fix: typos (#18151)
Signed-off-by: omahs <73983677+omahs@users.noreply.github.com>
This commit is contained in:
@@ -142,7 +142,7 @@ def mamba_v2_sharded_weight_loader(
|
||||
) -> LoaderFunction:
|
||||
"""Create a weight loader for mamba v2. This ensures that the projections
|
||||
are correctly sharded so that they can be split into x, B, C. It also
|
||||
ensures the the all the groups corresponding to a head shard is placed
|
||||
ensures that all the groups corresponding to a head shard is placed
|
||||
together with it.
|
||||
"""
|
||||
|
||||
|
||||
@@ -21,7 +21,7 @@
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
"""Inference-only IBM Granite speeech model."""
|
||||
"""Inference-only IBM Granite speech model."""
|
||||
import math
|
||||
from collections.abc import Iterable, Mapping
|
||||
from typing import Optional, TypedDict, Union
|
||||
@@ -626,7 +626,7 @@ class GraniteSpeechForConditionalGeneration(
|
||||
audio_embed_sizes: torch.Tensor,
|
||||
) -> torch.Tensor:
|
||||
"""Calculate the input features mask, which will generally be used
|
||||
to mask the the padded features for all entries in the batch except
|
||||
to mask the padded features for all entries in the batch except
|
||||
for those with the most audio features.
|
||||
|
||||
Args:
|
||||
|
||||
@@ -91,9 +91,9 @@ class ConformerEncoderLayer(nn.Module):
|
||||
if set to True, use GLULinear module,
|
||||
otherwise, used GLUPointWiseConv module.
|
||||
default to False.
|
||||
attention_innner_dim: int, optional
|
||||
attention_inner_dim: int, optional
|
||||
if equal to -1, attention dim for linears k/q/v is
|
||||
equal to d_model. otherwise attention_innner_dim is used.
|
||||
equal to d_model. otherwise attention_inner_dim is used.
|
||||
default -1.
|
||||
attention_glu_type: str, optional
|
||||
activation function for glu used in the multihead attention,
|
||||
@@ -148,7 +148,7 @@ class ConformerEncoderLayer(nn.Module):
|
||||
conv_glu_type="sigmoid",
|
||||
bias_in_glu=True,
|
||||
linear_glu_in_convm=False,
|
||||
attention_innner_dim=-1,
|
||||
attention_inner_dim=-1,
|
||||
attention_glu_type="swish",
|
||||
activation_checkpointing="",
|
||||
export=False,
|
||||
@@ -169,7 +169,7 @@ class ConformerEncoderLayer(nn.Module):
|
||||
n_head,
|
||||
d_model,
|
||||
dropout_rate,
|
||||
attention_innner_dim,
|
||||
attention_inner_dim,
|
||||
attention_glu_type,
|
||||
bias_in_glu,
|
||||
use_pt_scaled_dot_product_attention=
|
||||
|
||||
Reference in New Issue
Block a user