Labyricorn/voicebox
1
0
Fork 0
mirror of https://github.com/jamiepine/voicebox.git synced 2026-09-18 14:20:42 -07:00
Code Issues Packages Projects Releases Wiki Activity

fix(tada): run voice-prompt encode under torch.inference_mode (#955)

Browse Source 
Encoder.eval() alone still builds an autograd graph because parameters
require grad by default. On 8GB GPUs that ballooned TADA encode VRAM far
past the model footprint (issue 890). Wrap the encode forward in
inference_mode and add a unit test that asserts the flag is set.

Co-authored-by: fooSynaptic <[email protected]>
This commit is contained in:
fooSynaptic
2026-07-26 23:32:25 -07:00
committed by Jamie Pine
co-authored by fooSynaptic
parent f7b45cc89e
commit 0d31687638
2 changed files with 74 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/backends/hume_backend.py
+6 -2
View File
@@ -247,9 +247,13 @@ class HumeTadaBackend:
audio = audio.T # (samples, channels) -> (channels, samples)
audio = audio.to(device)
# Encode with forced alignment
# Encode with forced alignment.
# Must run under inference_mode: encoder params still require
# grad by default, and an autograd graph across the DAC/Snake
# stack can balloon VRAM far past the model footprint (#890).
text_arg = [reference_text] if reference_text else None
prompt = self.encoder(audio, text=text_arg, sample_rate=sr)
with torch.inference_mode():
prompt = self.encoder(audio, text=text_arg, sample_rate=sr)
# Serialize EncoderOutput to a dict of CPU tensors for caching
prompt_dict = {}