Merge branch 'v0.9.2-dev-bmm-2cd-0206-rebase-newest' into 'v0.9.2-dev'

去掉DTBmm的一个冗余条件

See merge request dcutoolkit/deeplearing/vllm!416

vllm/v1/attention/backends/mla/common.py

@@ -936,7 +936,6 @@ class MLACommonImpl(MLAAttentionImpl[M], Generic[M]):
         
     def enable_fused_DTBmm(self):
         if envs.VLLM_USE_FUSED_DTBMM and \
-            self.kv_cache_dtype == "fp8_e4m3" and \
                 torch.cuda.get_device_properties("cuda").gcnArchName.split(':')[0] == "gfx938":
             return True
         else: