A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 16:00 UTC
[Bugfix]: Fix is_embedding_layer condition in VocabParallelEmbedding (#15824)
Signed-off-by: alexwl <alexey.a.kiryushin@gmail.com>
Alexey Kiryushin committed
Apr 1, 2025 at 05:57 UTC
63d8eabed05a632679f0e9e929d52afb08d2ec7f
1 file changed
+1
-1
vllm/model_executor/layers/vocab_parallel_embedding.py
+1
-1
@@ -235,7 +235,7 @@ class VocabParallelEmbedding(torch.nn.Module):
235
# If we are making an embedding layer, then our quantization linear
236
# method must implement the embedding operation. If we are another
237
# layer type like ParallelLMHead, this is not important.
238
- is_embedding_layer = type(self.__class__) is VocabParallelEmbedding
238
+ is_embedding_layer = type(self) is VocabParallelEmbedding
239
quant_method_implements_embedding = method_has_implemented_embedding(
240
type(quant_method))
241
if is_embedding_layer and not quant_method_implements_embedding: