Spaces:
Paused
Paused
File size: 5,260 Bytes
101093d |
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 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 |
import streamlit as st
from meta_utils import run_subprocess, load_metagraphs
# from opendashboards.assets import io, inspect, metric, plot
from meta_plotting import plot_trace, plot_cabals
DEFAULT_SRC = 'miner'
DEFAULT_NTOP = 10
DEFAULT_UID_NTOP = 10
# Set app config
st.set_page_config(
page_title='Validator Dashboard',
menu_items={
'Report a bug': "https://github.com/opentensor/dashboards/issues",
'About': """
This dashboard is part of the OpenTensor project. \n
"""
},
layout = "centered"
)
st.title('Metagraph :red[Analysis] Dashboard :eyes:')
# add vertical space
st.markdown('#')
st.markdown('#')
with st.spinner(text=f'Loading data...'):
df = load_metagraphs()
blocks = df.block.unique()
# metric.wandb(df_runs)
# add vertical space
st.markdown('#')
st.markdown('#')
tab1, tab2, tab3, tab4 = st.tabs(["Health", "Miners", "Validators", "Block"])
### Wandb Runs ###
with tab1:
st.markdown('#')
st.header(":violet[Wandb] Runs")
run_msg = st.info("Select a single run or compare multiple runs")
selected_runs = st.multiselect(f'Runs ({len(df_runs)})', df_runs.id, default=DEFAULT_SELECTED_RUNS, key='runs')
# Load data if new runs selected
if not selected_runs:
# open a dialog to select runs
run_msg.error("Please select at least one run")
st.snow()
st.stop()
df = io.load_data(df_runs.loc[df_runs.id.isin(selected_runs)], load=True, save=True)
df_long = inspect.explode_data(df)
df_weights = inspect.weights(df)
metric.runs(df, df_long, selected_runs)
with st.expander(f'Show :violet[raw] data for {len(selected_runs)} selected runs'):
inspect.run_event_data(df_runs,df, selected_runs)
### UID Health ###
with tab2:
st.markdown('#')
st.header("UID :violet[Health]")
st.info(f"Showing UID health metrics for **{len(selected_runs)} selected runs**")
uid_src = st.radio('Select one:', ['followup', 'answer'], horizontal=True, key='uid_src')
metric.uids(df_long, uid_src)
with st.expander(f'Show UID **{uid_src}** weights data for **{len(selected_runs)} selected runs**'):
uids = st.multiselect('UID:', sorted(df_long[f'{uid_src}_uids'].unique()), key='uid')
st.markdown('#')
st.subheader(f"UID {uid_src.title()} :violet[Weights]")
plot.weights(
df_weights,
uids=uids,
)
with st.expander(f'Show UID **{uid_src}** leaderboard data for **{len(selected_runs)} selected runs**'):
st.markdown('#')
st.subheader(f"UID {uid_src.title()} :violet[Leaderboard]")
uid_col1, uid_col2 = st.columns(2)
uid_ntop = uid_col1.slider('Number of UIDs:', min_value=1, max_value=50, value=DEFAULT_UID_NTOP, key='uid_ntop')
uid_agg = uid_col2.selectbox('Aggregation:', ('mean','min','max','size','nunique'), key='uid_agg')
plot.leaderboard(
df,
ntop=uid_ntop,
group_on=f'{uid_src}_uids',
agg_col=f'{uid_src}_rewards',
agg=uid_agg
)
with st.expander(f'Show UID **{uid_src}** diversity data for **{len(selected_runs)} selected runs**'):
st.markdown('#')
st.subheader(f"UID {uid_src.title()} :violet[Diversity]")
rm_failed = st.checkbox(f'Remove failed **{uid_src}** completions', value=True)
plot.uid_diversty(df, rm_failed)
### Completions ###
with tab3:
st.markdown('#')
st.subheader('Completion :violet[Leaderboard]')
completion_info = st.empty()
msg_col1, msg_col2 = st.columns(2)
completion_src = msg_col1.radio('Select one:', ['followup', 'answer'], horizontal=True, key='completion_src')
completion_info.info(f"Showing **{completion_src}** completions for **{len(selected_runs)} selected runs**")
completion_ntop = msg_col2.slider('Top k:', min_value=1, max_value=50, value=DEFAULT_COMPLETION_NTOP, key='completion_ntop')
completion_col = f'{completion_src}_completions'
reward_col = f'{completion_src}_rewards'
uid_col = f'{completion_src}_uids'
completions = inspect.completions(df_long, completion_col)
# Get completions with highest average rewards
plot.leaderboard(
df,
ntop=completion_ntop,
group_on=completion_col,
agg_col=reward_col,
agg='mean',
alias=True
)
with st.expander(f'Show **{completion_src}** completion rewards data for **{len(selected_runs)} selected runs**'):
st.markdown('#')
st.subheader('Completion :violet[Rewards]')
completion_select = st.multiselect('Completions:', completions.index, default=completions.index[:3].tolist())
# completion_regex = st.text_input('Completion regex:', value='', key='completion_regex')
plot.completion_rewards(
df,
completion_col=completion_col,
reward_col=reward_col,
uid_col=uid_col,
ntop=completion_ntop,
completions=completion_select,
)
### Prompt-based scoring ###
with tab4:
# coming soon
st.info('Prompt-based scoring coming soon')
# st.dataframe(df_long_long.filter(regex=prompt_src).head())
|