_from_trainer
-- dataset_size:24100
-- loss:MatryoshkaLoss
-- loss:CachedMultipleNegativesRankingLoss
-- loss:MultipleNegativesRankingLoss
+  - sentence-transformers
+  - sentence-similarity
+  - feature-extraction
+  - mauritian-creole
+  - kreol-morisien
+  - matryoshka
 base_model: intfloat/multilingual-e5-base
-widget:
-- source_sentence: 'Li reponn:'
-  sentences:
-  - 'Jesus answered, in the Scriptures, it is wriiten:'
-  - 'Il met les deux mains l’une dans l’autre, puis il leur dit :'
-  - 'Jesus also answered, the Scriptures also say:'
-  - 'Mary said:'
-  - 'Jésus lui répondit :'
-  - 'They ran thus:'
-- source_sentence: Kouma li kapav fer tousa?
-  sentences:
-  - What do you do exactly?
-  - Il croit qu’il pourra tout faire tout seul.
-  - What can they do about it?
-  - Comment peut-il faire tout cela?
-  - How to explain all this?
-  - How to make him understand?
-- source_sentence: Tape e laport pou ouver pou twa.
-  sentences:
-  - Choose times and places to collect when people are more likely to be able to remember.
-  - Mettez un tablier avant de travailler dans la cuisine.
-  - The parcel bursts open.
-  - And the door will be opened for everyone who knocks.
-  - Frappez et la porte sera ouverte pour vous.
-  - Just turn right and go all the way to end of this terminal and you'll be in Terminal
-    3.
-- source_sentence: Zezi dir zot, komie manze ena?
-  sentences:
-  - Jésus a demandé, lorsque le propriétaire de cette vigne viendra, que croyez-vous
-    qu'il va faire à ces vignerons?
-  - 'Jesus answered:'
-  - Jesus replied, surely you know that the Scriptures say,
-  - Then Jesus asked them, but who do you say I am?
-  - Jesus replied, do you see these buildings?
-  - Jesus asked them how much food they had.
-- source_sentence: Pa bizin per.
-  sentences:
-  - N'ayez pas peur.
-  - Don't be afraid of people.
-  - They're not frightened of being wrong.
-  - Don't trouble me.
-  - Don't let anyone fool you.
-  - It does not hurt much.
-pipeline_tag: sentence-similarity
-library_name: sentence-transformers
-metrics:
-- cosine_accuracy@1
-- cosine_accuracy@3
-- cosine_accuracy@5
-- cosine_accuracy@10
-- cosine_precision@1
-- cosine_precision@3
-- cosine_precision@5
-- cosine_precision@10
-- cosine_recall@1
-- cosine_recall@3
-- cosine_recall@5
-- cosine_recall@10
-- cosine_ndcg@10
-- cosine_mrr@10
-- cosine_map@100
-model-index:
-- name: SentenceTransformer based on intfloat/multilingual-e5-base
-  results:
-  - task:
-      type: information-retrieval
-      name: Information Retrieval
-    dataset:
-      name: morisienmt dev eng
-      type: morisienmt-dev-eng
-    metrics:
-    - type: cosine_accuracy@1
-      value: 0.96
-      name: Cosine Accuracy@1
-    - type: cosine_accuracy@3
-      value: 0.98
-      name: Cosine Accuracy@3
-    - type: cosine_accuracy@5
-      value: 0.98
-      name: Cosine Accuracy@5
-    - type: cosine_accuracy@10
-      value: 0.982
-      name: Cosine Accuracy@10
-    - type: cosine_precision@1
-      value: 0.96
-      name: Cosine Precision@1
-    - type: cosine_precision@3
-      value: 0.3266666666666666
-      name: Cosine Precision@3
-    - type: cosine_precision@5
-      value: 0.196
-      name: Cosine Precision@5
-    - type: cosine_precision@10
-      value: 0.09820000000000001
-      name: Cosine Precision@10
-    - type: cosine_recall@1
-      value: 0.96
-      name: Cosine Recall@1
-    - type: cosine_recall@3
-      value: 0.98
-      name: Cosine Recall@3
-    - type: cosine_recall@5
-      value: 0.98
-      name: Cosine Recall@5
-    - type: cosine_recall@10
-      value: 0.982
-      name: Cosine Recall@10
-    - type: cosine_ndcg@10
-      value: 0.9728072904313593
-      name: Cosine Ndcg@10
-    - type: cosine_mrr@10
-      value: 0.9696666666666667
-      name: Cosine Mrr@10
-    - type: cosine_map@100
-      value: 0.9699554965211892
-      name: Cosine Map@100
+datasets:
+  - prajdabre/MorisienMT
+  - jhu-clsp/kreyol-mt
 ---
 
