[PD]添加VLLM_PP_LAYER_PARTITION_D控制d层切分

vllm/envs.py

@@ -47,6 +47,7 @@ if TYPE_CHECKING:
     VLLM_TRACE_FUNCTION: int = 0
     VLLM_USE_FLASHINFER_SAMPLER: bool | None = None
     VLLM_PP_LAYER_PARTITION: str | None = None
+    VLLM_PP_LAYER_PARTITION_D: Optional[str] = None
     VLLM_CPU_KVCACHE_SPACE: int | None = 0
     VLLM_CPU_OMP_THREADS_BIND: str = ""
     VLLM_CPU_NUM_OF_RESERVED_CPU: int | None = None
@@ -754,6 +755,11 @@ environment_variables: dict[str, Callable[[], Any]] = {
     else None,
     # Pipeline stage partition strategy
     "VLLM_PP_LAYER_PARTITION": lambda: os.getenv("VLLM_PP_LAYER_PARTITION", None),
+    
+    # Pipeline stage partition strategy
+    "VLLM_PP_LAYER_PARTITION_D":
+    lambda: os.getenv("VLLM_PP_LAYER_PARTITION_D", None),
+
     # (CPU backend only) CPU key-value cache space.
     # default is None and will be set as 4 GB
     "VLLM_CPU_KVCACHE_SPACE": lambda: int(os.getenv("VLLM_CPU_KVCACHE_SPACE", "0"))