mirror of
https://github.com/comfyanonymous/ComfyUI.git
synced 2026-04-25 01:42:36 +08:00
3d51d63490
When a full whisper checkpoint (encoder + decoder) is loaded via AudioEncoderLoader, two classes of spurious warnings were emitted: 1. 'unexpected audio encoder' for every decoder.* key - the decoder is not part of WhisperLargeV3, so these keys are always present in full whisper checkpoints and should be silently discarded. 2. 'missing audio encoder' for feature_extractor.mel_spectrogram buffers (window and mel_scale.fb) - these are torchaudio buffers computed deterministically from config at init time; they are never stored in standard whisper checkpoints but are always correctly initialised. Fix: strip decoder keys from the state-dict before loading, and suppress warnings for the two known torchaudio-computed buffer keys. Fixes #13276
107 lines
4.3 KiB
Python
107 lines
4.3 KiB
Python
from .wav2vec2 import Wav2Vec2Model
|
|
from .whisper import WhisperLargeV3
|
|
import comfy.model_management
|
|
import comfy.ops
|
|
import comfy.utils
|
|
import logging
|
|
import torchaudio
|
|
|
|
|
|
class AudioEncoderModel():
|
|
def __init__(self, config):
|
|
self.load_device = comfy.model_management.text_encoder_device()
|
|
offload_device = comfy.model_management.text_encoder_offload_device()
|
|
self.dtype = comfy.model_management.text_encoder_dtype(self.load_device)
|
|
model_type = config.pop("model_type")
|
|
model_config = dict(config)
|
|
model_config.update({
|
|
"dtype": self.dtype,
|
|
"device": offload_device,
|
|
"operations": comfy.ops.manual_cast
|
|
})
|
|
|
|
if model_type == "wav2vec2":
|
|
self.model = Wav2Vec2Model(**model_config)
|
|
elif model_type == "whisper3":
|
|
self.model = WhisperLargeV3(**model_config)
|
|
self.model.eval()
|
|
self.patcher = comfy.model_patcher.CoreModelPatcher(self.model, load_device=self.load_device, offload_device=offload_device)
|
|
self.model_sample_rate = 16000
|
|
comfy.model_management.archive_model_dtypes(self.model)
|
|
|
|
def load_sd(self, sd):
|
|
return self.model.load_state_dict(sd, strict=False, assign=self.patcher.is_dynamic())
|
|
|
|
def get_sd(self):
|
|
return self.model.state_dict()
|
|
|
|
def encode_audio(self, audio, sample_rate):
|
|
comfy.model_management.load_model_gpu(self.patcher)
|
|
audio = torchaudio.functional.resample(audio, sample_rate, self.model_sample_rate)
|
|
out, all_layers = self.model(audio.to(self.load_device))
|
|
outputs = {}
|
|
outputs["encoded_audio"] = out
|
|
outputs["encoded_audio_all_layers"] = all_layers
|
|
outputs["audio_samples"] = audio.shape[2]
|
|
return outputs
|
|
|
|
|
|
def load_audio_encoder_from_sd(sd, prefix=""):
|
|
sd = comfy.utils.state_dict_prefix_replace(sd, {"wav2vec2.": ""})
|
|
if "encoder.layer_norm.bias" in sd: #wav2vec2
|
|
embed_dim = sd["encoder.layer_norm.bias"].shape[0]
|
|
if embed_dim == 1024:# large
|
|
config = {
|
|
"model_type": "wav2vec2",
|
|
"embed_dim": 1024,
|
|
"num_heads": 16,
|
|
"num_layers": 24,
|
|
"conv_norm": True,
|
|
"conv_bias": True,
|
|
"do_normalize": True,
|
|
"do_stable_layer_norm": True
|
|
}
|
|
elif embed_dim == 768: # base
|
|
config = {
|
|
"model_type": "wav2vec2",
|
|
"embed_dim": 768,
|
|
"num_heads": 12,
|
|
"num_layers": 12,
|
|
"conv_norm": False,
|
|
"conv_bias": False,
|
|
"do_normalize": False, # chinese-wav2vec2-base has this False
|
|
"do_stable_layer_norm": False
|
|
}
|
|
else:
|
|
raise RuntimeError("ERROR: audio encoder file is invalid or unsupported embed_dim: {}".format(embed_dim))
|
|
elif "model.encoder.embed_positions.weight" in sd:
|
|
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 = {
|
|
"model_type": "whisper3",
|
|
}
|
|
else:
|
|
raise RuntimeError("ERROR: audio encoder not supported.")
|
|
|
|
audio_encoder = AudioEncoderModel(config)
|
|
m, u = audio_encoder.load_sd(sd)
|
|
if len(m) > 0:
|
|
# 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:
|
|
logging.warning("unexpected audio encoder: {}".format(u))
|
|
|
|
return audio_encoder
|