Tcid / app.py
manueldeprada's picture
Upload folder using huggingface_hub
891c332 verified
raw
history blame
25.1 kB
import matplotlib.pyplot as plt
import matplotlib
import pandas as pd
import gradio as gr
from data import CIResults
from utils import logger
from summary_page import create_summary_page
from model_page import plot_model_stats
from time_series_gradio import create_time_series_summary_gradio, create_model_time_series_gradio
# Configure matplotlib to prevent memory warnings and set dark background
matplotlib.rcParams['figure.facecolor'] = '#000000'
matplotlib.rcParams['axes.facecolor'] = '#000000'
matplotlib.rcParams['savefig.facecolor'] = '#000000'
plt.ioff() # Turn off interactive mode to prevent figure accumulation
# Load data once at startup
Ci_results = CIResults()
Ci_results.load_data()
# Start the auto-reload scheduler
Ci_results.schedule_data_reload()
# Function to check if a model has failures
def model_has_failures(model_name):
"""Check if a model has any failures (AMD or NVIDIA)."""
if Ci_results.df is None or Ci_results.df.empty:
return False
# Normalize model name to match DataFrame index
model_name_lower = model_name.lower()
# Check if model exists in DataFrame
if model_name_lower not in Ci_results.df.index:
return False
row = Ci_results.df.loc[model_name_lower]
# Check for failures in both AMD and NVIDIA
amd_multi_failures = row.get('failed_multi_no_amd', 0)
amd_single_failures = row.get('failed_single_no_amd', 0)
nvidia_multi_failures = row.get('failed_multi_no_nvidia', 0)
nvidia_single_failures = row.get('failed_single_no_nvidia', 0)
return any([
amd_multi_failures > 0,
amd_single_failures > 0,
nvidia_multi_failures > 0,
nvidia_single_failures > 0,
])
# Function to get current description text
def get_description_text():
"""Get description text with integrated last update time."""
msg = [
"Transformer CI Dashboard",
"-",
"AMD runs on MI325",
"NVIDIA runs on A10",
]
msg = ["**" + x + "**" for x in msg] + [""]
if Ci_results.latest_update_msg:
msg.append(f"*({Ci_results.latest_update_msg})*")
else:
msg.append("*(loading...)*")
return "<br>".join(msg)
# Load CSS from external file
def load_css():
try:
with open("styles.css", "r") as f:
css_content = f.read()
return css_content
except FileNotFoundError:
logger.warning("styles.css not found, using minimal default styles")
return "body { background: #000; color: #fff; }"
# Create the Gradio interface with sidebar and dark theme
with gr.Blocks(title="Model Test Results Dashboard", css=load_css()) as demo:
with gr.Row():
# Sidebar for model selection
with gr.Column(scale=1, elem_classes=["sidebar"]):
gr.Markdown("# 🤖 TCID", elem_classes=["sidebar-title"])
# Description with integrated last update time
description_text = get_description_text()
description_display = gr.Markdown(description_text, elem_classes=["sidebar-description"])
# View toggle buttons
with gr.Row(elem_classes=["view-toggle-row"]):
current_view_button = gr.Button(
"current\n📊",
variant="primary",
size="lg",
elem_classes=["view-toggle-button", "view-toggle-active"]
)
historical_view_button = gr.Button(
"history\n📈",
variant="secondary",
size="lg",
elem_classes=["view-toggle-button"]
)
# Date selection toggle button (initially hidden)
date_toggle_button = gr.Button(
"► Date Selection",
variant="secondary",
elem_classes=["date-header"],
visible=False
)
# Date selection container (collapsible) - start folded
with gr.Column(visible=True, elem_classes=["date-selection", "date-selection-hidden"]) as date_selection:
start_date = gr.Dropdown(
choices=Ci_results.available_dates,
value=Ci_results.available_dates[-1] if Ci_results.available_dates else None, # Last date (oldest)
label="Start Date",
elem_classes=["date-dropdown"]
)
end_date = gr.Dropdown(
choices=Ci_results.available_dates,
value=Ci_results.available_dates[0] if Ci_results.available_dates else None, # First date (newest)
label="End Date",
elem_classes=["date-dropdown"]
)
load_historical_button = gr.Button(
"Reload Historical Data",
variant="primary",
size="sm",
elem_classes=["load-historical-button"]
)
# Summary button (for current view)
summary_button = gr.Button(
"summary\n📊",
variant="primary",
size="lg",
elem_classes=["summary-button"]
)
# Model selection header (clickable toggle)
model_toggle_button = gr.Button(
f"► Select model ({len(Ci_results.available_models)})",
variant="secondary",
elem_classes=["model-header"]
)
# Model buttons container (collapsible) - start folded
with gr.Column(elem_classes=["model-list", "model-list-hidden"]) as model_list_container:
# Create individual buttons for each model
model_buttons = []
model_choices = [model.lower() for model in Ci_results.available_models] if Ci_results.available_models else ["auto", "bert", "clip", "llama"]
print(f"Creating {len(model_choices)} model buttons: {model_choices}")
for model_name in model_choices:
# Check if model has failures to determine styling
has_failures = model_has_failures(model_name)
button_classes = ["model-button"]
if has_failures:
button_classes.append("model-button-failed")
btn = gr.Button(
model_name,
variant="secondary",
size="sm",
elem_classes=button_classes
)
model_buttons.append(btn)
# CI job links at bottom of sidebar
ci_links_display = gr.Markdown("🔗 **CI Jobs:** *Loading...*", elem_classes=["sidebar-links"])
# Main content area
with gr.Column(scale=4, elem_classes=["main-content"]):
# Current view components
with gr.Column(visible=True, elem_classes=["current-view"]) as current_view:
# Summary display (default view)
summary_display = gr.Plot(
value=create_summary_page(Ci_results.df, Ci_results.available_models),
label="",
format="png",
elem_classes=["plot-container"],
visible=True
)
# Detailed view components (hidden by default)
with gr.Column(visible=False, elem_classes=["detail-view"]) as detail_view:
# Back button for current view detail
back_to_summary_current_button = gr.Button(
"← Back to Summary",
variant="secondary",
size="sm",
elem_classes=["back-button"]
)
# Create the plot output
plot_output = gr.Plot(
label="",
format="png",
elem_classes=["plot-container"]
)
# Create two separate failed tests displays in a row layout
with gr.Row():
with gr.Column(scale=1):
amd_failed_tests_output = gr.Textbox(
value="",
lines=8,
max_lines=8,
interactive=False,
container=False,
elem_classes=["failed-tests"]
)
with gr.Column(scale=1):
nvidia_failed_tests_output = gr.Textbox(
value="",
lines=8,
max_lines=8,
interactive=False,
container=False,
elem_classes=["failed-tests"]
)
# Historical view components (hidden by default)
with gr.Column(visible=False, elem_classes=["historical-view"]) as historical_view:
# Time-series summary displays (multiple Gradio plots)
time_series_failure_rates = gr.LinePlot(
label="",
elem_classes=["plot-container"]
)
time_series_amd_tests = gr.LinePlot(
label="",
elem_classes=["plot-container"]
)
time_series_nvidia_tests = gr.LinePlot(
label="",
elem_classes=["plot-container"]
)
# Time-series model view (hidden by default)
with gr.Column(visible=False, elem_classes=["time-series-detail-view"]) as time_series_detail_view:
# Back button for time-series model view
back_to_summary_button = gr.Button(
"← Back to Summary",
variant="secondary",
size="sm",
elem_classes=["back-button"]
)
# Time-series plots for specific model (with spacing)
time_series_amd_model_plot = gr.LinePlot(
label="",
elem_classes=["plot-container"]
)
time_series_nvidia_model_plot = gr.LinePlot(
label="",
elem_classes=["plot-container"]
)
# Set up click handlers for model buttons
for i, btn in enumerate(model_buttons):
model_name = model_choices[i]
btn.click(
fn=lambda selected_model=model_name: plot_model_stats(Ci_results.df, selected_model),
outputs=[plot_output, amd_failed_tests_output, nvidia_failed_tests_output]
).then(
fn=lambda: [gr.update(visible=False), gr.update(visible=True)],
outputs=[summary_display, detail_view]
)
# Model toggle functionality
def toggle_model_list(current_visible):
"""Toggle the visibility of the model list."""
