Update app.py
Browse files
app.py
CHANGED
@@ -6,9 +6,9 @@ import pyarrow.parquet as pq
|
|
6 |
import os
|
7 |
import requests
|
8 |
from io import BytesIO
|
9 |
-
import
|
10 |
|
11 |
-
# Define pipeline tags
|
12 |
PIPELINE_TAGS = [
|
13 |
'text-generation',
|
14 |
'text-to-image',
|
@@ -60,57 +60,59 @@ MODEL_SIZE_RANGES = {
|
|
60 |
}
|
61 |
|
62 |
# Filter functions for tags - keeping the same from provided code
|
63 |
-
def is_audio_speech(
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
|
|
69 |
|
70 |
-
def is_music(
|
71 |
-
|
72 |
-
return
|
73 |
|
74 |
-
def is_robotics(
|
75 |
-
|
76 |
-
return
|
77 |
|
78 |
-
def is_biomed(
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
|
83 |
-
def is_timeseries(
|
84 |
-
|
85 |
-
return
|
86 |
|
87 |
-
def is_science(
|
88 |
-
|
89 |
-
return
|
90 |
|
91 |
-
def is_video(
|
92 |
-
|
93 |
-
return
|
94 |
|
95 |
-
def is_image(
|
96 |
-
|
97 |
-
return
|
98 |
|
99 |
-
def is_text(
|
100 |
-
|
101 |
-
return
|
102 |
|
103 |
# Add model size filter function
|
104 |
-
def is_in_size_range(
|
105 |
if size_range is None:
|
106 |
return True
|
107 |
|
108 |
min_size, max_size = MODEL_SIZE_RANGES[size_range]
|
109 |
|
110 |
# Get model size in GB from safetensors total (if available)
|
111 |
-
|
|
|
112 |
# Convert bytes to GB
|
113 |
-
size_gb =
|
114 |
return min_size <= size_gb < max_size
|
115 |
|
116 |
return False
|
@@ -127,251 +129,421 @@ TAG_FILTER_FUNCS = {
|
|
127 |
"Sciences": is_science,
|
128 |
}
|
129 |
|
130 |
-
def
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
sorted_stats = sorted(
|
141 |
-
[(
|
142 |
-
org_id,
|
143 |
-
sum(model[count_by] for model in models if combined_filter(model))
|
144 |
-
) for org_id, models in org_stats.items()],
|
145 |
-
key=lambda x: x[1],
|
146 |
-
reverse=True,
|
147 |
-
)
|
148 |
|
149 |
-
#
|
150 |
-
|
151 |
-
|
|
|
152 |
|
153 |
-
|
154 |
-
|
155 |
-
for org, st in res:
|
156 |
-
if org == "Others...":
|
157 |
-
res_plot_df += [("Others...", "other", st * 100 / total_st if total_st > 0 else 0)]
|
158 |
-
else:
|
159 |
-
for model in org_stats[org]:
|
160 |
-
if combined_filter(model):
|
161 |
-
res_plot_df += [(org, model["id"], model[count_by] * 100 / total_st if total_st > 0 else 0)]
|
162 |
|
163 |
-
|
164 |
-
|
165 |
-
def
|
166 |
-
|
|
|
|
|
167 |
|
168 |
-
#
|
169 |
-
|
170 |
-
if tag_filter:
|
171 |
-
filter_func = TAG_FILTER_FUNCS[tag_filter]
|
172 |
-
elif pipeline_filter:
|
173 |
-
filter_func = lambda dct: dct.get("pipeline_tag", None) and dct.get("pipeline_tag", "") == pipeline_filter
|
174 |
-
else:
|
175 |
-
filter_func = lambda dct: True
|
176 |
|
177 |
-
#
|
178 |
-
|
|
|
179 |
|
180 |
-
#
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
)
|
188 |
|
189 |
-
|
|
|
190 |
|
191 |
-
#
|
192 |
-
|
193 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
194 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
195 |
fig.update_layout(
|
196 |
margin=dict(t=50, l=25, r=25, b=25)
|
197 |
)
|
198 |
|
|
|
|
|
|
|
|
|
|
|
|
|
199 |
return fig
|
200 |
|
201 |
-
def
|
202 |
-
"""Download
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
203 |
try:
|
204 |
# Create a cache directory
|
205 |
if not os.path.exists('data'):
|
206 |
os.makedirs('data')
|
207 |
|
208 |
# Check if we have cached data
|
209 |
-
if os.path.exists('data/processed_models.
|
210 |
-
|
211 |
-
|
212 |
-
|
|
|
|
|
213 |
|
214 |
# URL to the models.parquet file
|
215 |
url = "https://huggingface.co/datasets/cfahlgren1/hub-stats/resolve/main/models.parquet"
|
216 |
|
|
|
|
|
217 |
print(f"Downloading models data from {url}...")
|
218 |
-
|
219 |
-
|
220 |
-
|
|
|
|
|
|
|
221 |
|
222 |
# Read the parquet file
|
223 |
-
table = pq.read_table(BytesIO(
|
224 |
df = table.to_pandas()
|
225 |
|
226 |
print(f"Downloaded {len(df)} models")
|
227 |
|
228 |
-
|
229 |
-
|
230 |
|
231 |
-
|
232 |
-
|
233 |
-
|
234 |
-
|
235 |
-
if '/' in model_id:
|
236 |
-
org_id = model_id.split('/')[0]
|
237 |
-
else:
|
238 |
-
org_id = "unaffiliated"
|
239 |
-
|
240 |
-
# Create model entry with needed fields
|
241 |
-
model_entry = {
|
242 |
-
"id": model_id,
|
243 |
-
"downloads": row.get('downloads', 0),
|
244 |
-
"likes": row.get('likes', 0),
|
245 |
-
"pipeline_tag": row.get('pipeline_tag'),
|
246 |
-
"tags": row.get('tags', []),
|
247 |
-
}
|
248 |
-
|
249 |
-
# Add safetensors information if available
|
250 |
-
if 'safetensors' in row and row['safetensors']:
|
251 |
-
if isinstance(row['safetensors'], dict) and 'total' in row['safetensors']:
|
252 |
-
model_entry["safetensors"] = {"total": row['safetensors']['total']}
|
253 |
-
elif isinstance(row['safetensors'], str):
|
254 |
-
# Try to parse JSON string
|
255 |
try:
|
256 |
-
|
257 |
-
if isinstance(safetensors, dict) and 'total' in safetensors:
|
258 |
-
model_entry["safetensors"] = {"total": safetensors['total']}
|
259 |
except:
|
260 |
-
|
|
|
261 |
|
262 |
-
|
263 |
-
|
264 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
265 |
|
266 |
-
|
267 |
|
268 |
# Cache the processed data
|
269 |
-
|
270 |
-
|
|
|
|
|
|
|
|
|
271 |
|
272 |
-
return
|
273 |
|
274 |
except Exception as e:
|
275 |
print(f"Error downloading or processing data: {e}")
|
|
|
|
|
276 |
# Return sample data for testing if real data unavailable
|
277 |
return create_sample_data()
|
278 |
|
279 |
-
def create_sample_data():
|
280 |
"""Create sample data for testing when real data is unavailable"""
|
281 |
print("Creating sample data for testing...")
|
282 |
|
283 |
-
|
284 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
285 |
|
286 |
-
|
287 |
-
|
288 |
-
|
|
|
|
|
|
|
|
|
|
|
289 |
|
290 |
for i in range(num_models):
|
291 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
292 |
|
293 |
-
|
294 |
-
|
295 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
296 |
|
297 |
-
#
|
298 |
-
tags = [pipeline_tag
|
299 |
|
300 |
-
|
301 |
-
|
302 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
303 |
|
304 |
-
#
|
305 |
-
|
|
|
|
|
|
|
306 |
|
307 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
308 |
"id": model_id,
|
309 |
"downloads": downloads,
|
|
|
310 |
"likes": likes,
|
311 |
"pipeline_tag": pipeline_tag,
|
312 |
"tags": tags,
|
313 |
-
"safetensors": {"total":
|
314 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
315 |
|
316 |
-
|
|
|
|
|
|
|
317 |
|
318 |
# Create Gradio interface
|
319 |
with gr.Blocks() as demo:
|
320 |
-
models_data = gr.State(
|
321 |
-
|
322 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
323 |
gr.Markdown("""
|
324 |
-
|
325 |
|
326 |
This app shows how different organizations contribute to the HuggingFace ecosystem with their models.
|
327 |
Use the filters to explore models by different metrics, tags, pipelines, and model sizes.
|
|
|
|
|
328 |
""")
|
329 |
|
330 |
-
with gr.Row():
|
331 |
with gr.Column(scale=1):
|
332 |
count_by_dropdown = gr.Dropdown(
|
333 |
label="Metric",
|
334 |
-
choices=["downloads", "likes"],
|
335 |
-
value="downloads"
|
|
|
336 |
)
|
337 |
|
338 |
filter_choice_radio = gr.Radio(
|
339 |
-
label="Filter
|
340 |
choices=["None", "Tag Filter", "Pipeline Filter"],
|
341 |
-
value="None"
|
|
|
342 |
)
|
343 |
|
344 |
tag_filter_dropdown = gr.Dropdown(
|
345 |
label="Select Tag",
|
346 |
choices=list(TAG_FILTER_FUNCS.keys()),
|
347 |
value=None,
|
348 |
-
visible=False
|
|
|
349 |
)
|
350 |
|
351 |
pipeline_filter_dropdown = gr.Dropdown(
|
352 |
label="Select Pipeline Tag",
|
353 |
choices=PIPELINE_TAGS,
|
354 |
value=None,
|
355 |
-
visible=False
|
|
|
356 |
)
|
357 |
|
358 |
size_filter_dropdown = gr.Dropdown(
|
359 |
label="Model Size Filter",
|
360 |
choices=["None"] + list(MODEL_SIZE_RANGES.keys()),
|
361 |
-
value="None"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
362 |
)
|
363 |
|
364 |
-
generate_plot_button = gr.Button("Generate Plot")
|
365 |
|
366 |
with gr.Column(scale=3):
|
367 |
plot_output = gr.Plot()
|
|
|
368 |
|
369 |
-
def generate_plot_on_click(count_by, filter_choice, tag_filter, pipeline_filter, size_filter,
|
370 |
-
print(f"Generating plot with: Metric={count_by}, Filter={filter_choice}, Tag={tag_filter}, Pipeline={pipeline_filter}, Size={size_filter}")
|
371 |
|
372 |
-
if
|
373 |
-
|
374 |
-
return None
|
375 |
|
376 |
selected_tag_filter = None
|
377 |
selected_pipeline_filter = None
|
@@ -384,15 +556,47 @@ with gr.Blocks() as demo:
|
|
384 |
|
385 |
if size_filter != "None":
|
386 |
selected_size_filter = size_filter
|
387 |
-
|
388 |
-
|
|
|
|
|
389 |
count_by=count_by,
|
390 |
-
|
391 |
tag_filter=selected_tag_filter,
|
392 |
pipeline_filter=selected_pipeline_filter,
|
393 |
-
|
394 |
)
|
395 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
396 |
|
397 |
def update_filter_visibility(filter_choice):
|
398 |
if filter_choice == "Tag Filter":
|
@@ -408,11 +612,17 @@ with gr.Blocks() as demo:
|
|
408 |
outputs=[tag_filter_dropdown, pipeline_filter_dropdown]
|
409 |
)
|
410 |
|
411 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
412 |
demo.load(
|
413 |
-
fn=
|
414 |
inputs=[],
|
415 |
-
outputs=[models_data]
|
416 |
)
|
417 |
|
418 |
# Button click event to generate plot
|
@@ -424,9 +634,10 @@ with gr.Blocks() as demo:
|
|
424 |
tag_filter_dropdown,
|
425 |
pipeline_filter_dropdown,
|
426 |
size_filter_dropdown,
|
|
|
427 |
models_data
|
428 |
],
|
429 |
-
outputs=[plot_output]
|
430 |
)
|
431 |
|
432 |
|
|
|
6 |
import os
|
7 |
import requests
|
8 |
from io import BytesIO
|
9 |
+
import numpy as np
|
10 |
|
11 |
+
# Define pipeline tags from the provided code
|
12 |
PIPELINE_TAGS = [
|
13 |
'text-generation',
|
14 |
'text-to-image',
|
|
|
60 |
}
|
61 |
|
62 |
# Filter functions for tags - keeping the same from provided code
|
63 |
+
def is_audio_speech(model_dict):
|
64 |
+
tags = model_dict.get("tags", [])
|
65 |
+
pipeline_tag = model_dict.get("pipeline_tag", "")
|
66 |
+
|
67 |
+
return (pipeline_tag and ("audio" in pipeline_tag.lower() or "speech" in pipeline_tag.lower())) or \
|
68 |
+
any("audio" in tag.lower() for tag in tags) or \
|
69 |
+
any("speech" in tag.lower() for tag in tags)
|
70 |
|
71 |
+
def is_music(model_dict):
|
72 |
+
tags = model_dict.get("tags", [])
|
73 |
+
return any("music" in tag.lower() for tag in tags)
|
74 |
|
75 |
+
def is_robotics(model_dict):
|
76 |
+
tags = model_dict.get("tags", [])
|
77 |
+
return any("robot" in tag.lower() for tag in tags)
|
78 |
|
79 |
+
def is_biomed(model_dict):
|
80 |
+
tags = model_dict.get("tags", [])
|
81 |
+
return any("bio" in tag.lower() for tag in tags) or \
|
82 |
+
any("medic" in tag.lower() for tag in tags)
|
83 |
|
84 |
+
def is_timeseries(model_dict):
|
85 |
+
tags = model_dict.get("tags", [])
|
86 |
+
return any("series" in tag.lower() for tag in tags)
|
87 |
|
88 |
+
def is_science(model_dict):
|
89 |
+
tags = model_dict.get("tags", [])
|
90 |
+
return any("science" in tag.lower() and "bigscience" not in tag for tag in tags)
|
91 |
|
92 |
+
def is_video(model_dict):
|
93 |
+
tags = model_dict.get("tags", [])
|
94 |
+
return any("video" in tag.lower() for tag in tags)
|
95 |
|
96 |
+
def is_image(model_dict):
|
97 |
+
tags = model_dict.get("tags", [])
|
98 |
+
return any("image" in tag.lower() for tag in tags)
|
99 |
|
100 |
+
def is_text(model_dict):
|
101 |
+
tags = model_dict.get("tags", [])
|
102 |
+
return any("text" in tag.lower() for tag in tags)
|
103 |
|
104 |
