“WadoodAbdul”
added evaluation metric type radio button
f738aa2
raw
history blame
26.5 kB
import subprocess
import gradio as gr
import pandas as pd
from apscheduler.schedulers.background import BackgroundScheduler
from huggingface_hub import snapshot_download
from src.about import (
CITATION_BUTTON_LABEL,
CITATION_BUTTON_TEXT,
EVALUATION_QUEUE_TEXT,
INTRODUCTION_TEXT,
LLM_BENCHMARKS_TEXT_1,
EVALUATION_EXAMPLE_IMG,
LLM_BENCHMARKS_TEXT_2,
ENTITY_DISTRIBUTION_IMG,
LLM_BENCHMARKS_TEXT_3,
TITLE,
LOGO
)
from src.display.css_html_js import custom_css
from src.display.utils import (
DATASET_BENCHMARK_COLS,
TYPES_BENCHMARK_COLS,
DATASET_COLS,
Clinical_TYPES_COLS,
EVAL_COLS,
EVAL_TYPES,
NUMERIC_INTERVALS,
TYPES,
AutoEvalColumn,
ModelType,
ModelArch,
PromptTemplateName,
Precision,
WeightType,
fields,
)
from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN
from src.populate import get_evaluation_queue_df, get_leaderboard_df
from src.submission.submit import add_new_eval, PLACEHOLDER_DATASET_WISE_NORMALIZATION_CONFIG
def restart_space():
API.restart_space(repo_id=REPO_ID)
try:
print(EVAL_REQUESTS_PATH)
snapshot_download(
repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
)
except Exception:
restart_space()
try:
print(EVAL_RESULTS_PATH)
snapshot_download(
repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
)
except Exception:
restart_space()
# Span based results
_, span_based_datasets_original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, DATASET_COLS, DATASET_BENCHMARK_COLS, "SpanBasedWithPartialOverlap", "datasets")
span_based_datasets_leaderboard_df = span_based_datasets_original_df.copy()
_, span_based_types_original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, Clinical_TYPES_COLS, TYPES_BENCHMARK_COLS, "SpanBasedWithPartialOverlap", "clinical_types")
span_based_types_leaderboard_df = span_based_types_original_df.copy()
# Token based results
_, token_based_datasets_original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, DATASET_COLS, DATASET_BENCHMARK_COLS, "TokenBasedWithMacroAverage", "datasets")
token_based_datasets_leaderboard_df = token_based_datasets_original_df.copy()
_, token_based_types_original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, Clinical_TYPES_COLS, TYPES_BENCHMARK_COLS, "TokenBasedWithMacroAverage", "clinical_types")
token_based_types_leaderboard_df = token_based_types_original_df.copy()
(
finished_eval_queue_df,
running_eval_queue_df,
pending_eval_queue_df,
) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
def update_df(evaluation_metric, shown_columns, subset="datasets"):
print(evaluation_metric)
if subset == "datasets":
match evaluation_metric:
case "Span Based":
leaderboard_table_df = span_based_datasets_leaderboard_df.copy()
hidden_leader_board_df = span_based_datasets_original_df
case "Token Based":
leaderboard_table_df = token_based_datasets_leaderboard_df.copy()
hidden_leader_board_df = token_based_datasets_original_df
case _:
pass
else:
match evaluation_metric:
case "Span Based":
leaderboard_table_df = span_based_types_leaderboard_df.copy()
hidden_leader_board_df = span_based_types_original_df
case "Token Based":
leaderboard_table_df = token_based_types_leaderboard_df.copy()
hidden_leader_board_df = token_based_types_original_df
case _:
pass
value_cols = [c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns
return leaderboard_table_df[value_cols], hidden_leader_board_df
# Searching and filtering
def update_table(