-# SentenceTransformer based on intfloat/multilingual-e5-base
-
-This is a [sentence-transformers](https://www.SBERT.net) model finetuned from [intfloat/multilingual-e5-base](https://huggingface.co/intfloat/multilingual-e5-base). It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, classification, clustering, and more.
-
-## Model Details
-
-### Model Description
-- **Model Type:** Sentence Transformer
-- **Base model:** [intfloat/multilingual-e5-base](https://huggingface.co/intfloat/multilingual-e5-base) <!-- at revision d128750597153bb5987e10b1c3493a34e5a4502a -->
-- **Maximum Sequence Length:** 512 tokens
-- **Output Dimensionality:** 768 dimensions
-- **Similarity Function:** Cosine Similarity
-- **Supported Modality:** Text
-<!-- - **Training Dataset:** Unknown -->
-<!-- - **Language:** Unknown -->
-<!-- - **License:** Unknown -->
-
-### Model Sources
+# morisien-embed
 
-- **Documentation:** [Sentence Transformers Documentation](https://sbert.net)
-- **Repository:** [Sentence Transformers on GitHub](https://github.com/huggingface/sentence-transformers)
-- **Hugging Face:** [Sentence Transformers on Hugging Face](https://huggingface.co/models?library=sentence-transformers)
+To our knowledge, the first dedicated text embedding model for **Mauritian Creole (Kreol Morisien,
+`mfe`)** — the home language of roughly 90% of Mauritius (2022 census).
 
-### Full Model Architecture
+Fine-tuned from [multilingual-e5-base](https://huggingface.co/intfloat/multilingual-e5-base) on
+effectively all publicly available Creole↔{English, French} parallel data, it outperforms every
+general multilingual embedding model we evaluated — including
+[LaBSE](https://huggingface.co/sentence-transformers/LaBSE), the strongest of them on this task — on
+held-out Creole retrieval in both directions.
 
-```
-SentenceTransformer(
-  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'XLMRobertaModel'})
-  (1): Pooling({'embedding_dimension': 768, 'pooling_mode': 'mean', 'include_prompt': True})
-  (2): Normalize({})
-)
-```
+Use it for semantic search, retrieval, RAG, bitext mining, or clustering over Kreol Morisien text.
 
 ## 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("Singaraj/morisien-embed")
-# Run inference
-queries = [
-    'Pa bizin per.',
-]
-documents = [
-    "N'ayez pas peur.",
-    "Don't trouble me.",
-    "Don't let anyone fool you.",
-]
-query_embeddings = model.encode_query(queries)
-document_embeddings = model.encode_document(documents)
-print(query_embeddings.shape, document_embeddings.shape)
-# [1, 768] [3, 768]
-
-# Get the similarity scores for the embeddings
-similarities = model.similarity(query_embeddings, document_embeddings)
-print(similarities)
-# tensor([[0.7254, 0.5502, 0.4745]])
-```
-<!--
-### 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.*
--->
-
-## Evaluation
-
-### Metrics
-
-#### Information Retrieval
-
-* Dataset: `morisienmt-dev-eng`
-* Evaluated with [<code>InformationRetrievalEvaluator</code>](https://sbert.net/docs/package_reference/sentence_transformer/evaluation.html#sentence_transformers.sentence_transformer.evaluation.InformationRetrievalEvaluator)
-
-| Metric              | Value      |
-|:--------------------|:-----------|
-| cosine_accuracy@1   | 0.96       |
-| cosine_accuracy@3   | 0.98       |
-| cosine_accuracy@5   | 0.98       |
-| cosine_accuracy@10  | 0.982      |
-| cosine_precision@1  | 0.96       |
-| cosine_precision@3  | 0.3267     |
-| cosine_precision@5  | 0.196      |
-| cosine_precision@10 | 0.0982     |
-| cosine_recall@1     | 0.96       |
-| cosine_recall@3     | 0.98       |
-| cosine_recall@5     | 0.98       |
-| cosine_recall@10    | 0.982      |
-| **cosine_ndcg@10**  | **0.9728** |
-| cosine_mrr@10       | 0.9697     |
-| cosine_map@100      | 0.97       |
-
-<!--
-## 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
+creole = ["Mo pe al bazar aster.", "Bann zanfan pe zwe dan lakour."]
+english = ["I am going to the market now.", "The children are playing in the yard."]
 
