[Misc] Remove no-op self-assignments across vLLM (#55988)
Signed-off-by: shaolila <shaolila@buaa.edu.cn> Co-authored-by: shaolila <shaolila@buaa.edu.cn>
This commit is contained in:
@@ -615,14 +615,15 @@ class RoutedExperts(PluggableLayer):
|
|||||||
# compressed-tensors checkpoints with packed weights are stored flipped
|
# compressed-tensors checkpoints with packed weights are stored flipped
|
||||||
# TODO (mgoin): check self.quant_method.quant_config.quant_format
|
# TODO (mgoin): check self.quant_method.quant_config.quant_format
|
||||||
# against known CompressionFormat enum values that have this quality
|
# against known CompressionFormat enum values that have this quality
|
||||||
if quant_method_name in (
|
if (
|
||||||
"CompressedTensorsWNA16MoEMethod",
|
quant_method_name
|
||||||
"CompressedTensorsW4A16FlydslMoEMethod",
|
in (
|
||||||
|
"CompressedTensorsWNA16MoEMethod",
|
||||||
|
"CompressedTensorsW4A16FlydslMoEMethod",
|
||||||
|
)
|
||||||
|
and is_transposed
|
||||||
):
|
):
|
||||||
if is_transposed:
|
loaded_weight = loaded_weight.t().contiguous()
|
||||||
loaded_weight = loaded_weight.t().contiguous()
|
|
||||||
else:
|
|
||||||
loaded_weight = loaded_weight
|
|
||||||
|
|
||||||
if shard_id not in ("w1", "w2", "w3"):
|
if shard_id not in ("w1", "w2", "w3"):
|
||||||
raise ValueError(f"shard_id must be ['w1','w2','w3'] but got {shard_id}.")
|
raise ValueError(f"shard_id must be ['w1','w2','w3'] but got {shard_id}.")
|
||||||
|
|||||||
@@ -780,9 +780,7 @@ class DeepCLIPVisionEmbeddings(CLIPVisionEmbeddings):
|
|||||||
self, pixel_values: torch.Tensor, patch_embeds: torch.Tensor | None = None
|
self, pixel_values: torch.Tensor, patch_embeds: torch.Tensor | None = None
|
||||||
) -> torch.Tensor:
|
) -> torch.Tensor:
|
||||||
batch_size = pixel_values.shape[0]
|
batch_size = pixel_values.shape[0]
|
||||||
if patch_embeds is not None:
|
if patch_embeds is None:
|
||||||
patch_embeds = patch_embeds
|
|
||||||
else:
|
|
||||||
patch_embeds = self.patch_embedding(pixel_values)
|
patch_embeds = self.patch_embedding(pixel_values)
|
||||||
patch_embeds = patch_embeds.flatten(2).transpose(1, 2)
|
patch_embeds = patch_embeds.flatten(2).transpose(1, 2)
|
||||||
|
|
||||||
|
|||||||
@@ -964,7 +964,6 @@ class MiMoV2OmniMultiModalProcessor(BaseMultiModalProcessor[MiMoV2OmniProcessing
|
|||||||
assert isinstance(va_item, (list, tuple)) and len(va_item) == 2
|
assert isinstance(va_item, (list, tuple)) and len(va_item) == 2
|
||||||
vid, audio_src = va_item
|
vid, audio_src = va_item
|
||||||
va_item = VideoAudioInput(video=vid, audio=audio_src)
|
va_item = VideoAudioInput(video=vid, audio=audio_src)
|
||||||
vid = vid
|
|
||||||
# Convert video frames to (TCHW, timestamps) if needed
|
# Convert video frames to (TCHW, timestamps) if needed
|
||||||
if (
|
if (
|
||||||
isinstance(vid, tuple)
|
isinstance(vid, tuple)
|
||||||
|
|||||||
@@ -173,7 +173,6 @@ class Step3p5Attention(nn.Module):
|
|||||||
rope_scaling = None
|
rope_scaling = None
|
||||||
|
|
||||||
if sliding_window is not None and enable_sliding_window:
|
if sliding_window is not None and enable_sliding_window:
|
||||||
sliding_window = sliding_window
|
|
||||||
if swa_num_attention_heads is not None:
|
if swa_num_attention_heads is not None:
|
||||||
num_heads = swa_num_attention_heads
|
num_heads = swa_num_attention_heads
|
||||||
self.total_num_heads = swa_num_attention_heads
|
self.total_num_heads = swa_num_attention_heads
|
||||||
|
|||||||
Reference in New Issue
Block a user