[1/N][Cleanup] Standardize on use of is_quantized_kv_cache (#38659)
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
@@ -61,6 +61,10 @@ MODELOPT_TO_VLLM_KV_CACHE_DTYPE_MAP = {
|
||||
T = TypeVar("T")
|
||||
|
||||
|
||||
def is_quantized_kv_cache(kv_cache_dtype: str) -> bool:
|
||||
return kv_cache_dtype.startswith("fp8")
|
||||
|
||||
|
||||
def is_strictly_contiguous(t: torch.Tensor) -> bool:
|
||||
"""
|
||||
Check if tensor is contiguous AND has no degenerate strides.
|
||||
|
||||
Reference in New Issue
Block a user