-### Training Dataset
-
-#### Unnamed Dataset
-
-* Size: 24,100 training samples
-* Columns: <code>anchor</code>, <code>positive</code>, <code>negative_1</code>, <code>negative_2</code>, <code>negative_3</code>, <code>negative_4</code>, and <code>negative_5</code>
-* Approximate statistics based on the first 100 samples:
-  |          | anchor                                                                            | positive                                                                          | negative_1                                                                        | negative_2                                                                        | negative_3                                                                        | negative_4                                                                        | negative_5                                                                        |
-  |:---------|:----------------------------------------------------------------------------------|:----------------------------------------------------------------------------------|:----------------------------------------------------------------------------------|:----------------------------------------------------------------------------------|:----------------------------------------------------------------------------------|:----------------------------------------------------------------------------------|:----------------------------------------------------------------------------------|
-  | type     | string                                                                            | string                                                                            | string                                                                            | string                                                                            | string                                                                            | string                                                                            | string                                                                            |
-  | modality | text                                                                              | text                                                                              | text                                                                              | text                                                                              | text                                                                              | text                                                                              | text                                                                              |
-  | details  | <ul><li>min: 5 tokens</li><li>mean: 14.52 tokens</li><li>max: 38 tokens</li></ul> | <ul><li>min: 5 tokens</li><li>mean: 14.02 tokens</li><li>max: 34 tokens</li></ul> | <ul><li>min: 5 tokens</li><li>mean: 15.32 tokens</li><li>max: 40 tokens</li></ul> | <ul><li>min: 4 tokens</li><li>mean: 14.96 tokens</li><li>max: 39 tokens</li></ul> | <ul><li>min: 4 tokens</li><li>mean: 14.99 tokens</li><li>max: 39 tokens</li></ul> | <ul><li>min: 5 tokens</li><li>mean: 14.47 tokens</li><li>max: 30 tokens</li></ul> | <ul><li>min: 5 tokens</li><li>mean: 14.65 tokens</li><li>max: 38 tokens</li></ul> |
-* Samples:
-  | anchor                                                           | positive                                                                                      | negative_1                                                              | negative_2                                                                                                            | negative_3                                                                                        | negative_4                                                              | negative_5                                           |
-  |:-----------------------------------------------------------------|:----------------------------------------------------------------------------------------------|:------------------------------------------------------------------------|:----------------------------------------------------------------------------------------------------------------------|:--------------------------------------------------------------------------------------------------|:------------------------------------------------------------------------|:-----------------------------------------------------|
-  | <code>pou determinn lepeser ek mouvman so lakok glason e.</code> | <code>to determine the thickness and dynamics of the moon's ice shell and.</code>             | <code>You can tell what a tree is like by the fruit it produces.</code> | <code>The crisis’ vigour and tempo can also be determined by the internal re-organization of the ruling block.</code> | <code>Find out from the storyteller or others what rules exist that determine performance.</code> | <code>Vous pouvez dire le temps qu'il fera en regardant le ciel.</code> | <code>How to Recognize and Collect Folktales.</code> |