# Add model size filter function
|
105 |
+
def is_in_size_range(model_dict, size_range):
|
106 |
if size_range is None:
|
107 |
return True
|
108 |
|
109 |
min_size, max_size = MODEL_SIZE_RANGES[size_range]
|
110 |
|
111 |
# Get model size in GB from safetensors total (if available)
|
112 |
+
safetensors = model_dict.get("safetensors", None)
|
113 |
+
if safetensors and isinstance(safetensors, dict) and "total" in safetensors:
|
114 |
# Convert bytes to GB
|
115 |
+
size_gb = safetensors["total"] / (1024 * 1024 * 1024)
|
116 |
return min_size <= size_gb < max_size
|
117 |
|
118 |
return False
|
|
|
129 |
"Sciences": is_science,
|
130 |
}
|
131 |
|
132 |
+
def extract_org_from_id(model_id):
|
133 |
+
"""Extract organization name from model ID"""
|
134 |
+
if "/" in model_id:
|
135 |
+
return model_id.split("/")[0]
|
136 |
+
return "unaffiliated"
|
137 |
+
|
138 |
+
def make_treemap_data(df, count_by, top_k=25, tag_filter=None, pipeline_filter=None, size_filter=None):
|
139 |
+
"""Process DataFrame into treemap format with filters applied"""
|
140 |
+
# Create a copy to avoid modifying the original
|
141 |
+
filtered_df = df.copy()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
142 |
|
143 |
+
# Apply filters
|
144 |
+
if tag_filter and tag_filter in TAG_FILTER_FUNCS:
|
145 |
+
filter_func = TAG_FILTER_FUNCS[tag_filter]
|
146 |
+
filtered_df = filtered_df[filtered_df.apply(filter_func, axis=1)]
|
147 |
|
148 |
+
if pipeline_filter:
|
149 |
+
filtered_df = filtered_df[filtered_df["pipeline_tag"] == pipeline_filter]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
150 |
|
151 |
+
if size_filter and size_filter in MODEL_SIZE_RANGES:
|
152 |
+
# Create a function to check if a model is in the size range
|
153 |
+
def check_size(row):
|
154 |
+
return is_in_size_range(row, size_filter)
|
155 |
+
|
156 |
+
filtered_df = filtered_df[filtered_df.apply(check_size, axis=1)]
|
157 |
|
158 |
+
# Add organization column
|
159 |
+
filtered_df["organization"] = filtered_df["id"].apply(extract_org_from_id)
|
|
|
|
|
|
|
|
|
|
|
|
|
160 |
|
161 |
+
# Aggregate by organization
|
162 |
+
org_totals = filtered_df.groupby("organization")[count_by].sum().reset_index()
|
163 |
+
org_totals = org_totals.sort_values(by=count_by, ascending=False)
|
164 |
|
165 |
+
# Get top organizations
|
166 |
+
top_orgs = org_totals.head(top_k)["organization"].tolist()
|
167 |
+
|
168 |
+
# Filter to only include models from top organizations
|
169 |
+
filtered_df = filtered_df[filtered_df["organization"].isin(top_orgs)]
|
170 |
+
|
171 |
+
# Prepare data for treemap
|
172 |
+
treemap_data = filtered_df[["id", "organization", count_by]].copy()
|
173 |
|
174 |
+
# Add a root node
|
175 |
+
treemap_data["root"] = "models"
|
176 |
|
177 |
+
# Ensure numeric values
|
178 |
+
treemap_data[count_by] = pd.to_numeric(treemap_data[count_by], errors="coerce").fillna(0)
|
179 |
+
|
180 |
+
return treemap_data
|
181 |
+
|
182 |
+
def create_treemap(treemap_data, count_by, title=None):
|
183 |
+
"""Create a Plotly treemap from the prepared data"""
|
184 |
+
if treemap_data.empty:
|
185 |
+
# Create an empty figure with a message
|
186 |
+
fig = px.treemap(
|
187 |
+
names=["No data matches the selected filters"],
|
188 |
+
values=[1]
|
189 |
+
)
|
190 |
+
fig.update_layout(
|
191 |
+
title="No data matches the selected filters",
|
192 |
+
margin=dict(t=50, l=25, r=25, b=25)
|
193 |
+
)
|
194 |
+
return fig
|
195 |
|
196 |
+
# Create the treemap
|
197 |
+
fig = px.treemap(
|
198 |
+
treemap_data,
|
199 |
+
path=["root", "organization", "id"],
|
200 |
+
values=count_by,
|
201 |
+
title=title or f"HuggingFace Models - {count_by.capitalize()} by Organization"
|
202 |
+
)
|
203 |
+
|
204 |
+
# Update layout
|
205 |
fig.update_layout(
|
206 |
margin=dict(t=50, l=25, r=25, b=25)
|
207 |
)
|
208 |
|
209 |
+
# Update traces for better readability
|
210 |
+
fig.update_traces(
|
211 |
+
textinfo="label+value+percent root",
|
212 |
+
hovertemplate="<b>%{label}</b><br>%{value:,} " + count_by + "<br>%{percentRoot:.2%} of total<extra></extra>"
|
213 |
+
)
|
214 |
+
|
215 |
return fig
|
216 |
|
217 |
+
def download_with_progress(url, progress=None):
|
218 |
+
"""Download a file with progress tracking"""
|
219 |
+
response = requests.get(url, stream=True)
|
220 |
+
total_size = int(response.headers.get('content-length', 0))
|
221 |
+
block_size = 1024 # 1 Kibibyte
|
222 |
+
data = BytesIO()
|
223 |
+
|
224 |
+
if total_size == 0:
|
225 |
+
# If content length is unknown, we can't show accurate progress
|
226 |
+
if progress:
|
227 |
+
progress(0, "Starting download...")
|
228 |
+
|
229 |
+
for chunk in response.iter_content(block_size):
|
230 |
+
data.write(chunk)
|
231 |
+
if progress:
|
232 |
+
progress(0, f"Downloading... (unknown size)")
|
233 |
+
else:
|
234 |
+
downloaded = 0
|
235 |
+
for chunk in response.iter_content(block_size):
|
236 |
+
downloaded += len(chunk)
|
237 |
+
data.write(chunk)
|
238 |
+
if progress:
|
239 |
+
percent = int(100 * downloaded / total_size)
|
240 |
+
progress(percent / 100, f"Downloading... {percent}% ({downloaded//(1024*1024)}MB/{total_size//(1024*1024)}MB)")
|
241 |
+
|
242 |
+
return data.getvalue()
|
243 |
+
|
244 |
+
def download_and_process_models(progress=None):
|
245 |
+
"""Download and process the models data from HuggingFace dataset with progress tracking"""
|
246 |
try:
|
247 |
# Create a cache directory
|
248 |
if not os.path.exists('data'):
|
249 |
os.makedirs('data')
|
250 |
|
251 |
# Check if we have cached data
|
252 |
+
if os.path.exists('data/processed_models.parquet'):
|
253 |
+
if progress:
|
254 |
+
progress(1.0, "Loading from cache...")
|
255 |
+
print("Loading models from cache...")
|
256 |
+
df = pd.read_parquet('data/processed_models.parquet')
|
257 |
+
return df
|
258 |
|
259 |
# URL to the models.parquet file
|
260 |
url = "https://huggingface.co/datasets/cfahlgren1/hub-stats/resolve/main/models.parquet"
|
261 |
|
262 |
+
if progress:
|
263 |
+
progress(0.0, "Starting download...")
|
264 |
print(f"Downloading models data from {url}...")
|
265 |
+
|
266 |
+
# Download with progress tracking
|
267 |
+
file_content = download_with_progress(url, progress)
|
268 |
+
|
269 |
+
if progress:
|
270 |
+
progress(0.9, "Parsing parquet file...")
|
271 |
|
272 |
# Read the parquet file
|
273 |
+
table = pq.read_table(BytesIO(file_content))
|
274 |
df = table.to_pandas()
|
275 |
|
276 |
print(f"Downloaded {len(df)} models")
|
277 |
|
278 |
+
if progress:
|
279 |
+
progress(0.95, "Processing data...")
|
280 |
|
281 |
+
# Process the safetensors column if it's a string (JSON)
|
282 |
+
if 'safetensors' in df.columns:
|
283 |
+
def parse_safetensors(val):
|
284 |
+
if isinstance(val, str):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
285 |
try:
|
286 |
+
return json.loads(val)
|
|
|
|
|
287 |
except:
|
288 |
+
return None
|
289 |
+
return val
|
290 |
|
291 |
+
df['safetensors'] = df['safetensors'].apply(parse_safetensors)
|
292 |
+
|
293 |
+
# Process the tags column if needed
|
294 |
+
if 'tags' in df.columns and not isinstance(df['tags'].iloc[0], list):
|
295 |
+
def parse_tags(val):
|
296 |
+
if isinstance(val, str):
|
297 |
+
try:
|
298 |
+
return json.loads(val)
|
299 |
+
except:
|
300 |
+
return []
|
301 |
+
return val if isinstance(val, list) else []
|
302 |
|
303 |
+
df['tags'] = df['tags'].apply(parse_tags)
|
304 |
|
305 |
# Cache the processed data
|
306 |
+
if progress:
|
307 |
+
progress(0.98, "Saving to cache...")
|
308 |
+
df.to_parquet('data/processed_models.parquet')
|
309 |
+
|
310 |
+
if progress:
|
311 |
+
progress(1.0, "Data ready!")
|
312 |
|
313 |
+
return df
|
314 |
|
315 |
except Exception as e:
|
316 |
print(f"Error downloading or processing data: {e}")
|
317 |
+
if progress:
|
318 |
+
progress(1.0, "Using sample data (download failed)")
|
319 |
# Return sample data for testing if real data unavailable
|
320 |
return create_sample_data()
|
321 |
|
322 |
+
def create_sample_data(progress=None):
|
323 |
"""Create sample data for testing when real data is unavailable"""
|
324 |
print("Creating sample data for testing...")
|
325 |
|
326 |
+
if progress:
|
327 |
+
progress(0.3, "Creating sample data...")
|
328 |
+
|
329 |
+
# Sample organizations
|
330 |
+
orgs = ['openai', 'meta', 'google', 'microsoft', 'anthropic', 'nvidia', 'huggingface',
|
331 |
+
'deepseek-ai', 'stability-ai', 'mistralai', 'cerebras', 'databricks', 'together',
|
332 |
+
'facebook', 'amazon', 'deepmind', 'cohere', 'nvidia', 'bigscience', 'eleutherai']
|
333 |
+
|
334 |
+
# Common model name formats
|
335 |
+
model_name_patterns = [
|
336 |
+
"model-{size}-{version}",
|
337 |
+
"{prefix}-{size}b",
|
338 |
+
"{prefix}-{size}b-{variant}",
|
339 |
+
"llama-{size}b-{variant}",
|
340 |
+
"gpt-{variant}-{size}b",
|
341 |
+
"{prefix}-instruct-{size}b",
|
342 |
+
"{prefix}-chat-{size}b",
|
343 |
+
"{prefix}-coder-{size}b",
|
344 |
+
"stable-diffusion-{version}",
|
345 |
+
"whisper-{size}",
|
346 |
+
"bert-{size}-{variant}",
|
347 |
+
"roberta-{size}",
|
348 |
+
"t5-{size}",
|
349 |
+
"{prefix}-vision-{size}b"
|
350 |
+
]
|
351 |
+
|
352 |
+
# Common name parts
|
353 |
+
prefixes = ["falcon", "llama", "mistral", "gpt", "phi", "gemma", "qwen", "yi", "mpt", "bloom"]
|
354 |
+
sizes = ["7", "13", "34", "70", "1", "3", "7b", "13b", "70b", "8b", "2b", "1b", "0.5b", "small", "base", "large", "huge"]
|
355 |
+
variants = ["chat", "instruct", "base", "v1.0", "v2", "beta", "turbo", "fast", "xl", "xxl"]
|
356 |
|
357 |
+
# Generate sample data
|
358 |
+
data = []
|
359 |
+
total_models = sum(np.random.randint(5, 20) for _ in orgs)
|
360 |
+
models_created = 0
|
361 |
+
|
362 |
+
for org_idx, org in enumerate(orgs):
|
363 |
+
# Create 5-20 models per organization
|
364 |
+
num_models = np.random.randint(5, 20)
|
365 |
|
366 |
for i in range(num_models):
|
367 |
+
# Create realistic model name
|
368 |
+
pattern = np.random.choice(model_name_patterns)
|
369 |
+
prefix = np.random.choice(prefixes)
|
370 |
+
size = np.random.choice(sizes)
|
371 |
+
version = f"v{np.random.randint(1, 4)}"
|
372 |
+
variant = np.random.choice(variants)
|
373 |
+
|
374 |
+
model_name = pattern.format(
|
375 |
+
prefix=prefix,
|
376 |
+
size=size,
|
377 |
+
version=version,
|
378 |
+
variant=variant
|
379 |
+
)
|
380 |
|
381 |
+
model_id = f"{org}/{model_name}"
|
382 |
+
|
383 |
+
# Select a realistic pipeline tag based on name
|
384 |
+
if "diffusion" in model_name or "image" in model_name:
|
385 |
+
pipeline_tag = np.random.choice(["text-to-image", "image-to-image", "image-segmentation"])
|
386 |
+
elif "whisper" in model_name or "speech" in model_name:
|
387 |
+
pipeline_tag = np.random.choice(["automatic-speech-recognition", "text-to-speech"])
|
388 |
+
elif "coder" in model_name or "code" in model_name:
|
389 |
+
pipeline_tag = "text-generation"
|
390 |
+
elif "bert" in model_name or "roberta" in model_name:
|
391 |
+
pipeline_tag = np.random.choice(["fill-mask", "text-classification", "token-classification"])
|
392 |
+
elif "vision" in model_name:
|
393 |
+
pipeline_tag = np.random.choice(["image-classification", "image-to-text", "visual-question-answering"])
|
394 |
+
else:
|
395 |
+
pipeline_tag = "text-generation" # Most common
|
396 |
|
397 |
+
# Generate realistic tags
|
398 |
+
tags = [pipeline_tag]
|
399 |
|
400 |
+
if "text-generation" in pipeline_tag:
|
401 |
+
tags.extend(["language-model", "text", "gpt", "llm"])
|
402 |
+
if "instruct" in model_name:
|
403 |
+
tags.append("instruction-following")
|
404 |
+
if "chat" in model_name:
|
405 |
+
tags.append("chat")
|
406 |
+
elif "speech" in pipeline_tag:
|
407 |
+
tags.extend(["audio", "speech", "voice"])
|
408 |
+
elif "image" in pipeline_tag:
|
409 |
+
tags.extend(["vision", "image", "diffusion"])
|
410 |
|
411 |
+
# Add language tags
|
412 |
+
if np.random.random() < 0.8: # 80% chance for English
|
413 |
+
tags.append("en")
|
414 |
+
if np.random.random() < 0.3: # 30% chance for multilingual
|
415 |
+
tags.append("multilingual")
|
416 |
|
417 |
+
# Generate downloads and likes (weighted by org position for variety)
|
418 |
+
# Earlier orgs get more downloads to make the visualization interesting
|
419 |
+
popularity_factor = (len(orgs) - org_idx) / len(orgs) # 1.0 to 0.0
|
420 |
+
base_downloads = 1000 * (10 ** (2 * popularity_factor))
|
421 |
+
downloads = int(base_downloads * np.random.uniform(0.3, 3.0))
|
422 |
+
likes = int(downloads * np.random.uniform(0.01, 0.1)) # 1-10% like ratio
|
423 |
+
|
424 |
+
# Generate model size (in bytes for safetensors total)
|
425 |
+
# Model size should correlate somewhat with the size in the name
|
426 |
+
size_indicator = 1
|
427 |
+
for s in ["70b", "13b", "7b", "3b", "2b", "1b", "large", "huge", "xl", "xxl"]:
|
428 |
+
if s in model_name.lower():
|
429 |
+
size_indicator = float(s.replace("b", "")) if s[0].isdigit() else 3
|
430 |
+
break
|
431 |
+
|
432 |
+
# Size in GB, then convert to bytes
|
433 |
+
size_gb = np.random.uniform(0.1, 2.0) * size_indicator
|
434 |
+
if size_gb > 50: # Cap at 100GB
|
435 |
+
size_gb = min(size_gb, 100)
|
436 |
+
size_bytes = int(size_gb * 1e9)
|
437 |
+
|
438 |
+
# Create model entry
|
439 |
+
model = {
|
440 |
"id": model_id,
|
441 |
"downloads": downloads,
|
442 |
+
"downloadsAllTime": int(downloads * np.random.uniform(1.5, 3.0)), # All-time higher than recent
|
443 |
"likes": likes,
|
444 |
"pipeline_tag": pipeline_tag,
|
445 |
"tags": tags,
|
446 |
+
"safetensors": {"total": size_bytes}
|
447 |
+
}
|
448 |
+
|
449 |
+
data.append(model)
|
450 |
+
models_created += 1
|
451 |
+
|
452 |
+
if progress and i % 5 == 0:
|
453 |
+
progress(0.3 + 0.6 * (models_created / total_models), f"Created {models_created}/{total_models} sample models...")
|
454 |
+
|
455 |
+
# Convert to DataFrame
|
456 |
+
df = pd.DataFrame(data)
|
457 |
|
458 |
+
if progress:
|
459 |
+
progress(0.95, "Finalizing sample data...")
|
460 |
+
|
461 |
+
return df
|
462 |
|
463 |
# Create Gradio interface
|
464 |
with gr.Blocks() as demo:
|
465 |
+
models_data = gr.State() # To store loaded data
|
466 |
+
|
467 |
+
# Loading screen components
|
468 |
+
with gr.Row(visible=True) as loading_screen:
|
469 |
+
with gr.Column(scale=1):
|
470 |
+
gr.Markdown("""
|
471 |
+
# HuggingFace Models TreeMap Visualization
|
472 |
+
|
473 |
+
Loading data... This might take a moment.
|
474 |
+
""")
|
475 |
+
data_loading_progress = gr.Progress()
|
476 |
+
|
477 |
+
# Main application components (initially hidden)
|
478 |
+
with gr.Row(visible=False) as main_app:
|
479 |
gr.Markdown("""
|
480 |
+
# HuggingFace Models TreeMap Visualization
|
481 |
|
482 |
This app shows how different organizations contribute to the HuggingFace ecosystem with their models.
|
483 |
Use the filters to explore models by different metrics, tags, pipelines, and model sizes.
|
484 |
+
|
485 |
+
The treemap visualizes models grouped by organization, with the size of each box representing the selected metric (downloads or likes).
