app.py

import json
import re
from pathlib import Path

import requests
import streamlit as st
import yaml
from huggingface_hub import hf_hub_download
from streamlit_tags import st_tags

# exact same regex as in the Hub server. Please keep in sync.
REGEX_YAML_BLOCK = re.compile(r"---[\n\r]+([\S\s]*?)[\n\r]+---[\n\r]")

with open("languages.json") as f:
    lang2name = json.load(f)


def try_parse_yaml(yaml_block):
    try:
        metadata = yaml.load(yaml_block, yaml.SafeLoader)
    except yaml.YAMLError as e:
        print("Error while parsing the metadata YAML:")
        if hasattr(e, "problem_mark"):
            if e.context is not None:
                st.error(
                    str(e.problem_mark)
                    + "\n  "
                    + str(e.problem)
                    + " "
                    + str(e.context)
                    + "\nPlease correct the README.md and retry."
                )
            else:
                st.error(
                    str(e.problem_mark)
                    + "\n  "
                    + str(e.problem)
                    + "\nPlease correct the README.md and retry."
                )
        else:
            st.error(
                "Something went wrong while parsing the metadata. "
                "Make sure it's written according to the YAML spec!"
            )
        return None
    return metadata


def main():
    st.markdown("## 1. Load your model's metadata")
    st.markdown("Enter your model's path below.")
    model_id = st.text_input("", placeholder="<username>/<model>")
    if not model_id.strip():
        st.stop()
    try:
        readme_path = hf_hub_download(model_id, filename="README.md")
    except requests.exceptions.HTTPError:
        st.error(
            f"ERROR: https://huggingface.co/{model_id}/blob/main/README.md "
            f"not found, make sure you've entered a correct model path!"
        )
        st.stop()

    content = Path(readme_path).read_text()
    match = REGEX_YAML_BLOCK.search(content)
    if match:
        meta_yaml = match.group(1)
    else:
        st.error(
            "ERROR: Couldn't find the metadata section inside your model's `README.md`. Do you have some basic metadata "
            "enclosed in `---` as described in [the Hub documentation](https://huggingface.co/docs/hub/model-repos#model-card-metadata)?"
        )
        st.stop()

    metadata = try_parse_yaml(meta_yaml)
    if metadata is None:
        st.stop()
    else:
        st.success("Successfully loaded the metadata!")
    with st.expander("Inspect the parsed metadata for debugging"):
        st.json(metadata)

    st.markdown("## 2. Edit the data")

    ############################
    # LANGUAGES
    ############################
    st.markdown("### Language(s)")
    st.markdown(
        "For each spoken language that your model handles, enter an "
        "[ISO 639-1](https://en.wikipedia.org/wiki/List_of_ISO_639-1_codes) language code, or "
        "find an appropriate alternative from "
        "[our list here](https://huggingface.co/spaces/huggingface/hf-speech-bench/blob/main/languages.json). "
        "When in doubt, use the most generic language code, e.g. `en` instead of `en-GB` and `en-US`."
    )
    st.markdown("*Example*: `cs, hsb, pl`")
    metadata["language"] = metadata["language"] if "language" in metadata else []
    metadata["language"] = (
        metadata["language"]
        if isinstance(metadata["language"], list)
        else [metadata["language"]]
    )

    languages = st_tags(
        label="", text="add more if needed, and press enter", value=metadata["language"]
    )
    lang_names = [lang2name[lang] if lang in lang2name else lang for lang in languages]
    st.markdown("These languages will be parsed by the leaderboard as: ")
    st.code(", ".join(lang_names))

    ############################
    # TRAIN DATASETS
    ############################
    st.markdown("### Training dataset(s)")
    st.markdown("List the datasets that your model was trained on.")
    st.markdown("*Example*: `librispeech_asr, mozilla-foundation/common_voice_8_0`")

    if "datasets" not in metadata:
        metadata["datasets"] = []

    train_datasets = st_tags(
        label="", text="add more if needed, and press enter", value=metadata["datasets"]
    )
    if "common_voice" in train_datasets:
        st.warning(
            "WARNING: `common_voice` is deprecated, please replace it with its equivalent: "
            "`mozilla-foundation/common_voice_6_1`"
        )

    ############################
    # MODEL NAME
    ############################
    st.markdown("### Model name")
    st.markdown("Enter a descriptive name for your model.")
    st.markdown("*Example*: `XLS-R Wav2Vec2 LM Spanish by Jane Doe`")

    if "model_index" not in metadata:
        metadata["model_index"] = [{}]
    if "name" not in ["model_index"][0]:
        metadata["model_index"][0]["name"] = model_id.split("/")[-1]
    model_name = st.text_input("", value=metadata["model_index"][0]["name"])

    ############################
    # EVAL DATASETS
    ############################
    st.markdown("### Evaluation metrics")


if __name__ == "__main__":
    main()