Spaces:
Runtime error
Runtime error
Commit
•
cd05b03
1
Parent(s):
dbf0e3c
Update app.py
Browse files
app.py
CHANGED
@@ -1,7 +1,6 @@
|
|
1 |
import pandas as pd
|
2 |
import streamlit as st
|
3 |
-
from huggingface_hub import HfApi
|
4 |
-
from huggingface_hub.repocard import metadata_load
|
5 |
from utils import ascending_metrics, metric_ranges
|
6 |
import numpy as np
|
7 |
from st_aggrid import AgGrid, GridOptionsBuilder, JsCode
|
@@ -80,9 +79,9 @@ def parse_metrics_rows(meta, only_verified=False):
|
|
80 |
continue
|
81 |
yield row
|
82 |
|
|
|
83 |
@st.cache(ttl=0)
|
84 |
def get_data_wrapper():
|
85 |
-
|
86 |
def get_data(dataframe=None, verified_dataframe=None):
|
87 |
data = []
|
88 |
verified_data = []
|
@@ -126,107 +125,69 @@ def get_data_wrapper():
|
|
126 |
|
127 |
return dataframe
|
128 |
|
|
|
129 |
dataframe = get_data_wrapper()
|
130 |
|
131 |
-
st.markdown("# 🤗
|
132 |
|
133 |
query_params = st.experimental_get_query_params()
|
134 |
if "first_query_params" not in st.session_state:
|
135 |
st.session_state.first_query_params = query_params
|
136 |
first_query_params = st.session_state.first_query_params
|
137 |
|
138 |
-
default_task = first_query_params.get("task", [None])[0]
|
139 |
-
default_only_verified = bool(int(first_query_params.get("only_verified", [0])[0]))
|
140 |
-
print(default_only_verified)
|
141 |
-
default_dataset = first_query_params.get("dataset", [None])[0]
|
142 |
-
default_split = first_query_params.get("split", [None])[0]
|
143 |
default_config = first_query_params.get("config", [None])[0]
|
144 |
default_metric = first_query_params.get("metric", [None])[0]
|
145 |
|
146 |
-
only_verified_results =
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
)
|
151 |
-
|
152 |
-
selectable_tasks = list(set(dataframe.pipeline_tag))
|
153 |
-
if None in selectable_tasks:
|
154 |
-
selectable_tasks.remove(None)
|
155 |
-
selectable_tasks.sort(key=lambda name: name.lower())
|
156 |
-
selectable_tasks = ["-any-"] + selectable_tasks
|
157 |
-
|
158 |
-
task = st.sidebar.selectbox(
|
159 |
-
"Task",
|
160 |
-
selectable_tasks,
|
161 |
-
index=(selectable_tasks).index(default_task) if default_task in selectable_tasks else 0,
|
162 |
-
help="Filter the selectable datasets by task. Leave as \"-any-\" to see all selectable datasets."
|
163 |
-
)
|
164 |
-
|
165 |
-
if task != "-any-":
|
166 |
-
dataframe = dataframe[dataframe.pipeline_tag == task]
|
167 |
-
|
168 |
-
selectable_datasets = ["-any-"] + sorted(list(set(dataframe.dataset.tolist())), key=lambda name: name.lower())
|
169 |
-
if "" in selectable_datasets:
|
170 |
-
selectable_datasets.remove("")
|
171 |
-
|
172 |
-
dataset = st.sidebar.selectbox(
|
173 |
-
"Dataset",
|
174 |
-
selectable_datasets,
|
175 |
-
index=selectable_datasets.index(default_dataset) if default_dataset in selectable_datasets else 0,
|
176 |
-
help="Select a dataset to see the leaderboard!"
