Simplify training script

[AmitMY]

As an idea -

The setup script relies on the config file

multimodalhugs-setup \
  --modality "pose2text" \
  --config_path signwriting_transcription/config.yaml

Now that the training script can also just rely on the config file, how about the following?

multimodalhugs-train \
    --task "translation" \
    --config-path "signwriting_transcription/config.yaml"

It can automatically figure out these parameters

    --model_name_or_path $MODEL_PATH \
    --processor_name_or_path $PROCESSOR_PATH \
    --run_name $MODEL_NAME \
    --dataset_dir $DATA_PATH \
    --output_dir $OUTPUT_PATH \

And take everything else from the config.

OPTIONALLY the user could also specify more parameters (for example, learning_rate) but they should probably all go in the config

[GerrySant]

With the next version it will be possible to manage almost everything from the configuration itself. However, since we use the HuggingFace Trainer and it requires that we always specify the --output_dir, for now we can simplify the command as:

multimodalhugs-train \
    --task "translation" \
    --config-path "signwriting_transcription/config.yaml" \
    --output_dir "/path/to/your/output_directory"