wangbo/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2026-04-15 13:02:35 +08:00
Code Issues Actions 17 Packages Projects Releases Wiki Activity

Merge 3d51d63490 into a2840e7552

Browse Source
This commit is contained in:
Octopus 2026-04-11 13:15:43 +08:00 committed by GitHub
commit 52f08c3565
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 15 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
comfy/audio_encoders/audio_encoders.py
View File

@ -76,6 +76,9 @@ def load_audio_encoder_from_sd(sd, prefix=""):
raise RuntimeError("ERROR: audio encoder file is invalid or unsupported embed_dim: {}".format(embed_dim)) raise RuntimeError("ERROR: audio encoder file is invalid or unsupported embed_dim: {}".format(embed_dim))
elif "model.encoder.embed_positions.weight" in sd: elif "model.encoder.embed_positions.weight" in sd:
sd = comfy.utils.state_dict_prefix_replace(sd, {"model.": ""}) sd = comfy.utils.state_dict_prefix_replace(sd, {"model.": ""})
# Full whisper checkpoints include decoder weights; discard them since
# WhisperLargeV3 is encoder-only.
sd = {k: v for k, v in sd.items() if not k.startswith("decoder.")}
config = { config = {
"model_type": "whisper3", "model_type": "whisper3",
} }
@ -85,7 +88,18 @@ def load_audio_encoder_from_sd(sd, prefix=""):
audio_encoder = AudioEncoderModel(config) audio_encoder = AudioEncoderModel(config)
m, u = audio_encoder.load_sd(sd) m, u = audio_encoder.load_sd(sd)
if len(m) > 0: if len(m) > 0:
logging.warning("missing audio encoder: {}".format(m)) # torchaudio registers the mel-spectrogram window and filterbank as
# buffers whose values are deterministically computed from the config
# at init time; they are never saved in standard whisper checkpoints.
# Suppress warnings for these expected-missing buffers so that users
# are only alerted about genuinely unexpected missing weights.
whisper_computed_buffers = {
"feature_extractor.mel_spectrogram.spectrogram.window",
"feature_extractor.mel_spectrogram.mel_scale.fb",
}
significant_missing = [k for k in m if k not in whisper_computed_buffers]
if significant_missing:
logging.warning("missing audio encoder: {}".format(significant_missing))
if len(u) > 0: if len(u) > 0:
logging.warning("unexpected audio encoder: {}".format(u)) logging.warning("unexpected audio encoder: {}".format(u))