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 6, 14:30 UTC
Fix EagleMistralLarge3Model initialization (#37232)
Signed-off-by: juliendenize <julien.denize@mistral.ai>
Julien Denize committed
Mar 16, 2026 at 23:41 UTC
7961486a9b749b1b60d8b6fd5fb7d61596a9b041
1 file changed
+1
vllm/model_executor/models/mistral_large_3_eagle.py
+1
@@ -74,6 +74,7 @@ class EagleMistralLarge3Model(DeepseekV2Model):
74
prefix=maybe_prefix(prefix, "fc"),
75
)
76
self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
77
+ self.aux_hidden_state_layers: tuple[int, ...] = ()
78
self.make_empty_intermediate_tensors = make_empty_intermediate_tensors_factory(
79
["hidden_states", "residual"], config.hidden_size
80
)