| --- |
| library_name: transformers |
| pipeline_tag: text-classification |
| tags: |
| - email |
| - text-classification |
| language: |
| - en |
| datasets: |
| - FlowRank/labeled_emails |
| base_model: distilbert-base-uncased |
| license: mit |
| metrics: |
| - accuracy |
| model-index: |
| - name: mailSort |
| results: |
| - task: |
| type: text-classification |
| dataset: |
| name: FlowRank/labeled_emails |
| type: FlowRank/labeled_emails |
| metrics: |
| - type: accuracy |
| value: 0.9601 |
| widget: |
| - text: "Subject: Insurance claim\n\nBody: Hello, I need to update my policy urgently before the deadline." |
| --- |
| |
| ## mailSort |
|
|
| Classifieur d’e-mails **20 classes** (catégorie + urgence), basé sur DistilBERT. |
|
|
| Entraîné sur [`FlowRank/labeled_emails`](https://huggingface.co/datasets/FlowRank/labeled_emails) (révision `main`, ~19,8k exemples). |
| **Accuracy test : 96,01 %** (1 900 / 1 979). |
|
|
| | Version | Labels | Révision dataset | Révision modèle | |
| |---------|--------|------------------|-----------------| |
| | **v2** (courante) | 20 | `main` ou `v2.0` | `main` ou `v2.0` | |
| | v1 | 10 | `v1.0` | `v1.0` | |
|
|
| ### Labels (v2) |
|
|
| Chaque catégorie est déclinée en `_urgent` / `_not_urgent` : `family`, `finance`, `games`, `human resources`, `medical`, `pets`, `school`, `software engineering`, `sport`, `work/airbus`. |
|
|
| ## Inférence |
|
|
| ```python |
| from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipeline |
| |
| tok = AutoTokenizer.from_pretrained("FlowRank/mailSort", subfolder="model") |
| model = AutoModelForSequenceClassification.from_pretrained("FlowRank/mailSort", subfolder="model") |
| clf = pipeline("text-classification", model=model, tokenizer=tok, truncation=True) |
| |
| text = "Subject: Insurance claim\n\nBody: I need to update my policy urgently." |
| print(clf(text)) |
| ``` |
|
|
| ## Développement local |
|
|
| Repo Python minimal pour entraîner, évaluer et publier. Le script principal est `mailsort.train`. |
|
|
| Par défaut, le code charge la **dernière version du dataset** (`FlowRank/labeled_emails@main`). |
|
|
| ### Installation |
|
|
| ```bash |
| uv sync |
| ``` |
|
|
| ### Entraîner + évaluer |
|
|
| ```bash |
| uv run python -m mailsort.train \ |
| --model-name distilbert-base-uncased \ |
| --hub-model-id FlowRank/mailSort \ |
| --num-train-epochs 2 |
| ``` |
|
|
| Pour épingler une version précise du dataset : |
|
|
| ```bash |
| uv run python -m mailsort.train --dataset-revision v1.0 |
| ``` |
|
|
| ### Évaluer le modèle publié |
|
|
| ```bash |
| uv run python -m mailsort.eval --split test |
| ``` |
|
|
| Évalue par défaut `FlowRank/mailSort` (sous-dossier `model/`) sur `FlowRank/labeled_emails@main`. |
|
|
| ### Publier via Git (README + `model/`) |
|
|
| ```bash |
| uv run python -m mailsort.prepare_model --outputs-dir outputs --model-dir model |
| git add README.md model |
| git commit -m "chore(model): update published artifacts" |
| git push |
| ``` |
|
|
| Le push Hub automatique via `Trainer` nécessite `HF_TOKEN` : |
|
|
| ```bash |
| export HF_TOKEN="..." |
| uv run python -m mailsort.train --hub-model-id FlowRank/mailSort |
| ``` |
|
|
