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 12, 23:30 UTC
[config] Expose `get_total_num_hidden_layers()` in ModelConfig (#28961)
Signed-off-by: tovam <tovam@pliops.com> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com> Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
Tova Movshovitz committed
Nov 19, 2025 at 13:37 UTC
ba558c029ad65ab4f040c8320607ebd87612cf08
1 file changed
+10
-5
vllm/config/model.py
+10
-5
@@ -1369,11 +1369,7 @@ class ModelConfig:
1369
# Coerce to 0 if explicitly set to None
1370
return num_experts or 0
1371
1372
- def get_layers_start_end_indices(
1373
- self, parallel_config: ParallelConfig
1374
- ) -> tuple[int, int]:
1375
- from vllm.distributed.utils import get_pp_indices
1376
-
1372
+ def get_total_num_hidden_layers(self) -> int:
1373
if (
1374
self.hf_text_config.model_type == "deepseek_mtp"
1375
or self.hf_config.model_type == "mimo_mtp"
@@ -1393,6 +1389,15 @@ class ModelConfig:
1389
total_num_hidden_layers = getattr(
1390
self.hf_text_config, "num_hidden_layers", 0
1391
)
1392
+ return total_num_hidden_layers
1393
+
1394
+ def get_layers_start_end_indices(
1395
+ self, parallel_config: ParallelConfig
1396
+ ) -> tuple[int, int]:
1397
+ from vllm.distributed.utils import get_pp_indices
1398
+
1399
+ total_num_hidden_layers = self.get_total_num_hidden_layers()
1400
+
1401
# the layout order is: DP x PP x TP
1402
pp_rank = (
1403
parallel_config.rank // parallel_config.tensor_parallel_size