# Hugging Face Space Training

This project keeps internal orchestration as an external orchestration microservice. Model training
is handled by a separate Hugging Face Docker Space:

```text
mortadhabbb/train_model_chatbot
```

The local Django chatbot generates `service_intents_model_training.csv`, uploads
it to the Space, and the Space trains a service-intent extractor with Hugging
Face resources.

## Space Code

The local clone prepared for upload is:

```text
D:\Nouveau dossier\huggingface_spaces\train_model_chatbot
```

It contains:

- `app.py`: FastAPI job API.
- `training/train_service_intent_extractor.py`: standalone trainer.
- `Dockerfile`: Hugging Face Docker Space runtime.
- `requirements.txt`: Python dependencies.

## Space Secrets

Set these in the Hugging Face Space settings when needed:

```text
TRAINING_API_KEY=change-me
HF_TOKEN=hf_...
```

`TRAINING_API_KEY` protects the training API. `HF_TOKEN` is only required if the
Space should push the trained model to a Hugging Face model repository.

## Local Settings

Configure these in `.env`:

```text
SERVICE_INTENT_TRAINING_SPACE_URL=https://mortadhabbb-train-model-chatbot.hf.space
SERVICE_INTENT_TRAINING_SPACE_API_KEY=
SERVICE_INTENT_TRAINING_OUTPUT_REPO_ID=
SERVICE_INTENT_TRAINING_POLL_SECONDS=20
```

## Submit Training From Django

Rebuild the datasets and submit a remote training job:

```powershell
python manage.py submit_remote_service_intent_training --rebuild-datasets
```

Submit and wait until the Space finishes:

```powershell
python manage.py submit_remote_service_intent_training --rebuild-datasets --wait
```

Wait and download the trained artifact ZIP:

```powershell
python manage.py submit_remote_service_intent_training --rebuild-datasets --wait --download-artifact artifacts/service_intent_model_remote.zip
```

Ask the Space to push the trained model to a Hugging Face model repo:

```powershell
python manage.py submit_remote_service_intent_training --rebuild-datasets --wait --push-to-hub --output-repo-id mortadhabbb/service-intent-extractor
```

For free CPU Spaces, start with a smaller model:

```powershell
python manage.py submit_remote_service_intent_training --model-id google/flan-t5-small --rebuild-datasets --wait
```

Use `google/flan-t5-base` when the Space has enough CPU/GPU memory and time.