[GLM-5.3-Flash] Drop the NextN layer from the KDA state pool (#44098)

Co-authored-by: Mohammad Angkad <mohammad.angkad@radixark.ai>
This commit is contained in:
Mohammad Miadh Angkad
2026-10-13 19:31:02 -07:00
committed by GitHub
co-authored by Mohammad Angkad
parent ac187c0cd5
commit 7f0434e4af
2 changed files with 3 additions and 11 deletions
-5
View File
@@ -250,11 +250,6 @@ class Glm5NextTextConfig(PretrainedConfig):
def linear_layer_ids(self):
return [i for i in range(self.num_hidden_layers) if self.is_kda_layer(i)]
@property
def nextn_layer_ids(self):
num_nextn_layers = self.num_nextn_predict_layers or 0
return [self.num_hidden_layers + i for i in range(num_nextn_layers)]
@property
def full_attention_layer_ids(self):
return [i for i in range(self.num_hidden_layers) if not self.is_kda_layer(i)]
@@ -1438,16 +1438,13 @@ class KVCacheConfigurator:
)
def _get_mamba_layer_ids_for_req_pool(self) -> list:
mamba_layer_ids = [
# Target layers on this rank only; a draft holding linear state gets its own
# pool via clone_with_new_mamba.
return [
i
for i in self.mambaish_config.mamba2_cache_params.layers
if self.layer_info.start_layer <= i < self.layer_info.end_layer
]
if max_speculative_num_draft_tokens():
for layer_id in getattr(self.mambaish_config, "nextn_layer_ids", []):
if layer_id not in mamba_layer_ids:
mamba_layer_ids.append(layer_id)
return mamba_layer_ids
def _get_ple_req_pool_kwargs(self) -> dict[str, Any]:
from sglang.srt.configs.qwen4_exp import Qwen4ExpTextConfig