Commit History
25 days ago
-
Time each stage of a take: ZeroGPU scheduling, worker start, encode, capture, prefill, windows
Verified signature0cbfe0b -
Rename the Script heading to Examples and tighten the examples section
Verified signatured8542ef -
Remove the app's daily take allowance; ZeroGPU's own quota applies. Inline direction tags on /legacy too
Verified signature35cc1f9 -
Show direction tags inline and always visible, in a compact wrapping layout
Verified signatureecad1cd -
Default to xlarge ZeroGPU; request GPU time closer to use; README measurements
Verified signature07a0f50 -
Log the GPU and the decode step time at warm-up
Verified signatured158e24 -
Load the preloaded checkpoint offline-first; log startup steps
Verified signature1ef6ea1 -
Freeze parameters; warm up under inference_mode
Verified signatureae1c571 -
Compile in the GPU worker, after ZeroGPU restores torch.cuda
Verified signature9b3c95d -
Restore torch.cuda._maybe_exchange_device in the GPU worker; warm up at startup
Verified signature252e871 -
Run Drama 3 in-process on ZeroGPU instead of the H200 endpoint
Verified signaturec430fcf
26 days ago
-
Update technical report link to arXiv (#13)
- Update technical report link to arXiv (34b0658303afb737716f9e84866a76e939f0ffe0)3294a04 -
Fix ZeroGPU initialization (#11)
- Fix ZeroGPU initialization (070eaa6bcc74f329881918c951e80e4437784c73) - fixing (a3e529d3b490d03dcab837b46334e4ff70f696f8) Co-authored-by: Andres Marafioti <andito@users.noreply.huggingface.co>c1f6d2b -
Fix native Gradio dependency install (#9)
- Fix native Gradio dependency install (0ffc5f0adffbe83c6968f505eff6df13e8091169) Co-authored-by: Andres Marafioti <andito@users.noreply.huggingface.co>fe2290f -
Fix ZeroGPU startup with the native Gradio SDK (#7)
- Fix ZeroGPU startup with the native Gradio SDK (c41fab9382e9564d9afc3c7cba6607e8d3c85487) Co-authored-by: Andres Marafioti <andito@users.noreply.huggingface.co>9e9375b -
Replace ZeroGPU demo with Gradio studio and H200 endpoint (#6)
- Replace ZeroGPU demo with Gradio studio and H200 endpoint (3964d7cccceba79eb78cd48f4d76a695d697b01a) Co-authored-by: Andres Marafioti <andito@users.noreply.huggingface.co>e9cf2ae -
Center the UI (#4)
- Center the UI (091b9a8fe92ef849a01f55b66e54740b540cdd29) Co-authored-by: Apolinário from multimodal AI art <multimodalart@users.noreply.huggingface.co>98baacb -
add voice clone consent gate (#5)
- Add voice consent gate (LLM consent sentence + Whisper verification) before cloning (e4a9301df8dbced41e44a157dfe234b7f67bf9ce) - Add consent_confirmed port to clone node (dc2c2d29c7927f930aa334f86a9535704ea1e9a1) Co-authored-by: AK <akhaliq@users.noreply.huggingface.co>0e36c4e -
[hf staff] fix dark mode (#3)
- [hf staff] fix dark mode (a992baeb90d6f4a365e40c6127dd5a3ef7b350e0) Co-authored-by: Linoy Tsaban <linoyts@users.noreply.huggingface.co>5505e80 -
add workflow back (#2)
- Add the workflow canvas back behind the App/Workflow toggle; UI unchanged (41eb56a3fc0332170e5b3ad349aaabdd6eabfc8d) - Single-speaker clone graph for the workflow canvas (d7a81d5f7ff77a3dce9dd5c17fa4a412ccb0d999) Co-authored-by: AK <akhaliq@users.noreply.huggingface.co>0296177
27 days ago
-
Restyle the voice-clone page in the fish.audio look.
zhaochenyang20 +1: CursorKeep the same fields, instructions, links, and acknowledgments on a cream page with white cards and a black pill button. Co-authored-by: Cursor <cursoragent@cursor.com>d183993 -
Shorten the thanks to AK for setting up the Space.
zhaochenyang20 +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>c89aeca -
Point SGLang Omni at its GitHub repo and ask for a star.
zhaochenyang20 +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>9307daa -
Thank AK for reaching out and helping set up the Space.
zhaochenyang20 +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>c417ff7 -
Explain the voice-clone fields and credit the people behind the Space.
zhaochenyang20 +1: CursorThe page now says what each input does, links out to SGLang Omni, and thanks AK and both teams. Co-authored-by: Cursor <cursoragent@cursor.com>12f96a8 -
Simplify Drama 3 to a single-speaker voice clone page.
zhaochenyang20 +1: CursorA click now synthesizes the requested line from one reference clip and its transcript, instead of running the Workflow canvas and a revision pass. Co-authored-by: Cursor <cursoragent@cursor.com>a0978a5 -
workflow app (#1)
- Add requirements (gradio 6.30 workflow + fish_speech deps) (b0f829d720656e08825b050e3a64ae9fa261d198) - Gradio Workflow app (gr.Workflow + drama3_perform node) (87c2232b7f4a8d33379d056a025cdc4da5c66e2f) - Add workflow graph (Perform + Revise pipelines) (4eaefccdfb50e9b41112b85c2a5a084d1fcf76f5) - README: merged parent yaml + workflow docs (55926f4c2419f6db48a885a8b67393be797330ad) - Vendor fish_speech package (official PyTorch inference path) (b4bb39aa57a1ed496f24a44c1ef5eabfae112bdf) - Vendor fish_speech package (ffe2592df805a05fb59ee774d93d7c5588ef1a05) - Vendor fish_speech package (e3c01e9aa92c8423cae85e771ace7c5b73c90454) - Vendor fish_speech package (f9f146bf61eb72bae45b9d3b3467d9d96113665c) - Vendor fish_speech package (5175e6cc1a888d837273eead46c5a24884f9e80a) - Vendor fish_speech package (34ed18ef1a9e7a8c16a38d9b5cc6d4cfa1f5f240) - Vendor fish_speech package (61a5812f545dd660953d22cc9a1c6d94c7613200) - Vendor fish_speech package (c4689eb95deb36a05047c25f1b76c9e0ddfdfcc1) - Ve…80c54ec -
initial commit
Verified signature4568dd7