Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

fix Parameter dtype in audio models #30310

Merged
merged 1 commit into from
Apr 18, 2024
Merged

fix Parameter dtype in audio models #30310

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions src/transformers/models/data2vec/modeling_data2vec_audio.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -822,7 +822,7 @@ def __init__(self, config: Data2VecAudioConfig):

# model only needs masking vector if mask prob is > 0.0
if config.mask_time_prob > 0.0 or config.mask_feature_prob > 0.0:
self.masked_spec_embed = nn.Parameter(torch.FloatTensor(config.hidden_size).uniform_())
self.masked_spec_embed = nn.Parameter(torch.Tensor(config.hidden_size).uniform_())

self.encoder = Data2VecAudioEncoder(config)

Expand Down Expand Up @@ -858,7 +858,7 @@ def _mask_hidden_states(

if mask_time_indices is not None:
# apply SpecAugment along time axis with given mask_time_indices
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed
elif self.config.mask_time_prob > 0 and self.training:
mask_time_indices = _compute_mask_indices(
(batch_size, sequence_length),
Expand All @@ -868,7 +868,7 @@ def _mask_hidden_states(
min_masks=self.config.mask_time_min_masks,
)
mask_time_indices = torch.tensor(mask_time_indices, device=hidden_states.device, dtype=torch.bool)
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed

if self.config.mask_feature_prob > 0 and self.training:
# generate indices & apply SpecAugment along feature axis
Expand Down
6 changes: 3 additions & 3 deletions src/transformers/models/hubert/modeling_hubert.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -974,7 +974,7 @@ def __init__(self, config: HubertConfig):
self.feature_projection = HubertFeatureProjection(config)

if config.mask_time_prob > 0.0 or config.mask_feature_prob > 0.0:
self.masked_spec_embed = nn.Parameter(torch.FloatTensor(config.hidden_size).uniform_())
self.masked_spec_embed = nn.Parameter(torch.Tensor(config.hidden_size).uniform_())

if config.do_stable_layer_norm:
self.encoder = HubertEncoderStableLayerNorm(config)
Expand Down Expand Up @@ -1005,7 +1005,7 @@ def _mask_hidden_states(

if mask_time_indices is not None:
# apply SpecAugment along time axis with given mask_time_indices
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed
elif self.config.mask_time_prob > 0 and self.training:
mask_time_indices = _compute_mask_indices(
(batch_size, sequence_length),
Expand All @@ -1015,7 +1015,7 @@ def _mask_hidden_states(
min_masks=self.config.mask_time_min_masks,
)
mask_time_indices = torch.tensor(mask_time_indices, device=hidden_states.device, dtype=torch.bool)
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed

if self.config.mask_feature_prob > 0 and self.training:
# generate indices & apply SpecAugment along feature axis
Expand Down
6 changes: 3 additions & 3 deletions src/transformers/models/sew/modeling_sew.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -834,7 +834,7 @@ def __init__(self, config: SEWConfig):
self.feature_dropout = nn.Dropout(config.feat_proj_dropout)

if config.mask_time_prob > 0.0 or config.mask_feature_prob > 0.0:
self.masked_spec_embed = nn.Parameter(torch.FloatTensor(config.hidden_size).uniform_())
self.masked_spec_embed = nn.Parameter(torch.Tensor(config.hidden_size).uniform_())

self.encoder = SEWEncoder(config)

Expand Down Expand Up @@ -862,7 +862,7 @@ def _mask_hidden_states(

if mask_time_indices is not None:
# apply SpecAugment along time axis with given mask_time_indices
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed
elif self.config.mask_time_prob > 0 and self.training:
mask_time_indices = _compute_mask_indices(
(batch_size, sequence_length),
Expand All @@ -872,7 +872,7 @@ def _mask_hidden_states(
min_masks=self.config.mask_time_min_masks,
)
mask_time_indices = torch.tensor(mask_time_indices, device=hidden_states.device, dtype=torch.bool)
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed

if self.config.mask_feature_prob > 0 and self.training:
# generate indices & apply SpecAugment along feature axis
Expand Down
6 changes: 3 additions & 3 deletions src/transformers/models/sew_d/modeling_sew_d.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1360,7 +1360,7 @@ def __init__(self, config: SEWDConfig):
self.feature_dropout = nn.Dropout(config.feat_proj_dropout)

if config.mask_time_prob > 0.0 or config.mask_feature_prob > 0.0:
self.masked_spec_embed = nn.Parameter(torch.FloatTensor(config.hidden_size).uniform_())
self.masked_spec_embed = nn.Parameter(torch.Tensor(config.hidden_size).uniform_())

self.encoder = SEWDEncoder(config)