new_visible = not current_visible
arrow = "▼" if new_visible else "►"
button_text = f"{arrow} Select model ({len(Ci_results.available_models)})"
# Use CSS classes instead of Gradio visibility
css_classes = ["model-list"]
if new_visible:
css_classes.append("model-list-visible")
else:
css_classes.append("model-list-hidden")
return gr.update(value=button_text), gr.update(elem_classes=css_classes), new_visible
# Track model list visibility state
model_list_visible = gr.State(False)
model_toggle_button.click(
fn=toggle_model_list,
inputs=[model_list_visible],
outputs=[model_toggle_button, model_list_container, model_list_visible]
)
# Date toggle functionality
def toggle_date_selection(current_visible):
"""Toggle the visibility of the date selection."""
new_visible = not current_visible
arrow = "▼" if new_visible else "►"
button_text = f"{arrow} Date Selection"
# Use CSS classes instead of Gradio visibility
css_classes = ["date-selection"]
if new_visible:
css_classes.append("date-selection-visible")
else:
css_classes.append("date-selection-hidden")
return gr.update(value=button_text), gr.update(elem_classes=css_classes), new_visible
# Track date selection visibility state
date_selection_visible = gr.State(False)
date_toggle_button.click(
fn=toggle_date_selection,
inputs=[date_selection_visible],
outputs=[date_toggle_button, date_selection, date_selection_visible]
)
# Summary button click handler
def show_summary_and_update_links():
"""Show summary page and update CI links."""
return create_summary_page(Ci_results.df, Ci_results.available_models), get_description_text(), get_ci_links()
summary_button.click(
fn=show_summary_and_update_links,
outputs=[summary_display, description_display, ci_links_display]
).then(
fn=lambda: [gr.update(visible=True), gr.update(visible=False)],
outputs=[summary_display, detail_view]
)
# Function to get CI job links
def get_ci_links():
"""Get CI job links from the most recent data."""
try:
# Check if df exists and is not empty
if Ci_results.df is None or Ci_results.df.empty:
return "🔗 **CI Jobs:** *Loading...*"
# Get links from any available model (they should be the same for all models in a run)
amd_multi_link = None
amd_single_link = None
nvidia_multi_link = None
nvidia_single_link = None
for model_name in Ci_results.df.index:
row = Ci_results.df.loc[model_name]
# Extract AMD links
if pd.notna(row.get('job_link_amd')) and (not amd_multi_link or not amd_single_link):
amd_link_raw = row.get('job_link_amd')
if isinstance(amd_link_raw, dict):
if 'multi' in amd_link_raw and not amd_multi_link:
amd_multi_link = amd_link_raw['multi']
if 'single' in amd_link_raw and not amd_single_link:
amd_single_link = amd_link_raw['single']
# Extract NVIDIA links
if pd.notna(row.get('job_link_nvidia')) and (not nvidia_multi_link or not nvidia_single_link):
nvidia_link_raw = row.get('job_link_nvidia')
if isinstance(nvidia_link_raw, dict):
if 'multi' in nvidia_link_raw and not nvidia_multi_link:
nvidia_multi_link = nvidia_link_raw['multi']
if 'single' in nvidia_link_raw and not nvidia_single_link:
nvidia_single_link = nvidia_link_raw['single']
# Break if we have all links
if amd_multi_link and amd_single_link and nvidia_multi_link and nvidia_single_link:
break
# Add FAQ link at the bottom
links_md = "❓ [**FAQ**](https://huggingface.co/spaces/transformers-community/transformers-ci-dashboard/blob/main/README.md)\n\n"
links_md += "🔗 **CI Jobs:**\n\n"
# AMD links
if amd_multi_link or amd_single_link:
links_md += "**AMD:**\n"
if amd_multi_link:
links_md += f"• [Multi GPU]({amd_multi_link})\n"
if amd_single_link:
links_md += f"• [Single GPU]({amd_single_link})\n"
links_md += "\n"
# NVIDIA links
if nvidia_multi_link or nvidia_single_link:
links_md += "**NVIDIA:**\n"
if nvidia_multi_link:
links_md += f"• [Multi GPU]({nvidia_multi_link})\n"
if nvidia_single_link:
links_md += f"• [Single GPU]({nvidia_single_link})\n"
if not (amd_multi_link or amd_single_link or nvidia_multi_link or nvidia_single_link):
links_md += "*No links available*"
return links_md
except Exception as e:
logger.error(f"getting CI links: {e}")
return "🔗 **CI Jobs:** *Error loading links*\n\n❓ **[FAQ](README.md)**"
# View toggle functionality
def toggle_to_current_view():
"""Switch to current view."""
return [
gr.update(visible=True), # current_view
gr.update(visible=False), # historical_view
gr.update(visible=False), # date_toggle_button
gr.update(visible=True), # summary_button
gr.update(variant="primary", elem_classes=["view-toggle-button", "view-toggle-active"]), # current_view_button
gr.update(variant="secondary", elem_classes=["view-toggle-button"]) # historical_view_button
]
def toggle_to_historical_view():
"""Switch to historical view first, then auto-load data."""
# First, just switch the view
return [
gr.update(visible=False), # current_view
gr.update(visible=True), # historical_view
gr.update(visible=True), # date_toggle_button
gr.update(visible=False), # summary_button
gr.update(variant="secondary", elem_classes=["view-toggle-button"]), # current_view_button
gr.update(variant="primary", elem_classes=["view-toggle-button", "view-toggle-active"]), # historical_view_button
gr.update(), # time_series_failure_rates
gr.update(), # time_series_amd_tests
gr.update(), # time_series_nvidia_tests
]
def auto_load_historical_data():
"""Auto-load data for preselected dates after view switch."""
# Get the preselected dates
start_date_val = Ci_results.available_dates[-1] if Ci_results.available_dates else None
end_date_val = Ci_results.available_dates[0] if Ci_results.available_dates else None
# Check if we already have data for these dates
if (hasattr(Ci_results, 'cached_start_date') and hasattr(Ci_results, 'cached_end_date') and
Ci_results.cached_start_date == start_date_val and Ci_results.cached_end_date == end_date_val and
not Ci_results.historical_df.empty):
# Use cached data - no loading indicator needed
plots = create_time_series_summary_gradio(Ci_results.historical_df)
return plots['failure_rates'], plots['amd_tests'], plots['nvidia_tests']
# Auto-load historical data if dates are available
if start_date_val and end_date_val:
try:
Ci_results.load_historical_data(start_date_val, end_date_val)
if not Ci_results.historical_df.empty:
# Cache the loaded data
Ci_results.cached_start_date = start_date_val
Ci_results.cached_end_date = end_date_val
plots = create_time_series_summary_gradio(Ci_results.historical_df)
return plots['failure_rates'], plots['amd_tests'], plots['nvidia_tests']
else:
return gr.update(), gr.update(), gr.update()
except Exception as e:
logger.error(f"Error auto-loading historical data: {e}")
return gr.update(), gr.update(), gr.update()
else:
return gr.update(), gr.update(), gr.update()
current_view_button.click(
fn=toggle_to_current_view,
outputs=[current_view, historical_view, date_toggle_button, summary_button, current_view_button, historical_view_button]
)
historical_view_button.click(
fn=toggle_to_historical_view,
outputs=[current_view, historical_view, date_toggle_button, summary_button, current_view_button, historical_view_button, time_series_failure_rates, time_series_amd_tests, time_series_nvidia_tests]
).then(
fn=auto_load_historical_data,
outputs=[time_series_failure_rates, time_series_amd_tests, time_series_nvidia_tests]
)
# Historical data loading functionality
def load_historical_data(start_date, end_date):
"""Load and display historical data."""