hidden_df: pd.DataFrame,
columns: list,
query: str,
type_query: list = None,
architecture_query: list = None,
size_query: list = None,
precision_query: str = None,
show_deleted: bool = False,
):
filtered_df = filter_models(hidden_df, type_query, architecture_query, size_query, precision_query, show_deleted)
filtered_df = filter_queries(query, filtered_df)
df = select_columns(filtered_df, columns, list(hidden_df.columns))
return df
def search_table(df: pd.DataFrame, query: str) -> pd.DataFrame:
return df[(df[AutoEvalColumn.model.name].str.contains(query, case=False))]
def select_columns(df: pd.DataFrame, columns: list, cols:list) -> pd.DataFrame:
always_here_cols = [
AutoEvalColumn.model_type_symbol.name,
AutoEvalColumn.model.name,
]
# We use COLS to maintain sorting
filtered_df = df[always_here_cols + [c for c in cols if c in df.columns and c in columns]]
return filtered_df
def filter_queries(query: str, filtered_df: pd.DataFrame) -> pd.DataFrame:
final_df = []
if query != "":
queries = [q.strip() for q in query.split(";")]
for _q in queries:
_q = _q.strip()
if _q != "":
temp_filtered_df = search_table(filtered_df, _q)
if len(temp_filtered_df) > 0:
final_df.append(temp_filtered_df)
if len(final_df) > 0:
filtered_df = pd.concat(final_df)
filtered_df = filtered_df.drop_duplicates(
subset=[
AutoEvalColumn.model.name,
# AutoEvalColumn.precision.name,
# AutoEvalColumn.revision.name,
]
)
return filtered_df
def filter_models(
df: pd.DataFrame, type_query: list, architecture_query: list, size_query: list, precision_query: list, show_deleted: bool
) -> pd.DataFrame:
# Show all models
# if show_deleted:
# filtered_df = df
# else: # Show only still on the hub models
# filtered_df = df[df[AutoEvalColumn.still_on_hub.name] == True]
filtered_df = df
if type_query is not None:
type_emoji = [t[0] for t in type_query]
filtered_df = filtered_df.loc[df[AutoEvalColumn.model_type_symbol.name].isin(type_emoji)]
if architecture_query is not None:
arch_types = [t for t in architecture_query]
filtered_df = filtered_df.loc[df[AutoEvalColumn.architecture.name].isin(arch_types)]
# filtered_df = filtered_df.loc[df[AutoEvalColumn.architecture.name].isin(architecture_query + ["None"])]
if precision_query is not None:
if AutoEvalColumn.precision.name in df.columns:
filtered_df = filtered_df.loc[df[AutoEvalColumn.precision.name].isin(precision_query + ["None"])]
if size_query is not None:
numeric_interval = pd.IntervalIndex(sorted([NUMERIC_INTERVALS[s] for s in size_query]))
params_column = pd.to_numeric(df[AutoEvalColumn.params.name], errors="coerce")
mask = params_column.apply(lambda x: any(numeric_interval.contains(x)))
filtered_df = filtered_df.loc[mask]
return filtered_df
def change_submit_request_form(model_architecture):
match model_architecture:
case "Encoder":
return (
gr.Textbox(label="Threshold for gliner models", visible=False),
gr.Radio(
choices=["True", "False"],
label="Load GLiNER Tokenizer",
visible=False
),
gr.Dropdown(
choices=[prompt_template.value for prompt_template in PromptTemplateName],
label="Prompt for generation",
multiselect=False,
# value="HTML Highlighted Spans",
interactive=True,
visible=False
)
)
case "Decoder":
return (
gr.Textbox(label="Threshold for gliner models", visible=False),
gr.Radio(
choices=["True", "False"],
label="Load GLiNER Tokenizer",
visible=False
),
gr.Dropdown(
choices=[prompt_template.value for prompt_template in PromptTemplateName],
label="Prompt for generation",
multiselect=False,
# value="HTML Highlighted Spans",
interactive=True,
visible=True
)
)
case "GLiNER Encoder":
return (
gr.Textbox(label="Threshold for gliner models", visible=True),
gr.Radio(
choices=["True", "False"],
label="Load GLiNER Tokenizer",
visible=True
),
gr.Dropdown(
choices=[prompt_template.value for prompt_template in PromptTemplateName],
label="Prompt for generation",
multiselect=False,
# value="HTML Highlighted Spans",
interactive=True,
visible=False
)
)
demo = gr.Blocks(css=custom_css)
with demo:
gr.HTML(TITLE)
gr.HTML(LOGO, elem_classes="logo")
gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
with gr.Tabs(elem_classes="tab-buttons") as tabs:
with gr.TabItem("🏅 NER Datasets", elem_id="llm-benchmark-tab-table", id=0):
with gr.Row():
with gr.Column():
with gr.Row():
search_bar = gr.Textbox(
placeholder=" 🔍 Search for your model (separate multiple queries with `;`) and press ENTER...",
show_label=False,
elem_id="search-bar",
)
with gr.Row():
shown_columns = gr.CheckboxGroup(
choices=[c.name for c in fields(AutoEvalColumn) if not c.hidden and not c.never_hidden and not c.clinical_type_col],
value=[
c.name
for c in fields(AutoEvalColumn)
if c.displayed_by_default and not c.hidden and not c.never_hidden and not c.clinical_type_col
],
label="Select columns to show",
elem_id="column-select",
interactive=True,
)
# with gr.Row():
# deleted_models_visibility = gr.Checkbox(
# value=False, label="Show gated/private/deleted models", interactive=True
# )
with gr.Column(min_width=320):
# with gr.Box(elem_id="box-filter"):
eval_metric = gr.Radio(
choices=["Span Based", "Token Based"],
value = "Span Based",
label="Evaluation Metric",
)
filter_columns_type = gr.CheckboxGroup(
label="Model Types",
choices=[t.to_str() for t in ModelType],
value=[t.to_str() for t in ModelType],
interactive=True,
elem_id="filter-columns-type",
)
filter_columns_architecture = gr.CheckboxGroup(
label="Architecture Types",
choices=[i.value.name for i in ModelArch],
value=[i.value.name for i in ModelArch],
interactive=True,
elem_id="filter-columns-architecture",
)
# filter_columns_size = gr.CheckboxGroup(
# label="Model sizes (in billions of parameters)",
# choices=list(NUMERIC_INTERVALS.keys()),
# value=list(NUMERIC_INTERVALS.keys()),
# interactive=True,
# elem_id="filter-columns-size",
# )
datasets_leaderboard_df, datasets_original_df = update_df(eval_metric.value, shown_columns.value, subset="datasets")
leaderboard_table = gr.components.Dataframe(
value=datasets_leaderboard_df[[c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value],
headers=[c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value,
datatype=TYPES,
elem_id="leaderboard-table",
interactive=False,
visible=True,
)
# Dummy leaderboard for handling the case when the user uses backspace key
hidden_leaderboard_table_for_search = gr.components.Dataframe(
value=datasets_original_df[DATASET_COLS],
headers=DATASET_COLS,
datatype=TYPES,
visible=False,
)
eval_metric.change(
lambda a, b: update_df(a,b, "datasets") ,
inputs=[
eval_metric,
shown_columns,
],
outputs=[
leaderboard_table,
hidden_leaderboard_table_for_search,
]
)
search_bar.submit(
update_table,
[
hidden_leaderboard_table_for_search,
shown_columns,
search_bar,
filter_columns_type,
filter_columns_architecture
],
leaderboard_table,
)
for selector in [
shown_columns,
filter_columns_type,
filter_columns_architecture,