-  | <code>pou determinn lepeser ek mouvman so lakok glason e.</code> | <code>pour déterminer l'épaisseur et la dynamique de la couche de glace de la lune et.</code> | <code>You can tell what a tree is like by the fruit it produces.</code> | <code>The crisis’ vigour and tempo can also be determined by the internal re-organization of the ruling block.</code> | <code>Find out from the storyteller or others what rules exist that determine performance.</code> | <code>Vous pouvez dire le temps qu'il fera en regardant le ciel.</code> | <code>How to Recognize and Collect Folktales.</code> |
-  | <code>karakteriz zeolozi Europa en detay.</code>                 | <code>characterize the surface geology of Europa in detail.</code>                            | <code>Folklore thus take in local and national history.</code>          | <code>Give detailed description of the event.</code>                                                                  | <code>The Irish book is the model for this book.</code>                                           | <code>Prophecy of the downfall of the British colonizers.</code>        | <code>It is called Beasts of England.</code>         |
-* Loss: [<code>MatryoshkaLoss</code>](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#matryoshkaloss) with these parameters:
-  ```json
-  {
-      "loss": "CachedMultipleNegativesRankingLoss",
-      "matryoshka_dims": [
-          768,
-          512,
-          256,
-          128,
-          64
-      ],
-      "matryoshka_weights": [
-          1,
-          1,
-          1,
-          1,
-          1
-      ],
-      "n_dims_per_step": -1
-  }
-  ```
-
-### Training Hyperparameters
-#### Non-Default Hyperparameters
-
-- `per_device_train_batch_size`: 128
-- `learning_rate`: 2e-05
-- `warmup_steps`: 0.1
-- `fp16`: True
-- `batch_sampler`: no_duplicates
-
-#### All Hyperparameters
-<details><summary>Click to expand</summary>
-
-- `do_predict`: False
-- `prediction_loss_only`: True
-- `per_device_train_batch_size`: 128
-- `per_device_eval_batch_size`: 8
-- `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`: 3.0
-- `max_steps`: -1
-- `lr_scheduler_type`: linear
-- `lr_scheduler_kwargs`: None
-- `warmup_ratio`: None
-- `warmup_steps`: 0.1
-- `log_level`: passive
-- `log_level_replica`: warning
-- `log_on_each_node`: True
-- `logging_nan_inf_filter`: True
-- `enable_jit_checkpoint`: False
-- `save_on_each_node`: False
-- `save_only_model`: False
-- `restore_callback_states_from_checkpoint`: False
-- `use_cpu`: False
-- `seed`: 42
-- `data_seed`: None
-- `bf16`: False
-- `fp16`: True
-- `bf16_full_eval`: False
-- `fp16_full_eval`: False
-- `tf32`: None
-- `local_rank`: -1
-- `ddp_backend`: None
-- `debug`: []
-- `dataloader_drop_last`: False
-- `dataloader_num_workers`: 0
-- `dataloader_prefetch_factor`: None
-- `disable_tqdm`: False
-- `remove_unused_columns`: True
-- `label_names`: None
-- `load_best_model_at_end`: False
-- `ignore_data_skip`: False
-- `fsdp`: []
-- `fsdp_config`: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': 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
-- `deepspeed`: None
-- `label_smoothing_factor`: 0.0
-- `optim`: adamw_torch_fused
-- `optim_args`: None
-- `group_by_length`: False
-- `length_column_name`: length
-- `project`: huggingface
-- `trackio_space_id`: trackio
-- `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
-- `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_for_metrics`: []
-- `eval_do_concat_batches`: True
-- `auto_find_batch_size`: False
-- `full_determinism`: False
-- `ddp_timeout`: 1800
-- `torch_compile`: False
-- `torch_compile_backend`: None
-- `torch_compile_mode`: None
-- `include_num_input_tokens_seen`: no
-- `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`: True
-- `use_cache`: False
-- `prompts`: None
-- `batch_sampler`: no_duplicates
-- `multi_dataset_batch_sampler`: proportional
-- `router_mapping`: {}
-- `learning_rate_mapping`: {}
-
-</details>
-
-### Training Logs
-| Epoch  | Step | Training Loss | morisienmt-dev-eng_cosine_ndcg@10 |
-|:------:|:----:|:-------------:|:---------------------------------:|
-| 0.2646 | 50   | 19.3371       | -                                 |
-| 0.5291 | 100  | 7.6897        | -                                 |
-| 0.7937 | 150  | 5.7943        | -                                 |
-| 1.0    | 189  | -             | 0.9698                            |
-| 1.0582 | 200  | 5.0894        | -                                 |
-| 1.3228 | 250  | 4.2876        | -                                 |
-| 1.5873 | 300  | 3.9644        | -                                 |
-| 1.8519 | 350  | 3.8261        | -                                 |
-| 2.0    | 378  | -             | 0.9724                            |
-| 2.1164 | 400  | 3.4445        | -                                 |
-| 2.3810 | 450  | 3.2138        | -                                 |
-| 2.6455 | 500  | 3.0999        | -                                 |
-| 2.9101 | 550  | 3.0612        | -                                 |
-| 3.0    | 567  | -             | 0.9728                            |
-
-
-### Training Time
-- **Training**: 30.6 minutes
-- **Evaluation**: 1.9 seconds
-- **Total**: 30.6 minutes
+similarity = model.similarity(model.encode(creole), model.encode(english))
+```
 
-### Framework Versions
-- Python: 3.12.13
-- Sentence Transformers: 5.7.0
-- Transformers: 5.0.0
-- PyTorch: 2.10.0+cu128
-- Accelerate: 1.14.0
-- Datasets: 5.0.1
-- Tokenizers: 0.22.2
+Trained with Matryoshka loss, so embeddings can be truncated for faster search at a small,
+measured accuracy cost (ndcg@10 on the benchmark below: 0.9591 at 256 dims, 0.9531 at 128):
 
-## Additional Resources
+```python
+model = SentenceTransformer("Singaraj/morisien-embed", truncate_dim=256)
+```
 
-- [Training and Finetuning Embedding Models with Sentence Transformers](https://huggingface.co/blog/train-sentence-transformers): the end-to-end guide for training or finetuning Sentence Transformer models.
-- [Introduction to Matryoshka Embedding Models](https://huggingface.co/blog/matryoshka): variable-size embeddings that can be truncated with minimal quality loss.
-- [Binary and Scalar Embedding Quantization for Significantly Faster & Cheaper Retrieval](https://huggingface.co/blog/embedding-quantization): post-training compression of embedding vectors.
-- [Multimodal Embedding & Reranker Models with Sentence Transformers](https://huggingface.co/blog/multimodal-sentence-transformers): use text, image, audio, and video models through the same API.
-- [Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers](https://huggingface.co/blog/train-multimodal-sentence-transformers): train multimodal embedding models, with a Visual Document Retrieval walkthrough.
+No prompt/prefix is required.
+
+## Results
+
+Creole→English retrieval on the held-out [MorisienMT](https://huggingface.co/datasets/prajdabre/MorisienMT)
+test split (1,000 queries, leak-free against training data — enforced in the data pipeline by exact
+matching and by a punctuation-, case- and accent-insensitive check):
+
+| Model | Params | ndcg@10 | accuracy@1 |
+|---|---|---|---|
+| paraphrase-multilingual-MiniLM-L12-v2 | 118M | 0.16 | 0.10 |
+| BAAI/bge-m3 | 568M | 0.46 | 0.36 |
+| intfloat/multilingual-e5-small | 118M | 0.54 | 0.42 |
+| intfloat/multilingual-e5-base | 278M | 0.64 | 0.53 |
+| intfloat/multilingual-e5-large | 560M | 0.73 | 0.65 |
+| sentence-transformers/LaBSE | 470M | 0.94 | 0.91 |
+| **morisien-embed** | **278M** | **0.9655** | **0.9440** |
+
+Creole→French, same protocol:
+
+| Model | ndcg@10 | accuracy@1 |
+|---|---|---|
+| sentence-transformers/LaBSE | 0.9475 | 0.9130 |
+| **morisien-embed** | **0.9751** | **0.9530** |
+
+Generalization to an independent domain — [FLORES+](https://huggingface.co/datasets/openlanguagedata/flores_plus)
+`mfe` devtest (1,012 professionally translated sentences from Wikinews, Wikijunior and Wikivoyage,
+zero overlap with training data):
+
+| Model | ndcg@10 | accuracy@1 |
+|---|---|---|
+| sentence-transformers/LaBSE | 0.9996 | 0.9990 |
+| **morisien-embed** | **1.0000** | **1.0000** |
+
+Both models sit at the ceiling of this benchmark — FLORES+ sentences are long and distinctive, so
+1,012-way retrieval saturates. Read this as evidence of zero out-of-domain degradation, not as a
+margin over LaBSE.
+
+Training was repeated with three random seeds; Creole→English test ndcg@10 across seeds:
+**0.9653 ± 0.0002** (accuracy@1 **0.9433 ± 0.0006**). The released checkpoint is seed 42, designated
+before results were seen.
+
+Every number is reproducible from the [training repository](https://github.com/LK-maker-007/morisien-embed).
+
+## Training
+
+- **Data:** 35,064 unique, leak-free Creole↔{English, French} pairs — effectively all publicly
+  available Mauritian Creole parallel text — merged from
+  [MorisienMT](https://huggingface.co/datasets/prajdabre/MorisienMT) (CC) and
+  [Kreyòl-MT](https://huggingface.co/datasets/jhu-clsp/kreyol-mt) (mixed licenses; used for training
+  only, not redistributed). Every MorisienMT dev/test sentence is removed from training by exact
+  matching and by a punctuation-, case- and accent-insensitive check.
+- **Recipe:** hard-negative mining with positive-aware false-negative filtering
+  (`mine_hard_negatives`: 5 negatives/anchor, `range_min=10`, `relative_margin=0.05`). The margin
+  filter is strict: 24,100 of the 35,064 pairs survived with a full negative set, and the released
+  checkpoint's contrastive stage trained on those 24,100 tuples (the stage-1 mining model itself was
+  trained on all 35,064). Contrastive training uses `CachedMultipleNegativesRankingLoss` (batch 128,
+  767 in-batch negatives per anchor) wrapped in `MatryoshkaLoss` (dims 768/512/256/128/64). 3 epochs,
+  lr 2e-5, warmup 10%, fp16, seed 42, single T4 GPU (~30 min).
+- **Base model:** [intfloat/multilingual-e5-base](https://huggingface.co/intfloat/multilingual-e5-base)
+  (278M parameters, MIT).
+
+## Limitations
+
+- **Not native-perfect.** Accuracy@1 around 0.944 means roughly one query in eighteen ranks
+  a wrong translation first. Strong, but below a human bilingual speaker.
+- **Register skew.** The available Creole data over-represents religious text, politics, and
+  literature; highly informal or technical registers are less covered.
+- **Small evaluation universe.** Retrieval is measured over ~1,000-passage corpora — standard for
+  bitext benchmarks, but absolute scores would be lower against web-scale corpora.
+- **One distribution family.** MorisienMT and Kreyòl-MT overlap heavily, and the only fully
+  independent evaluation domain for `mfe` (FLORES+) is saturated at this corpus size — so the margin
+  over LaBSE is demonstrated in-domain only.
+- **Haitian Creole proximity.** Like every multilingual embedder we tested, the model embeds Haitian
+  Creole close to Mauritian Creole: with same-meaning Haitian sentences injected into a FLORES-based
+  corpus, mfe→eng accuracy@1 drops from 1.00 to 0.71 (the Haitian twin outranks the English
+  translation). The fine-tune still discriminates the two creoles better than LaBSE does on the same
+  trap (306/400 correct vs LaBSE's 170/400), and wrong-meaning Haitian text is never confused — but
+  mixed mfe/hat corpora will degrade retrieval.
+- **Case sensitivity.** ALL-CAPS text embeds measurably differently from its lower-case form
+  (cosine ≈ 0.81 to the same sentence); caps-heavy text retrieves worse.
+- **Protocol note.** During recipe development the held-out test score was printed at the end of each
+  training run, so recipe selection had test visibility; an independent audit bounded the resulting
+  optimism at ≤ ~0.01 ndcg. The 3-seed replication was run after the recipe was frozen.
+- **Orthographic variation.** Training data mixes pre- and post-2011 (Lortograf Kreol Morisien)
+  spellings; performance on older orthography is untested.
 
