[BugFix][Qwen3.8] Support Qwen3.8-MXFP4 DCP by registing Qwen3_5 text-only archs in mamba radix cache whitelists (#35297)
This commit is contained in:
@@ -1680,6 +1680,10 @@ _MAMBA_RADIX_CACHE_ARCHS = frozenset(
|
|||||||
"InternS2PreviewForConditionalGeneration",
|
"InternS2PreviewForConditionalGeneration",
|
||||||
"InternS2MobiusForConditionalGeneration",
|
"InternS2MobiusForConditionalGeneration",
|
||||||
"Qwen3_5ForConditionalGeneration",
|
"Qwen3_5ForConditionalGeneration",
|
||||||
|
# Text-only entries of the same hybrid stack (models/qwen3_5_text.py);
|
||||||
|
# Qwen3.8-2.4T-A95B ships as Qwen3_5MoeForCausalLM.
|
||||||
|
"Qwen3_5MoeForCausalLM",
|
||||||
|
"Qwen3_5ForCausalLM",
|
||||||
"MiniCPMV4_6ForConditionalGeneration",
|
"MiniCPMV4_6ForConditionalGeneration",
|
||||||
"NemotronHForCausalLM",
|
"NemotronHForCausalLM",
|
||||||
"NemotronHPuzzleForCausalLM",
|
"NemotronHPuzzleForCausalLM",
|
||||||
@@ -1700,6 +1704,10 @@ _MAMBA_EXTRA_BUFFER_ARCHS = frozenset(
|
|||||||
"KimiLinearForCausalLM",
|
"KimiLinearForCausalLM",
|
||||||
"Qwen3_5ForConditionalGeneration",
|
"Qwen3_5ForConditionalGeneration",
|
||||||
"Qwen3_5MoeForConditionalGeneration",
|
"Qwen3_5MoeForConditionalGeneration",
|
||||||
|
# Text-only entries of the same hybrid stack (models/qwen3_5_text.py);
|
||||||
|
# Qwen3.8-2.4T-A95B ships as Qwen3_5MoeForCausalLM.
|
||||||
|
"Qwen3_5MoeForCausalLM",
|
||||||
|
"Qwen3_5ForCausalLM",
|
||||||
"Qwen3NextForCausalLM",
|
"Qwen3NextForCausalLM",
|
||||||
"InternS2PreviewForConditionalGeneration",
|
"InternS2PreviewForConditionalGeneration",
|
||||||
"MiniCPMV4_6ForConditionalGeneration",
|
"MiniCPMV4_6ForConditionalGeneration",
|
||||||
|
|||||||
Reference in New Issue
Block a user