Spaces:
Running
on
A100
Running
on
A100
Restart space every 6 hours as workaround, refactor garbage remover, add apscheduler as new deps
Browse files
app.py
CHANGED
@@ -15,6 +15,8 @@ from gradio_logsview.logsview import Log, LogsView, LogsViewRunner
|
|
15 |
from mergekit.config import MergeConfiguration
|
16 |
|
17 |
from clean_community_org import garbage_collect_empty_models
|
|
|
|
|
18 |
|
19 |
has_gpu = torch.cuda.is_available()
|
20 |
|
@@ -186,10 +188,29 @@ def merge(yaml_config: str, hf_token: str, repo_name: str) -> Iterable[List[Log]
|
|
186 |
)
|
187 |
yield runner.log(f"Model successfully uploaded to HF: {repo_url.repo_id}")
|
188 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
189 |
|
190 |
with gr.Blocks() as demo:
|
191 |
gr.Markdown(MARKDOWN_DESCRIPTION)
|
192 |
-
|
|
|
193 |
with gr.Row():
|
194 |
filename = gr.Textbox(visible=False, label="filename")
|
195 |
config = gr.Code(language="yaml", lines=10, label="config.yaml")
|
@@ -221,18 +242,5 @@ with gr.Blocks() as demo:
|
|
221 |
button.click(fn=merge, inputs=[config, token, repo_name], outputs=[logs])
|
222 |
|
223 |
|
224 |
-
# Run garbage collection every hour to keep the community org clean.
|
225 |
-
# Empty models might exists if the merge fails abruptly (e.g. if user leaves the Space).
|
226 |
-
def _garbage_collect_every_hour():
|
227 |
-
while True:
|
228 |
-
try:
|
229 |
-
garbage_collect_empty_models(token=COMMUNITY_HF_TOKEN)
|
230 |
-
except Exception as e:
|
231 |
-
print("Error running garbage collection", e)
|
232 |
-
time.sleep(3600)
|
233 |
-
|
234 |
-
|
235 |
-
pool = ThreadPoolExecutor()
|
236 |
-
pool.submit(_garbage_collect_every_hour)
|
237 |
|
238 |
demo.queue(default_concurrency_limit=1).launch()
|
|
|
15 |
from mergekit.config import MergeConfiguration
|
16 |
|
17 |
from clean_community_org import garbage_collect_empty_models
|
18 |
+
from apscheduler.schedulers.background import BackgroundScheduler
|
19 |
+
from datetime import datetime, timezone
|
20 |
|
21 |
has_gpu = torch.cuda.is_available()
|
22 |
|
|
|
188 |
)
|
189 |
yield runner.log(f"Model successfully uploaded to HF: {repo_url.repo_id}")
|
190 |
|
191 |
+
# This is workaround. As the space always getting stuck.
|
192 |
+
def _restart_space():
|
193 |
+
huggingface_hub.HfApi().restart_space(repo_id="arcee-ai/mergekit-gui", token=COMMUNITY_HF_TOKEN, factory_reboot=False)
|
194 |
+
# Run garbage collection every hour to keep the community org clean.
|
195 |
+
# Empty models might exists if the merge fails abruptly (e.g. if user leaves the Space).
|
196 |
+
def _garbage_remover():
|
197 |
+
try:
|
198 |
+
garbage_collect_empty_models(token=COMMUNITY_HF_TOKEN)
|
199 |
+
except Exception as e:
|
200 |
+
print("Error running garbage collection", e)
|
201 |
+
|
202 |
+
scheduler = BackgroundScheduler()
|
203 |
+
restart_space_job = scheduler.add_job(_restart_space, "interval", seconds=21600)
|
204 |
+
garbage_remover_job = scheduler.add_job(_garbage_remover, "interval", seconds=3600)
|
205 |
+
scheduler.start()
|
206 |
+
next_run_time_utc = restart_space_job.next_run_time.astimezone(timezone.utc)
|
207 |
+
|
208 |
+
NEXT_RESTART = f"Next Restart: {next_run_time_utc.strftime('%Y-%m-%d %H:%M:%S')} (UTC)"
|
209 |
|
210 |
with gr.Blocks() as demo:
|
211 |
gr.Markdown(MARKDOWN_DESCRIPTION)
|
212 |
+
gr.Markdown(NEXT_RESTART)
|
213 |
+
|
214 |
with gr.Row():
|
215 |
filename = gr.Textbox(visible=False, label="filename")
|
216 |
config = gr.Code(language="yaml", lines=10, label="config.yaml")
|
|
|
242 |
button.click(fn=merge, inputs=[config, token, repo_name], outputs=[logs])
|
243 |
|
244 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
245 |
|
246 |
demo.queue(default_concurrency_limit=1).launch()
|