 ## Citation
 
-### BibTeX
+If you use this model, please cite the data sources it builds on:
+[MorisienMT](https://arxiv.org/abs/2206.02421) (Dabre & Sukhoo, 2022) and
+[Kreyòl-MT](https://arxiv.org/abs/2405.05376) (Robinson et al., NAACL 2024).
 
-#### 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",
+@misc{morisien-embed,
+  author = {Singaraj B},
+  title = {morisien-embed: a dedicated text embedding model for Mauritian Creole},
+  year = {2026},
+  url = {https://huggingface.co/Singaraj/morisien-embed}
 }
 ```
-
-#### MatryoshkaLoss
-```bibtex
-@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}
-}
-```
-
-#### CachedMultipleNegativesRankingLoss
-```bibtex
-@misc{gao2021scaling,
-    title={Scaling Deep Contrastive Learning Batch Size under Memory Limited Setup},
-    author={Luyu Gao and Yunyi Zhang and Jiawei Han and Jamie Callan},
-    year={2021},
-    eprint={2101.06983},
-    archivePrefix={arXiv},
-    primaryClass={cs.LG}
-}
-```
-
-#### MultipleNegativesRankingLoss
-```bibtex
-@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},
-}
-```
-
-<!--
-## 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
-
-*Provides a way for people who have updates to the Model Card, suggestions, or questions, to contact the Model Card authors.*
--->