Deepseek 4: -sm tensor (#26490)

* DSV4: sm tensor

* set coarser granularity for head splits

* fix dspark

* add model saving for dsv4 + allow dflash to return on specific device

* add comment about dsv4 seq_rm

* simplify

* add shared expert delayed allreduce

* remove special test for dsv4
This commit is contained in:
Aman Gupta
2026-08-24 11:50:25 +05:30
committed by GitHub
parent c060ca974c
commit bf0a29cc16
7 changed files with 314 additions and 48 deletions
-1
View File
@@ -1060,7 +1060,6 @@ bool llm_arch_supports_sm_tensor(const llm_arch & arch) {
case LLM_ARCH_OLMOE:
case LLM_ARCH_DEEPSEEK2:
case LLM_ARCH_DEEPSEEK32:
case LLM_ARCH_DEEPSEEK4:
case LLM_ARCH_DOTS3NOTE:
case LLM_ARCH_GLM_DSA:
case LLM_ARCH_BITNET: