Spaces:
Running on CPU Upgrade
Running on CPU Upgrade
Commit History
6 days ago
-
Embedding visualizer
7c0c96e
-
agents.md
a727456 -
stratos fix
b3c6651 -
Grok Build attempt to fix sudden 500 on iframe
The Hub iframe 500 was not an xVASynth synthesis crash. The TTS log had no traceback around that time — last successful synth was 2026-09-06 13:10 UTC. Hugging Face status was also clean. Standalone https://pendrokar-xvasynth-tts.hf.space kept working because Gradio on port 7860 was still up. What actually broke the iframe The Space has been in Dev Mode since 30 Aug, with init-process supervising python app.py. That original process (PID 19) had been serving 7860 for 7.5 days. Dev Mode Refresh does not replace that listener. At 09:51 UTC it started a second python app.py (sh -c python app.py) without taking over 7860. The new process loaded models, failed to bind, and then sat in a hold loop. The old process kept the port. That split is exactly how you get iframe 500 / standalone 200: 1. Hugging Face’s Hub page does not use the same path as a direct .hf.space visit. The iframe goes through the Spaces proxy / health layer. Standalone hits Gradio on 7860. 2. On Linux, Gradio/uvicorn…ad84328
-
init sacrifice models
58aff5f -
init voice radio fix
11f9735 -
download stratos snapshot
513cb99 -
stratos xVAPitch model Running
cad7a0f
-
Added Skyrim creature voices
29ef189 -
app.py: Making Gradio v4.20 compatible HF Space
c59a4e3
-
DeepMoji tests & changes to tests
efa00c2 -
Add Gradio API tests for a V3 xVAPitch voice and the Lojban model.
94f18dd -
Do not SIGTERM other Gradio processes; that kills the Space supervisor child (exit 143) and the Hub iframe.
b65f88d -
Free port 7860 before Gradio launch so Space restarts can bind.
A leftover Gradio listener made HF's official python app.py fail with OSError on 7860-7860, which marked the application stopped and killed the Space iframe.80b8cab -
re-enable deepmoji
7693ae3 -
Fix Space 502 by binding Gradio to 0.0.0.0 and resolving HF model cache paths.
Gradio was listening on 127.0.0.1 only (share=True / default), so the Space proxy returned 502. Voice models were cached under /root/.cache, not the hardcoded /home/user path.26b9756
-
disable deepmoji
Verified signature5badca2 -
gradio==4.20.0
Verified signaturece7b3f4
-
rm gradio==4.20.0
Verified signaturec6cecb7 -
Update requirements.txt
Verified signature2f8e6e8 -
Update requirements.txt
Verified signatureda160e1 -
torchmoji model pointer to file
Verified signature5c24e0a -
weights_only=False
Verified signature79532a4 -
torchvision>=0.19.0+cpu
Verified signature0f90656 -
torch < 2.6
Verified signaturebbbe7f6 -
torch 2.5.1
Verified signaturec1073f4
-
Torchmoji files
Verified signature2cc999f -
mkdir
Verified signature62d614b -
vocab wget
Verified signature227354a -
vocab #2
Verified signature50b4672 -
vocab
Verified signature6fef268 -
vocab preload
Verified signaturee107bc2
-
torch 2.4 cpu others
Verified signature0aa5f59 -
removed no deps
Verified signature0e667d7
-
torch>=2.4.0+cpu
Verified signature9e2ed0e -
typing-extensions >=4.5
Verified signatureaedb823
-
torch any version
Verified signaturede5353f -
torch
Verified signature9bbde0c -
Update Dockerfile
Verified signature2ffbf93 -
Update Dockerfile
Verified signaturee8aa1a7 -
Use Dockerfile
Verified signaturecb548b8 -
numpy #2
Verified signature5a4d82b -
numpy pre-required
Verified signatureb4fbafa -
numpy
Verified signature7a9f859
-
enable DeepMoji on xvapitch
Verified signatureab4f315 -
panphon==0.17.0
d743855 -
panphon==0.18.0 to fix TypeError: unsupported operand type(s) for |: 'type' and 'NoneType'
4d3906c -
Merge remote-tracking branch 'origin/main'
ae765ca -
output info to HF logs
df6a167
-
lowercase model type
Verified signature725f6d6