|
486 |
""")
|
487 |
|
488 |
+
with gr.Row(visible=False) as control_panel:
|
489 |
with gr.Column(scale=1):
|
490 |
count_by_dropdown = gr.Dropdown(
|
491 |
label="Metric",
|
492 |
+
choices=["downloads", "downloadsAllTime", "likes"],
|
493 |
+
value="downloads",
|
494 |
+
info="Select the metric to determine box sizes"
|
495 |
)
|
496 |
|
497 |
filter_choice_radio = gr.Radio(
|
498 |
+
label="Filter Type",
|
499 |
choices=["None", "Tag Filter", "Pipeline Filter"],
|
500 |
+
value="None",
|
501 |
+
info="Choose how to filter the models"
|
502 |
)
|
503 |
|
504 |
tag_filter_dropdown = gr.Dropdown(
|
505 |
label="Select Tag",
|
506 |
choices=list(TAG_FILTER_FUNCS.keys()),
|
507 |
value=None,
|
508 |
+
visible=False,
|
509 |
+
info="Filter models by domain/category"
|
510 |
)
|
511 |
|
512 |
pipeline_filter_dropdown = gr.Dropdown(
|
513 |
label="Select Pipeline Tag",
|
514 |
choices=PIPELINE_TAGS,
|
515 |
value=None,
|
516 |
+
visible=False,
|
517 |
+
info="Filter models by specific pipeline"
|
518 |
)
|
519 |
|
520 |
size_filter_dropdown = gr.Dropdown(
|
521 |
label="Model Size Filter",
|
522 |
choices=["None"] + list(MODEL_SIZE_RANGES.keys()),
|
523 |
+
value="None",
|
524 |
+
info="Filter models by their size (in safetensors['total'])"
|
525 |
+
)
|
526 |
+
|
527 |
+
top_k_slider = gr.Slider(
|
528 |
+
label="Number of Top Organizations",
|
529 |
+
minimum=5,
|
530 |
+
maximum=50,
|
531 |
+
value=25,
|
532 |
+
step=5,
|
533 |
+
info="Number of top organizations to include"
|
534 |
)
|
535 |
|
536 |
+
generate_plot_button = gr.Button("Generate Plot", variant="primary")
|
537 |
|
538 |
with gr.Column(scale=3):
|
539 |
plot_output = gr.Plot()
|
540 |
+
stats_output = gr.Markdown("*Generate a plot to see statistics*")
|
541 |
|
542 |
+
def generate_plot_on_click(count_by, filter_choice, tag_filter, pipeline_filter, size_filter, top_k, data_df):
|
543 |
+
print(f"Generating plot with: Metric={count_by}, Filter={filter_choice}, Tag={tag_filter}, Pipeline={pipeline_filter}, Size={size_filter}, Top K={top_k}")
|
544 |
|
545 |
+
if data_df is None or len(data_df) == 0:
|
546 |
+
return None, "Error: No data available. Please try again."
|
|
|
547 |
|
548 |
selected_tag_filter = None
|
549 |
selected_pipeline_filter = None
|
|
|
556 |
|
557 |
if size_filter != "None":
|
558 |
selected_size_filter = size_filter
|
559 |
+
|
560 |
+
# Process data for treemap
|
561 |
+
treemap_data = make_treemap_data(
|
562 |
+
df=data_df,
|
563 |
count_by=count_by,
|
564 |
+
top_k=top_k,
|
565 |
tag_filter=selected_tag_filter,
|
566 |
pipeline_filter=selected_pipeline_filter,
|
567 |
+
size_filter=selected_size_filter
|
568 |
)
|
569 |
+
|
570 |
+
# Create plot
|
571 |
+
fig = create_treemap(
|
572 |
+
treemap_data=treemap_data,
|
573 |
+
count_by=count_by,
|
574 |
+
title=f"HuggingFace Models - {count_by.capitalize()} by Organization"
|
575 |
+
)
|
576 |
+
|
577 |
+
# Generate statistics
|
578 |
+
if treemap_data.empty:
|
579 |
+
stats_md = "No data matches the selected filters."
|
580 |
+
else:
|
581 |
+
total_models = len(treemap_data)
|
582 |
+
total_value = treemap_data[count_by].sum()
|
583 |
+
top_5_orgs = treemap_data.groupby("organization")[count_by].sum().sort_values(ascending=False).head(5)
|
584 |
+
|
585 |
+
stats_md = f"""
|
586 |
+
### Statistics
|
587 |
+
- **Total models shown**: {total_models:,}
|
588 |
+
- **Total {count_by}**: {total_value:,}
|
589 |
+
|
590 |
+
### Top 5 Organizations
|
591 |
+
| Organization | {count_by.capitalize()} | % of Total |
|
592 |
+
| --- | --- | --- |
|
593 |
+
"""
|
594 |
+
|
595 |
+
for org, value in top_5_orgs.items():
|
596 |
+
percentage = (value / total_value) * 100
|
597 |
+
stats_md += f"| {org} | {value:,} | {percentage:.2f}% |\n"
|
598 |
+
|
599 |
+
return fig, stats_md
|
600 |
|
601 |
def update_filter_visibility(filter_choice):
|
602 |
if filter_choice == "Tag Filter":
|
|
|
612 |
outputs=[tag_filter_dropdown, pipeline_filter_dropdown]
|
613 |
)
|
614 |
|
615 |
+
def load_data_with_progress(progress=gr.Progress()):
|
616 |
+
"""Load data with progress tracking and update UI visibility"""
|
617 |
+
data_df = download_and_process_models(progress)
|
618 |
+
# Return both the data and the visibility updates
|
619 |
+
return data_df, gr.update(visible=False), gr.update(visible=True), gr.update(visible=True)
|
620 |
+
|
621 |
+
# Load data once at startup with progress bar
|
622 |
demo.load(
|
623 |
+
fn=load_data_with_progress,
|
624 |
inputs=[],
|
625 |
+
outputs=[models_data, loading_screen, main_app, control_panel]
|
626 |
)
|
627 |
|
628 |
# Button click event to generate plot
|
|
|
634 |
tag_filter_dropdown,
|
635 |
pipeline_filter_dropdown,
|
636 |
size_filter_dropdown,
|
637 |
+
top_k_slider,
|
638 |
models_data
|
639 |
],
|
640 |
+
outputs=[plot_output, stats_output]
|
641 |
)
|
642 |
|
643 |
|