Expand Down Expand Up @@ -1388,7 +1388,7 @@ def _mask_hidden_states(

if mask_time_indices is not None:
# apply SpecAugment along time axis with given mask_time_indices
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed
elif self.config.mask_time_prob > 0 and self.training:
mask_time_indices = _compute_mask_indices(
(batch_size, sequence_length),
Expand All @@ -1398,7 +1398,7 @@ def _mask_hidden_states(
min_masks=self.config.mask_time_min_masks,
)
mask_time_indices = torch.tensor(mask_time_indices, device=hidden_states.device, dtype=torch.bool)
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed

if self.config.mask_feature_prob > 0 and self.training:
# generate indices & apply SpecAugment along feature axis
Expand Down
6 changes: 3 additions & 3 deletions src/transformers/models/speecht5/modeling_speecht5.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -517,7 +517,7 @@ def __init__(self, config):

# model only needs masking vector if mask prob is > 0.0
if config.mask_time_prob > 0.0 or config.mask_feature_prob > 0.0:
self.masked_spec_embed = nn.Parameter(torch.FloatTensor(config.hidden_size).uniform_())
self.masked_spec_embed = nn.Parameter(torch.Tensor(config.hidden_size).uniform_())

self.pos_conv_embed = SpeechT5PositionalConvEmbedding(config)
self.pos_sinusoidal_embed = SpeechT5SinusoidalPositionalEmbedding(
Expand Down Expand Up @@ -616,7 +616,7 @@ def _mask_hidden_states(

if mask_time_indices is not None:
# apply SpecAugment along time axis with given mask_time_indices
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed
elif self.config.mask_time_prob > 0 and self.training:
mask_time_indices = _compute_mask_indices(
(batch_size, sequence_length),
Expand All @@ -626,7 +626,7 @@ def _mask_hidden_states(
min_masks=self.config.mask_time_min_masks,
)
mask_time_indices = torch.tensor(mask_time_indices, device=hidden_states.device, dtype=torch.bool)
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed

if self.config.mask_feature_prob > 0 and self.training:
# generate indices & apply SpecAugment along feature axis
Expand Down
6 changes: 3 additions & 3 deletions src/transformers/models/unispeech/modeling_unispeech.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1090,7 +1090,7 @@ def __init__(self, config: UniSpeechConfig):
self.feature_projection = UniSpeechFeatureProjection(config)

if config.mask_time_prob > 0.0 or config.mask_feature_prob > 0.0:
self.masked_spec_embed = nn.Parameter(torch.FloatTensor(config.hidden_size).uniform_())
self.masked_spec_embed = nn.Parameter(torch.Tensor(config.hidden_size).uniform_())

if config.do_stable_layer_norm:
self.encoder = UniSpeechEncoderStableLayerNorm(config)
Expand Down Expand Up @@ -1121,7 +1121,7 @@ def _mask_hidden_states(

if mask_time_indices is not None:
# apply SpecAugment along time axis with given mask_time_indices
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed
elif self.config.mask_time_prob > 0 and self.training:
mask_time_indices = _compute_mask_indices(
(batch_size, sequence_length),
Expand All @@ -1131,7 +1131,7 @@ def _mask_hidden_states(
min_masks=self.config.mask_time_min_masks,
)
mask_time_indices = torch.tensor(mask_time_indices, device=hidden_states.device, dtype=torch.bool)
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed

if self.config.mask_feature_prob > 0 and self.training:
# generate indices & apply SpecAugment along feature axis
Expand Down
6 changes: 3 additions & 3 deletions src/transformers/models/unispeech_sat/modeling_unispeech_sat.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1108,7 +1108,7 @@ def __init__(self, config: UniSpeechSatConfig):
self.feature_extractor = UniSpeechSatFeatureEncoder(config)
self.feature_projection = UniSpeechSatFeatureProjection(config)

self.masked_spec_embed = nn.Parameter(torch.FloatTensor(config.hidden_size).uniform_())
self.masked_spec_embed = nn.Parameter(torch.Tensor(config.hidden_size).uniform_())

if config.do_stable_layer_norm:
self.encoder = UniSpeechSatEncoderStableLayerNorm(config)
Expand Down Expand Up @@ -1139,7 +1139,7 @@ def _mask_hidden_states(

if mask_time_indices is not None:
# apply SpecAugment along time axis with given mask_time_indices
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed
elif self.config.mask_time_prob > 0 and self.training:
mask_time_indices = _compute_mask_indices(
(batch_size, sequence_length),
Expand All @@ -1149,7 +1149,7 @@ def _mask_hidden_states(
min_masks=self.config.mask_time_min_masks,
)
mask_time_indices = torch.tensor(mask_time_indices, device=hidden_states.device, dtype=torch.bool)
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed

if self.config.mask_feature_prob > 0 and self.training:
# generate indices & apply SpecAugment along feature axis
Expand Down
6 changes: 3 additions & 3 deletions src/transformers/models/wav2vec2/modeling_wav2vec2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1445,7 +1445,7 @@ def __init__(self, config: Wav2Vec2Config):

# model only needs masking vector if mask prob is > 0.0
if config.mask_time_prob > 0.0 or config.mask_feature_prob > 0.0:
self.masked_spec_embed = nn.Parameter(torch.FloatTensor(config.hidden_size).uniform_())
self.masked_spec_embed = nn.Parameter(torch.Tensor(config.hidden_size).uniform_())

if config.do_stable_layer_norm:
self.encoder = Wav2Vec2EncoderStableLayerNorm(config)
Expand Down Expand Up @@ -1496,7 +1496,7 @@ def _mask_hidden_states(

if mask_time_indices is not None:
# apply SpecAugment along time axis with given mask_time_indices
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed
elif self.config.mask_time_prob > 0 and self.training:
mask_time_indices = _compute_mask_indices(
(batch_size, sequence_length),
Expand All @@ -1506,7 +1506,7 @@ def _mask_hidden_states(
min_masks=self.config.mask_time_min_masks,
)
mask_time_indices = torch.tensor(mask_time_indices, device=hidden_states.device, dtype=torch.bool)
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed

if self.config.mask_feature_prob > 0 and self.training:
# generate indices & apply SpecAugment along feature axis
Expand Down
6 changes: 3 additions & 3 deletions src/transformers/models/wav2vec2_bert/modeling_wav2vec2_bert.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1053,7 +1053,7 @@ def __init__(self, config: Wav2Vec2BertConfig):

# model only needs masking vector if mask prob is > 0.0
if config.mask_time_prob > 0.0 or config.mask_feature_prob > 0.0:
self.masked_spec_embed = nn.Parameter(torch.FloatTensor(config.hidden_size).uniform_())
self.masked_spec_embed = nn.Parameter(torch.Tensor(config.hidden_size).uniform_())

self.encoder = Wav2Vec2BertEncoder(config)

Expand Down Expand Up @@ -1087,7 +1087,7 @@ def _mask_hidden_states(

if mask_time_indices is not None:
# apply SpecAugment along time axis with given mask_time_indices
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed
elif self.config.mask_time_prob > 0 and self.training:
mask_time_indices = _compute_mask_indices(
(batch_size, sequence_length),
Expand All @@ -1097,7 +1097,7 @@ def _mask_hidden_states(
min_masks=self.config.mask_time_min_masks,
)
mask_time_indices = torch.tensor(mask_time_indices, device=hidden_states.device, dtype=torch.bool)
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed

if self.config.mask_feature_prob > 0 and self.training:
# generate indices & apply SpecAugment along feature axis
Expand Down
6 changes: 3 additions & 3 deletions src/transformers/models/wav2vec2_conformer/modeling_wav2vec2_conformer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1235,7 +1235,7 @@ def __init__(self, config: Wav2Vec2ConformerConfig):

# model only needs masking vector if mask prob is > 0.0
if config.mask_time_prob > 0.0 or config.mask_feature_prob > 0.0:
self.masked_spec_embed = nn.Parameter(torch.FloatTensor(config.hidden_size).uniform_())
self.masked_spec_embed = nn.Parameter(torch.Tensor(config.hidden_size).uniform_())

self.encoder = Wav2Vec2ConformerEncoder(config)

Expand Down Expand Up @@ -1273,7 +1273,7 @@ def _mask_hidden_states(

if mask_time_indices is not None:
# apply SpecAugment along time axis with given mask_time_indices
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed
elif self.config.mask_time_prob > 0 and self.training:
mask_time_indices = _compute_mask_indices(
(batch_size, sequence_length),
Expand All @@ -1283,7 +1283,7 @@ def _mask_hidden_states(
min_masks=self.config.mask_time_min_masks,
)
mask_time_indices = torch.tensor(mask_time_indices, device=hidden_states.device, dtype=torch.bool)
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed

if self.config.mask_feature_prob > 0 and self.training:
# generate indices & apply SpecAugment along feature axis
Expand Down
6 changes: 3 additions & 3 deletions src/transformers/models/wavlm/modeling_wavlm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1107,7 +1107,7 @@ def __init__(self, config: WavLMConfig):

# model only needs masking vector if mask prob is > 0.0
if config.mask_time_prob > 0.0 or config.mask_feature_prob > 0.0:
self.masked_spec_embed = nn.Parameter(torch.FloatTensor(config.hidden_size).uniform_())
self.masked_spec_embed = nn.Parameter(torch.Tensor(config.hidden_size).uniform_())

if config.do_stable_layer_norm:
self.encoder = WavLMEncoderStableLayerNorm(config)
Expand Down Expand Up @@ -1158,7 +1158,7 @@ def _mask_hidden_states(

if mask_time_indices is not None:
# apply SpecAugment along time axis with given mask_time_indices
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed
elif self.config.mask_time_prob > 0 and self.training:
mask_time_indices = _compute_mask_indices(
(batch_size, sequence_length),
Expand All @@ -1168,7 +1168,7 @@ def _mask_hidden_states(
min_masks=self.config.mask_time_min_masks,
)
mask_time_indices = torch.tensor(mask_time_indices, device=hidden_states.device, dtype=torch.bool)
hidden_states[mask_time_indices] = self.masked_spec_embed.to(hidden_states.dtype)
hidden_states[mask_time_indices] = self.masked_spec_embed

if self.config.mask_feature_prob > 0 and self.training:
# generate indices & apply SpecAugment along feature axis
Expand Down
Loading