biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[6/N] pass whole config to inner model (#10205)

Browse Source 
Signed-off-by: youkaichao <youkaichao@gmail.com>
This commit is contained in:
youkaichao
2024-11-10 22:41:46 -08:00
committed by GitHub
parent f0f2e5638e
commit f89d18ff74
69 changed files with 681 additions and 963 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/model_executor/models/eagle.py
View File

@@ -14,6 +14,8 @@ from vllm.model_executor.models import ModelRegistry
from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.sequence import IntermediateTensors
from .utils import maybe_prefix
class EAGLE(nn.Module):
"""This class implements the EAGLE draft model from the paper: https://arxiv.org/pdf/2401.15077
@@ -42,7 +44,8 @@ class EAGLE(nn.Module):
architectures = getattr(self.config.model, "architectures", [])
model_cls, _ = ModelRegistry.resolve_model_cls(architectures)
self.model = model_cls(vllm_config, prefix)
self.model = model_cls(vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"))
self.fc = nn.Linear(config.model.hidden_size * 2,
config.model.hidden_size,
bias=getattr(self.config, "eagle_fc_bias", False))