Update app.py

app.py

@@ -1,115 +1,89 @@
-# A simple script that loops over all public models and get their libary_name
+from datasets import load_dataset
+from huggingface_hub import ModelCard
+from huggingface_hub import HfApi
+
 import gradio as gr
 import pandas as pd
-import numpy as np
-
-from collections import Counter
-
-from huggingface_hub import HfApi
-from datasets import load_dataset
 
 api = HfApi()
-list_models = api.list_models()
-
-def fetch_dataset_and_init():
-    dataset = load_dataset("librarian-bots/model_cards_with_metadata", split="train")
-    library_names = dataset["library_name"]
-
-    string_counts = Counter(library_names)
-    string_counts_series = pd.Series(string_counts)
-
-    # Sort the series in descending order
-    df = string_counts_series.sort_values(ascending=False).to_frame()
-    df.columns = ["count"]
-    df = df.reset_index()
-    df = df.rename(columns={"index": "library_name"})
-
-    df.replace(to_replace=[None], value="No library_name", inplace=True)
-    df_log = df.copy()
-    df_log['count'] = np.log(df_log['count'])
-
-    return df, df_log
-
-df, df_log = fetch_dataset_and_init()
-
-def get_current_nb_models():
-    # We need this hack since `list_models` returns a generator..
-    total_models = sum(1 for _ in list_models)
-    diff_models = total_models - df["count"].sum()
-    return str(diff_models)
-
-plot_height = 512
-plot_width = 1512
-select_box = ["all"]
-top_k = len(df)
-
-def bar_plot_fn(display, top_k, select_box):
-    if display == "simple":
-        if select_box is not None and ("all" not in select_box or select_box != ["all"]):
-            current_df = df[df["library_name"].isin(select_box)]
-        else:
-            current_df = df[:top_k]
-
-        return gr.BarPlot(
-            current_df,
-            x="library_name",
-            y="count",
-            tooltip=["library_name", "count"],
-            height=plot_height,
-            width=plot_width
-        )
-    elif display == "log":
-        if select_box is not None and ("all" not in select_box or select_box != ["all"]):
-            current_df = df_log[df_log["library_name"].isin(select_box)]
-        else:
-            current_df = df_log[:top_k]
-
-        return gr.BarPlot(
-            current_df,
-            x="library_name",
-            y="count",
-            tooltip=["library_name", "count"],
-            height=plot_height,
-            width=plot_width
-        )
 
+repo_id = "librarian-bots/model_cards_with_metadata"
+
+dataset = load_dataset(repo_id, split='train')
+dataset = dataset.filter(lambda x: x['library_name']=='transformers')
+
+list_commits = api.list_repo_commits(repo_id, repo_type="dataset")
+commits_date_dict = {commit.created_at.strftime("%m/%d/%Y"):commit.commit_id for commit in list_commits}
+current_date = "latest"
+
+def get_data(commit_date="latest"):
+    ds_kwargs = {}
+    if commit_date != "latest":
+        current_date = commit_date
+        commit_id = commits_date_dict[commit_date]
+        ds_kwargs = {"revision": commit_id}
+
+    dataset = load_dataset(repo_id, split='train', **ds_kwargs)
+    dataset = dataset.filter(lambda x: x['library_name']=='transformers')
+
+    def pipeline_tag_not_in_card(card):
+        try:
+            model_card_data = ModelCard(card).data
+            if model_card_data.library_name is None:
+                return True
+            return False
+        except AttributeError:
+            return False
+        except Exception:
+            return False
+
+    ds = dataset.map(lambda x: {"missing_library_name": pipeline_tag_not_in_card(x['card'])}, num_proc=4)
+
+
+    data = pd.DataFrame(
+        {
+            "name": ["Total Number of transformers Model", "Total number of models with missing 'library_name: transformers' in model card."],
+            "count": [len(ds), sum(ds["missing_library_name"])],
+        }
+    )
+    return data
+
+def fetch_fn(commit_date="latest"):
+    data = get_data(commit_date=commit_date)
+    return gr.BarPlot(
+        data,
+        x="name",
+        y="count",
+        title="Count of Model cards with the correct library_name tag",
+        height=256,
+        width=1024,
+        tooltip=["name", "count"],
+        vertical=False
+    )
+
+data = get_data()
 
 with gr.Blocks() as bar_plot:
     with gr.Column():
+        with gr.Row():
+            plot = gr.BarPlot(
+                data,
+                x="name",
+                y="count",
+                title=f"Count of Model cards with the correct library_name tag at the date {current_date}",
+                height=256,
+                width=1024,
+                tooltip=["name", "count"],
+                vertical=False
+            )
         with gr.Column():
             display = gr.Dropdown(
-                choices=[
-                    "simple",
-                    "log",
-                ],
-                value="simple",
+                choices=list(commits_date_dict.keys()),
+                value="latest",
                 label="Type of Bar Plot",
+                
             )
-            top_k = gr.Slider(
-                label="Select top-K most used library_name (This leads to a no-op if you selected something else than 'all' in the columns below)",
-                value=len(df),
-                minimum=1,
-                maximum=len(df),
-                step=1,
-            )
-        with gr.Column():
-            plot = gr.BarPlot()
-
-        with gr.Row():
-            fetch_button = gr.Button(value="Fetch current number of models without model cards (takes up to 1min to fetch everything)")
-            text_box = gr.Textbox(value="", label="Number of models without model cards")
-
-        with gr.Column():
-            select_box = gr.Dropdown(
-                ["all"] + df["library_name"].tolist(), value=["all"], multiselect=True, label="Libraries to inspect", info="Select specific libraries to inspect"
-            )
-
-
-    top_k.change(bar_plot_fn, inputs=[display, top_k, select_box], outputs=plot)
-    display.change(bar_plot_fn, inputs=[display, top_k, select_box], outputs=plot)
-    select_box.change(bar_plot_fn, inputs=[display, top_k, select_box], outputs=plot)
-
-    fetch_button.click(get_current_nb_models, outputs=[text_box])
-    bar_plot.load(fn=bar_plot_fn, inputs=[display, top_k], outputs=plot)
+        
+        display.change(fetch_fn, inputs=display, outputs=plot)
 
 bar_plot.launch()