Update app.py

app.py

@@ -618,11 +618,12 @@ def main(dataset_url, model_type, output_dataset_name, range_specification, targ
 gradio_title = "🧐 WMT21 Dataset Translation"
 gradio_desc = """This tool translates english datasets using the WMT21 translation model.
 ## 💭 What Does This Tool Do:
-- Translates datasets with structures based on the selected model type.
+- Translates datasets (as parquet files) with structures based on the selected model type (see below).
 - The translation model (facebook/wmt21-dense-24-wide-en-x) supports as target languages: Hausa (ha), Icelandic (is), Japanese (ja), Czech (cs), Russian (ru), Chinese (zh), German (de)
-- Uploads the translated dataset to Hugging Face.
-- At the moment, this works only on CPU, and therefore is very very slow (>1 minute per item depending on string lengths)."""
+- Uploads the translated dataset as jsonl to Hugging Face.
+- At the moment, this works only on CPU, and therefore is very very slow."""
 datasets_desc = """## 📊 Dataset Types:
+Note: additional fields will be kept (untranslated), an additional index field is added, which makes it easier to verify results, i.a.
 - **mix**: 
   - `prompt`: List of dictionaries with 'content' and 'role' fields (multi-turn conversation).
   - `chosen`: Single dictionary with 'content' and 'role' fields.