# filter_columns_size,
# deleted_models_visibility,
]:
selector.change(
update_table,
[
hidden_leaderboard_table_for_search,
shown_columns,
search_bar,
filter_columns_type,
filter_columns_architecture,
],
leaderboard_table,
queue=True,
)
with gr.TabItem("🏅 Clinical Types", elem_id="llm-benchmark-tab-table", id=4):
with gr.Row():
with gr.Column():
with gr.Row():
search_bar = gr.Textbox(
placeholder=" 🔍 Search for your model (separate multiple queries with `;`) and press ENTER...",
show_label=False,
elem_id="search-bar",
)
with gr.Row():
shown_columns = gr.CheckboxGroup(
choices=[c.name for c in fields(AutoEvalColumn) if not c.hidden and not c.never_hidden and not c.dataset_task_col],
value=[
c.name
for c in fields(AutoEvalColumn)
if c.displayed_by_default and not c.hidden and not c.never_hidden and not c.dataset_task_col
],
label="Select columns to show",
elem_id="column-select",
interactive=True,
)
# with gr.Row():
# deleted_models_visibility = gr.Checkbox(
# value=False, label="Show gated/private/deleted models", interactive=True
# )
with gr.Column(min_width=320):
eval_metric = gr.Radio(
choices=["Span Based", "Token Based"],
value = "Span Based",
label="Evaluation Metric",
)
# with gr.Box(elem_id="box-filter"):
filter_columns_type = gr.CheckboxGroup(
label="Model Types",
choices=[t.to_str() for t in ModelType],
value=[t.to_str() for t in ModelType],
interactive=True,
elem_id="filter-columns-type",
)
filter_columns_architecture = gr.CheckboxGroup(
label="Architecture Types",
choices=[i.value.name for i in ModelArch],
value=[i.value.name for i in ModelArch],
interactive=True,
elem_id="filter-columns-architecture",
)
# filter_columns_precision = gr.CheckboxGroup(
# label="Precision",
# choices=[i.value.name for i in Precision],
# value=[i.value.name for i in Precision],
# interactive=True,
# elem_id="filter-columns-precision",
# )
# filter_columns_size = gr.CheckboxGroup(
# label="Model sizes (in billions of parameters)",
# choices=list(NUMERIC_INTERVALS.keys()),
# value=list(NUMERIC_INTERVALS.keys()),
# interactive=True,
# elem_id="filter-columns-size",
# )
types_leaderboard_df, types_original_df = update_df(eval_metric.value, shown_columns.value, subset="clinical_types")
leaderboard_table = gr.components.Dataframe(
value=types_leaderboard_df[[c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value],
headers=[c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value,
datatype=TYPES,
elem_id="leaderboard-table",
interactive=False,
visible=True,
)
# Dummy leaderboard for handling the case when the user uses backspace key
hidden_leaderboard_table_for_search = gr.components.Dataframe(
value=types_original_df[Clinical_TYPES_COLS],
headers=Clinical_TYPES_COLS,
datatype=TYPES,
visible=False,
)
eval_metric.change(
fn=lambda a, b: update_df(a,b, "clinical_types"),
inputs=[
eval_metric,
shown_columns,
],
outputs=[
leaderboard_table,
hidden_leaderboard_table_for_search
]
)
search_bar.submit(
update_table,
[
hidden_leaderboard_table_for_search,
shown_columns,
search_bar,
filter_columns_type,
filter_columns_architecture,
],
leaderboard_table,
)
for selector in [
shown_columns,
filter_columns_type,
filter_columns_architecture,
# filter_columns_precision,
# filter_columns_size,
# deleted_models_visibility,
]:
selector.change(
update_table,
[
hidden_leaderboard_table_for_search,
shown_columns,
search_bar,
filter_columns_type,
filter_columns_architecture,
],
leaderboard_table,
queue=True,
)
with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2):
gr.Markdown(LLM_BENCHMARKS_TEXT_1, elem_classes="markdown-text")
gr.HTML(EVALUATION_EXAMPLE_IMG, elem_classes="logo")
gr.Markdown(LLM_BENCHMARKS_TEXT_2, elem_classes="markdown-text")
gr.HTML(ENTITY_DISTRIBUTION_IMG, elem_classes="logo")
gr.Markdown(LLM_BENCHMARKS_TEXT_3, elem_classes="markdown-text")
with gr.TabItem("🚀 Submit here! ", elem_id="llm-benchmark-tab-table", id=3):
with gr.Column():
with gr.Row():
gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text")
with gr.Column():
with gr.Accordion(
f"✅ Finished Evaluations ({len(finished_eval_queue_df)})",
open=False,
):
with gr.Row():
finished_eval_table = gr.components.Dataframe(
value=finished_eval_queue_df,
headers=EVAL_COLS,
datatype=EVAL_TYPES,
row_count=5,
)
with gr.Accordion(
f"🔄 Running Evaluation Queue ({len(running_eval_queue_df)})",
open=False,
):
with gr.Row():
running_eval_table = gr.components.Dataframe(
value=running_eval_queue_df,
headers=EVAL_COLS,
datatype=EVAL_TYPES,
row_count=5,
)
with gr.Accordion(
f"⏳ Pending Evaluation Queue ({len(pending_eval_queue_df)})",
open=False,
):
with gr.Row():
pending_eval_table = gr.components.Dataframe(
value=pending_eval_queue_df,
headers=EVAL_COLS,
datatype=EVAL_TYPES,
row_count=5,
)
with gr.Row():
gr.Markdown("# ✉️✨ Submit your model here!", elem_classes="markdown-text")
with gr.Row():
with gr.Column():
model_name_textbox = gr.Textbox(label="Model name")
revision_name_textbox = gr.Textbox(label="Revision commit", placeholder="main")
model_arch = gr.Radio(
choices=[t.to_str(" : ") for t in ModelArch if t != ModelArch.Unknown],
label="Model Architecture",
)
model_type = gr.Dropdown(
choices=[t.to_str(" : ") for t in ModelType if t != ModelType.Unknown],
label="Model type",
multiselect=False,
value=None,
interactive=True,
)
with gr.Column():
label_normalization_map = gr.Textbox(lines=6, label="Label Normalization Map", placeholder=PLACEHOLDER_DATASET_WISE_NORMALIZATION_CONFIG)
gliner_threshold = gr.Textbox(label="Threshold for GLiNER models", visible=False)
gliner_tokenizer_bool = gr.Radio(
choices=["True", "False"],
label="Load GLiNER Tokenizer",
visible=False
)
prompt_name = gr.Dropdown(
choices=[prompt_template.value for prompt_template in PromptTemplateName],
label="Prompt for generation",
multiselect=False,
value="HTML Highlighted Spans",
interactive=True,
visible=False
)# should be a dropdown
# parsing_function - this is tied to the prompt & therefore does not need to be specified
# generation_parameters = gr.Textbox(label="Generation params in json format") just default for now
model_arch.change(fn=change_submit_request_form, inputs=model_arch, outputs=[
gliner_threshold,
gliner_tokenizer_bool,
prompt_name])
submit_button = gr.Button("Submit Eval")
submission_result = gr.Markdown()
submit_button.click(
add_new_eval,
[
model_name_textbox,
# base_model_name_textbox,
revision_name_textbox,
model_arch,
label_normalization_map,
gliner_threshold,
gliner_tokenizer_bool,
prompt_name,
# weight_type,
model_type,
],
submission_result,
)
with gr.Row():
with gr.Accordion("📙 Citation", open=False):
citation_button = gr.Textbox(
value=CITATION_BUTTON_TEXT,
label=CITATION_BUTTON_LABEL,
lines=20,
elem_id="citation-button",
show_copy_button=True,
)
scheduler = BackgroundScheduler()
scheduler.add_job(restart_space, "interval", seconds=1800)
scheduler.start()
demo.queue(default_concurrency_limit=40).launch(allowed_paths=['./assets/'])