[Model] add GLM-5.3-Flash support (#53906)

This commit is contained in:
Jiangyun Zhu
2026-09-03 09:40:35 -07:00
committed by GitHub
parent da8ec28268
commit 98ed0856f3
94 changed files with 17927 additions and 266 deletions
+7
View File
@@ -61,6 +61,7 @@ MTPModelTypes = Literal[
"hy_v4_mtp",
"gemma4_mtp",
"inkling_mtp",
"glm5_next_mtp",
]
NgramGPUTypes = Literal["ngram_gpu"]
DFlashModelTypes = Literal["dflash"]
@@ -1016,6 +1017,12 @@ class SpeculativeConfig:
hf_config.update(
{"n_predict": n_predict, "architectures": ["MiniMaxM3MTP"]}
)
if hf_config.model_type == "glm5_next":
hf_config.model_type = "glm5_next_mtp"
n_predict = hf_config.num_nextn_predict_layers
hf_config.update(
{"n_predict": n_predict, "architectures": ["Glm5NextMTPModel"]}
)
return hf_config
+3
View File
@@ -81,6 +81,9 @@ DEFAULT_BREAKABLE_CUDAGRAPH_ARCHITECTURES = frozenset(
"DeepSeekV4MTPModel",
"Dots3NoteForCausalLM",
"Dots3NoteMTPModel",
"Glm5NextForCausalLM",
"Glm5NextForConditionalGeneration",
"Glm5NextMTPModel",
"GlmMoeDsaForCausalLM",
"HYV4ForCausalLM",
"HYV4MTPModel",