|
177 |
-
)
|
178 |
|
179 |
dataframe = dataframe[dataframe.only_verified == only_verified_results]
|
180 |
|
181 |
-
current_query_params = {"dataset": [dataset], "only_verified": [int(only_verified_results)], "task": [task]
|
|
|
182 |
|
183 |
st.experimental_set_query_params(**current_query_params)
|
184 |
|
185 |
-
|
186 |
-
|
187 |
-
else:
|
188 |
-
dataset_df = dataframe
|
189 |
|
190 |
dataset_df = dataset_df.dropna(axis="columns", how="all")
|
191 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
192 |
if len(dataset_df) > 0:
|
193 |
selectable_configs = list(set(dataset_df["config"]))
|
194 |
selectable_configs.sort(key=lambda name: name.lower())
|
195 |
|
196 |
-
|
197 |
-
selectable_configs.remove("-unspecified-")
|
198 |
-
selectable_configs = ["-unspecified-"] + selectable_configs
|
199 |
|
200 |
if dataset != "-any-":
|
201 |
config = st.sidebar.selectbox(
|
202 |
-
"
|
203 |
selectable_configs,
|
204 |
-
index=
|
205 |
-
help="Filter the results on the current leaderboard by
|
206 |
)
|
207 |
dataset_df = dataset_df[dataset_df.config == config]
|
208 |
|
209 |
-
selectable_splits =
|
210 |
-
selectable_splits.sort(key=lambda name: name.lower())
|
211 |
-
|
212 |
-
if "-unspecified-" in selectable_splits:
|
213 |
-
selectable_splits.remove("-unspecified-")
|
214 |
-
selectable_splits = ["-unspecified-"] + selectable_splits
|
215 |
-
|
216 |
split = st.sidebar.selectbox(
|
217 |
"Split",
|
218 |
selectable_splits,
|
219 |
-
index=
|
220 |
-
help="Filter the results on the current leaderboard by the dataset split. Self-reported results might not report the split, which is why \"-unspecified-\" is an option."
|
221 |
)
|
222 |
|
223 |
-
current_query_params.update({"config": [config], "split": [split]})
|
224 |
-
|
225 |
-
st.experimental_set_query_params(**current_query_params)
|
226 |
-
|
227 |
-
dataset_df = dataset_df[dataset_df.split == split]
|
228 |
-
|
229 |
not_selectable_metrics = ["model_id", "dataset", "split", "config", "pipeline_tag", "only_verified"]
|
|
|
|
|
|
|
230 |
selectable_metrics = list(filter(lambda column: column not in not_selectable_metrics, dataset_df.columns))
|
231 |
|
232 |
dataset_df = dataset_df.filter(["model_id"] + (["dataset"] if dataset == "-any-" else []) + selectable_metrics)
|
@@ -248,18 +209,9 @@ if len(dataset_df) > 0:
|
|
248 |
)
|
249 |
|
250 |
st.markdown(
|
251 |
-
"Want to beat the leaderboard? Don't see your model here? Simply
|
252 |
-
)
|
253 |
-
|
254 |
-
st.markdown(
|
255 |
-
"If you do not see your self-reported results here, ensure that your results are in the expected range for all metrics. E.g., accuracy is 0-1, not 0-100."
|
256 |
)
|
257 |
|
258 |
-
if dataset == "-any-":
|
259 |
-
st.info(
|
260 |
-
"Note: you haven't chosen a dataset, so the leaderboard is showing the best scoring model for a random sample of the datasets available."
|
261 |
-
)
|
262 |
-
|
263 |
# Make the default metric appear right after model names and dataset names
|
264 |
cols = dataset_df.columns.tolist()
|
265 |
cols.remove(sorting_metric)
|
@@ -313,4 +265,4 @@ if len(dataset_df) > 0:
|
|
313 |
else:
|
314 |
st.markdown(
|
315 |
"No " + ("verified" if only_verified_results else "unverified") + " results to display. Try toggling the verified results filter."
