Metasebiya-21/geez-amharic-translator

--- title: English → French Translator emoji: 🇫🇷 colorFrom: blue colorTo: green sdk: gradio sdk_version: 4.44.1 app_file: app.py pinned: false python_version: "3.10" --- # English → French Neural Machine Translation BiLSTM encoder–decoder with attention, trained on an English–French parallel corpus. **Notebooks:** `en_fr_baseline.ipynb` (EN→FR) · `gez_to_amh_blstm.ipynb` (Ge'ez→Amharic) --- ## Model artifacts Weights are **not** in this Git repo (~400 MB). Host them on **Hugging Face Hub**: ``` your-username/en-fr-translator/ ├── best_model.keras ├── src_tokenizer.pkl ├── tgt_tokenizer.pkl └── meta.json ``` ### Upload artifacts to Hugging Face ```bash pip install huggingface_hub huggingface-cli login python scripts/upload_artifacts_to_hf.py YOUR_USERNAME/en-fr-translator ``` --- ## Deploy on Hugging Face Spaces (recommended) 1. Create a new Space at huggingface.co - **SDK:** Gradio - **Hardware:** CPU Basic (free) is enough for inference 2. Push this repo (or connect GitHub) — entry point is **`app.py`** 3. In Space **Settings → Variables and secrets**, add: | Name | Value | |------|--------| | `HF_MODEL_REPO` | `YOUR_USERNAME/en-fr-translator` | 4. The Space downloads weights from Hub on first run, then caches them. The YAML block at the top of this README configures the Space automatically when this file is `README.md` in the Space repo. --- ## Run locally (Gradio) 1. Place artifacts in `en_fr_artifacts/`, **or** set `HF_MODEL_REPO=YOUR_USERNAME/en-fr-translator` 2. Install and run: ```bash pip install -r requirements.txt python app.py ``` --- ## Run locally (Streamlit, optional) >>>>>> parent of 947660f (Fix Streamlit Cloud deps: remove uv.lock, use tensorflow for Linux) --- ## Project layout ``` ├── app.py # Hugging Face Space (Gradio) ├── main.py # Optional Streamlit UI ├── inference.py # Load artifacts, beam/greedy decode ├── requirements.txt # HF Space dependencies ├── en_fr_baseline.ip …