vllm.triton_utils
¶
Modules:
-
dispatcher–Registry for overriding Triton kernels with platform implementations.
-
force_first_config–Skip Triton autotuning under VLLM_TRITON_FORCE_FIRST_CONFIG.
-
tensor_descriptor–
Functions:
-
tensor_descriptor_compatible–Whether every tensor can back a tensor descriptor spanning its last dim.
-
use_tensor_descriptor–Tri-state VLLM_TRITON_USE_TD: unset=auto (on for XPU), 1/0=force on/off.
tensor_descriptor_compatible(*tensors)
¶
Whether every tensor can back a tensor descriptor spanning its last dim.
None (unused optional operand) is accepted.
Source code in vllm/triton_utils/tensor_descriptor.py
use_tensor_descriptor(override=None)
¶
Tri-state VLLM_TRITON_USE_TD: unset=auto (on for XPU), 1/0=force on/off.