|
316 |
-
)
|
|
|
1 |
import pandas as pd
|
2 |
import streamlit as st
|
3 |
+
from huggingface_hub import HfApi
|
|
|
4 |
from utils import ascending_metrics, metric_ranges
|
5 |
import numpy as np
|
6 |
from st_aggrid import AgGrid, GridOptionsBuilder, JsCode
|
|
|
79 |
continue
|
80 |
yield row
|
81 |
|
82 |
+
|
83 |
@st.cache(ttl=0)
|
84 |
def get_data_wrapper():
|
|
|
85 |
def get_data(dataframe=None, verified_dataframe=None):
|
86 |
data = []
|
87 |
verified_data = []
|
|
|
125 |
|
126 |
return dataframe
|
127 |
|
128 |
+
|
129 |
dataframe = get_data_wrapper()
|
130 |
|
131 |
+
st.markdown("# 🤗 Whisper Event: Final Leaderboard")
|
132 |
|
133 |
query_params = st.experimental_get_query_params()
|
134 |
if "first_query_params" not in st.session_state:
|
135 |
st.session_state.first_query_params = query_params
|
136 |
first_query_params = st.session_state.first_query_params
|
137 |
|
|
|
|
|
|
|
|
|
|
|
138 |
default_config = first_query_params.get("config", [None])[0]
|
139 |
default_metric = first_query_params.get("metric", [None])[0]
|
140 |
|
141 |
+
only_verified_results = False
|
142 |
+
task = "automatic-speech-recognition"
|
143 |
+
dataset = "mozilla-foundation/common_voice_11_0"
|
144 |
+
split = "test"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
145 |
|
146 |
dataframe = dataframe[dataframe.only_verified == only_verified_results]
|
147 |
|
148 |
+
current_query_params = {"dataset": [dataset], "only_verified": [int(only_verified_results)], "task": [task],
|
149 |
+
"split": [split]}
|
150 |
|
151 |
st.experimental_set_query_params(**current_query_params)
|
152 |
|
153 |
+
dataset_df = dataframe[dataframe.dataset == dataset]
|
154 |
+
dataset_df = dataset_df[dataset_df.split == split]
|
|
|
|
|
155 |
|
156 |
dataset_df = dataset_df.dropna(axis="columns", how="all")
|
157 |
|
158 |
+
selectable_datasets = [dataset]
|
159 |
+
dataset = st.sidebar.selectbox(
|
160 |
+
"Dataset",
|
161 |
+
selectable_datasets,
|
162 |
+
index=0,
|
163 |
+
)
|
164 |
+
|
165 |
if len(dataset_df) > 0:
|
166 |
selectable_configs = list(set(dataset_df["config"]))
|
167 |
selectable_configs.sort(key=lambda name: name.lower())
|
168 |
|
169 |
+
selectable_configs.remove("-unspecified-")
|
|
|
|
|
170 |
|
171 |
if dataset != "-any-":
|
172 |
config = st.sidebar.selectbox(
|
173 |
+
"Language",
|
174 |
selectable_configs,
|
175 |
+
index=0,
|
176 |
+
help="Filter the results on the current leaderboard by language."
|
177 |
)
|
178 |
dataset_df = dataset_df[dataset_df.config == config]
|
179 |
|
180 |
+
selectable_splits = [split]
|
|
|
|
|
|
|
|
|
|
|
|
|
181 |
split = st.sidebar.selectbox(
|
182 |
"Split",
|
183 |
selectable_splits,
|
184 |
+
index=0,
|
|
|
185 |
)
|
186 |
|
|
|
|
|
|
|
|
|
|
|
|
|
187 |
not_selectable_metrics = ["model_id", "dataset", "split", "config", "pipeline_tag", "only_verified"]
|
188 |
+
# also ignore irrelevant ASR metrics
|
189 |
+
not_selectable_metrics.extend(["wer_without_norm", "mer"])
|
190 |
+
|
191 |
selectable_metrics = list(filter(lambda column: column not in not_selectable_metrics, dataset_df.columns))
|
192 |
|
193 |
dataset_df = dataset_df.filter(["model_id"] + (["dataset"] if dataset == "-any-" else []) + selectable_metrics)
|
|
|
209 |
)
|
210 |
|
211 |
st.markdown(
|
212 |
+
"Want to beat the leaderboard? Don't see your model here? Simply ..."
|
|
|
|
|
|
|
|
|
213 |
)
|
214 |
|
|
|
|
|
|
|
|
|
|
|
215 |
# Make the default metric appear right after model names and dataset names
|
216 |
cols = dataset_df.columns.tolist()
|
217 |
cols.remove(sorting_metric)
|
|
|
265 |
else:
|
266 |
st.markdown(
|
267 |
"No " + ("verified" if only_verified_results else "unverified") + " results to display. Try toggling the verified results filter."
|
268 |
+
)
|