File size: 3,162 Bytes
0ba78e9
 
0d5e0f7
0ba78e9
 
0d5e0f7
0ba78e9
0d5e0f7
0ba78e9
 
0d5e0f7
0ba78e9
 
 
0d5e0f7
 
294f139
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f574f70
0d5e0f7
 
0ba78e9
 
 
 
 
0d5e0f7
544774d
294f139
544774d
 
0ba78e9
 
544774d
294f139
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0d5e0f7
0ba78e9
0d5e0f7
0ba78e9
0d5e0f7
 
 
 
0ba78e9
 
 
f574f70
0ba78e9
f574f70
0ba78e9
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
import os
from pathlib import Path

import streamlit as st
from dotenv import load_dotenv

from utils import get_compatible_models, get_metadata, http_post

if Path(".env").is_file():
    load_dotenv(".env")

HF_TOKEN = os.getenv("HF_TOKEN")
AUTOTRAIN_USERNAME = os.getenv("AUTOTRAIN_USERNAME")
AUTOTRAIN_BACKEND_API = os.getenv("AUTOTRAIN_BACKEND_API")


TASK_TO_ID = {
    "binary_classification": 1,
    "multi_class_classification": 2,
    "multi_label_classification": 3,
    "entity_extraction": 4,
    "extractive_question_answering": 5,
    "translation": 6,
    "summarization": 8,
    "single_column_regression": 10,
}

# TODO: remove this hardcorded logic and accept any dataset on the Hub
DATASETS_TO_EVALUATE = ["emotion", "conll2003"]

dataset_name = st.selectbox("Select a dataset", [f"lewtun/autoevaluate__{dset}" for dset in DATASETS_TO_EVALUATE])

with st.form(key="form"):

    # TODO: remove this step once we select real datasets
    # Strip out original dataset name
    original_dataset_name = dataset_name.split("/")[-1].split("__")[-1]

    # In general this will be a list of multiple configs => need to generalise logic here
    metadata = get_metadata(dataset_name)

    dataset_config = st.selectbox("Select a config", [metadata[0]["config"]])

    splits = metadata[0]["splits"]
    split_names = list(splits.values())
    eval_split = splits.get("eval_split", split_names[0])

    selected_split = st.selectbox("Select a split", split_names, index=split_names.index(eval_split))

    col_mapping = metadata[0]["col_mapping"]
    col_names = list(col_mapping.values())

    # TODO: figure out how to get all dataset column names (i.e. features) without download dataset itself
    st.markdown("**Map your data columns**")
    col1, col2 = st.columns(2)

    with col1:
        st.markdown("`text` column")
        st.text("")
        st.text("")
        st.text("")
        st.text("")
        st.markdown("`target` column")
    with col2:
        st.selectbox("This column should contain the text you want to classify", col_names, index=0)
        st.selectbox("This column should contain the labels you want to assign to the text", col_names, index=1)

    compatible_models = get_compatible_models(metadata[0]["task"], original_dataset_name)

    selected_models = st.multiselect("Select the models you wish to evaluate", compatible_models, compatible_models[0])

    submit_button = st.form_submit_button("Make Submission")

    if submit_button:
        for model in selected_models:
            payload = {
                "username": AUTOTRAIN_USERNAME,
                "task": TASK_TO_ID[metadata[0]["task_id"]],
                "model": model,
                "col_mapping": metadata[0]["col_mapping"],
                "split": selected_split,
                "dataset": original_dataset_name,
                "config": dataset_config,
            }
            json_resp = http_post(
                path="/evaluate/create", payload=payload, token=HF_TOKEN, domain=AUTOTRAIN_BACKEND_API
            ).json()

            st.success(f"✅ Successfully submitted model {model} for evaluation with job ID {json_resp['id']}")