[Docs] Replace rst style double-backtick with md single-backtick (#27091)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
@@ -21,7 +21,7 @@ from vllm.utils import cdiv, has_deep_gemm
|
||||
|
||||
@functools.cache
|
||||
def is_deep_gemm_supported() -> bool:
|
||||
"""Return ``True`` if DeepGEMM is supported on the current platform.
|
||||
"""Return `True` if DeepGEMM is supported on the current platform.
|
||||
Currently, only Hopper and Blackwell GPUs are supported.
|
||||
"""
|
||||
is_supported_arch = current_platform.is_cuda() and (
|
||||
@@ -33,7 +33,7 @@ def is_deep_gemm_supported() -> bool:
|
||||
|
||||
@functools.cache
|
||||
def is_deep_gemm_e8m0_used() -> bool:
|
||||
"""Return ``True`` if vLLM is configured to use DeepGEMM "
|
||||
"""Return `True` if vLLM is configured to use DeepGEMM "
|
||||
"E8M0 scale on a Hopper or Blackwell-class GPU.
|
||||
"""
|
||||
if not is_deep_gemm_supported():
|
||||
@@ -311,9 +311,9 @@ def calc_diff(x: torch.Tensor, y: torch.Tensor):
|
||||
"""Return a global difference metric for unit tests.
|
||||
|
||||
DeepGEMM kernels on Blackwell/B200 currently exhibit noticeable per-element
|
||||
error, causing ``torch.testing.assert_close`` to fail. Instead of checking
|
||||
error, causing `torch.testing.assert_close` to fail. Instead of checking
|
||||
every element, we compute a cosine-style similarity over the whole tensor
|
||||
and report ``1 - sim``. Once kernel accuracy improves this helper can be
|
||||
and report `1 - sim`. Once kernel accuracy improves this helper can be
|
||||
removed.
|
||||
"""
|
||||
|
||||
|
||||
+10
-10
@@ -34,7 +34,7 @@ FLASHINFER_CUBINS_REPOSITORY = os.environ.get(
|
||||
|
||||
@functools.cache
|
||||
def has_flashinfer() -> bool:
|
||||
"""Return ``True`` if FlashInfer is available."""
|
||||
"""Return `True` if FlashInfer is available."""
|
||||
# Use find_spec to check if the module exists without importing it
|
||||
# This avoids potential CUDA initialization side effects
|
||||
if importlib.util.find_spec("flashinfer") is None:
|
||||
@@ -114,13 +114,13 @@ autotune = _lazy_import_wrapper(
|
||||
|
||||
@functools.cache
|
||||
def has_flashinfer_comm() -> bool:
|
||||
"""Return ``True`` if FlashInfer comm module is available."""
|
||||
"""Return `True` if FlashInfer comm module is available."""
|
||||
return has_flashinfer() and importlib.util.find_spec("flashinfer.comm") is not None
|
||||
|
||||
|
||||
@functools.cache
|
||||
def has_flashinfer_all2all() -> bool:
|
||||
"""Return ``True`` if FlashInfer mnnvl all2all is available."""
|
||||
"""Return `True` if FlashInfer mnnvl all2all is available."""
|
||||
if not has_flashinfer_comm():
|
||||
return False
|
||||
|
||||
@@ -141,7 +141,7 @@ def has_flashinfer_all2all() -> bool:
|
||||
|
||||
@functools.cache
|
||||
def has_flashinfer_moe() -> bool:
|
||||
"""Return ``True`` if FlashInfer MoE module is available."""
|
||||
"""Return `True` if FlashInfer MoE module is available."""
|
||||
return (
|
||||
has_flashinfer()
|
||||
and importlib.util.find_spec("flashinfer.fused_moe") is not None
|
||||
@@ -150,7 +150,7 @@ def has_flashinfer_moe() -> bool:
|
||||
|
||||
@functools.cache
|
||||
def has_flashinfer_cutlass_fused_moe() -> bool:
|
||||
"""Return ``True`` if FlashInfer CUTLASS fused MoE is available."""
|
||||
"""Return `True` if FlashInfer CUTLASS fused MoE is available."""
|
||||
if not has_flashinfer_moe():
|
||||
return False
|
||||
|
||||
@@ -171,7 +171,7 @@ def has_flashinfer_cutlass_fused_moe() -> bool:
|
||||
|
||||
@functools.cache
|
||||
def has_nvidia_artifactory() -> bool:
|
||||
"""Return ``True`` if NVIDIA's artifactory is accessible.
|
||||
"""Return `True` if NVIDIA's artifactory is accessible.
|
||||
|
||||
This checks connectivity to the kernel inference library artifactory
|
||||
which is required for downloading certain cubin kernels like TRTLLM FHMA.
|
||||
@@ -218,9 +218,9 @@ def _force_use_trtllm_attention(env_value: bool | None) -> bool | None:
|
||||
|
||||
def force_use_trtllm_attention() -> bool | None:
|
||||
"""
|
||||
Return ``None`` if VLLM_USE_TRTLLM_ATTENTION is not set,
|
||||
return ``True`` if TRTLLM attention is forced to be used,
|
||||
return ``False`` if TRTLLM attention is forced to be not used.
|
||||
Return `None` if VLLM_USE_TRTLLM_ATTENTION is not set,
|
||||
return `True` if TRTLLM attention is forced to be used,
|
||||
return `False` if TRTLLM attention is forced to be not used.
|
||||
"""
|
||||
return _force_use_trtllm_attention(envs.VLLM_USE_TRTLLM_ATTENTION)
|
||||
|
||||
@@ -244,7 +244,7 @@ def use_trtllm_attention(
|
||||
has_sinks: bool = False,
|
||||
has_spec: bool = False,
|
||||
) -> bool:
|
||||
"""Return ``True`` if TRTLLM attention is used."""
|
||||
"""Return `True` if TRTLLM attention is used."""
|
||||
force_use_trtllm = force_use_trtllm_attention()
|
||||
|
||||
# Environment variable is set to 0 - respect it
|
||||
|
||||
Reference in New Issue
Block a user