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 16, 09:30 UTC
[Bugfix] Eagle: change config name for fc bias (#9580)
gopalsarda committed
Oct 22, 2024 at 21:44 UTC
08075c34483843c75b4420bac92377b59ff9a8ac
1 file changed
+1
-1
vllm/model_executor/models/eagle.py
+1
-1
@@ -44,7 +44,7 @@ class EAGLE(nn.Module):
44
self.model = model_cls(self.config.model, *args, **kwargs)
45
self.fc = nn.Linear(config.model.hidden_size * 2,
46
config.model.hidden_size,
47
- bias=getattr(self.config, "bias", False))
47
+ bias=getattr(self.config, "eagle_fc_bias", False))
48
49
self.orig_vocab_size = config.vocab_size
50
self.truncated_vocab_size = config.truncated_vocab_size