if not start_date or not end_date:
logger.error("No start or end date provided")
return gr.update(), gr.update(), gr.update()
try:
Ci_results.load_historical_data(start_date, end_date)
if Ci_results.historical_df.empty:
logger.error("No historical data found for the selected date range")
return gr.update(), gr.update(), gr.update()
# Create time-series summary plots
plots = create_time_series_summary_gradio(Ci_results.historical_df)
# Cache the loaded data
Ci_results.cached_start_date = start_date
Ci_results.cached_end_date = end_date
return plots['failure_rates'], plots['amd_tests'], plots['nvidia_tests']
except Exception as e:
logger.error(f"Error loading historical data: {e}")
return gr.update(), gr.update(), gr.update()
load_historical_button.click(
fn=load_historical_data,
inputs=[start_date, end_date],
outputs=[time_series_failure_rates, time_series_amd_tests, time_series_nvidia_tests]
)
# Time-series model selection functionality
def show_time_series_model(selected_model):
"""Show time-series view for a specific model."""
if Ci_results.historical_df.empty:
return gr.update(), gr.update()
try:
plots = create_model_time_series_gradio(Ci_results.historical_df, selected_model)
return plots['amd_plot'], plots['nvidia_plot']
except Exception as e:
logger.error(f"Error creating time-series for model {selected_model}: {e}")
return gr.update(), gr.update()
# Back button functionality
def back_to_summary():
"""Return from model time-series view to summary time-series view."""
return [
gr.update(visible=True), # time_series_failure_rates
gr.update(visible=True), # time_series_amd_tests
gr.update(visible=True), # time_series_nvidia_tests
gr.update(visible=False) # time_series_detail_view
]
back_to_summary_button.click(
fn=back_to_summary,
outputs=[time_series_failure_rates, time_series_amd_tests, time_series_nvidia_tests, time_series_detail_view]
)
# Back button functionality for current view
def back_to_summary_current():
"""Return from model detail view to summary view in current view."""
return [
gr.update(visible=True), # summary_display
gr.update(visible=False) # detail_view
]
back_to_summary_current_button.click(
fn=back_to_summary_current,
outputs=[summary_display, detail_view]
)
# Update model button handlers to work with both views
for i, btn in enumerate(model_buttons):
model_name = model_choices[i]
# Current view handler (existing functionality)
btn.click(
fn=lambda selected_model=model_name: plot_model_stats(Ci_results.df, selected_model),
outputs=[plot_output, amd_failed_tests_output, nvidia_failed_tests_output]
).then(
fn=lambda: [gr.update(visible=False), gr.update(visible=True)],
outputs=[summary_display, detail_view]
)
# Historical view handler (new functionality)
btn.click(
fn=lambda selected_model=model_name: show_time_series_model(selected_model),
outputs=[time_series_amd_model_plot, time_series_nvidia_model_plot]
).then(
fn=lambda: [gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=True)],
outputs=[time_series_failure_rates, time_series_amd_tests, time_series_nvidia_tests, time_series_detail_view]
)
# Auto-update CI links when the interface loads
demo.load(
fn=get_ci_links,
outputs=[ci_links_display]
)
# Gradio entrypoint
if __name__ == "__main__":
demo.launch()