fix some typos : supported_head_sizes (#14627)
This commit is contained in:
@@ -335,11 +335,11 @@ class BlocksparseFlashAttentionImpl(AttentionImpl):
|
|||||||
self.sparse_block_size = self.blocksparse_params.block_size
|
self.sparse_block_size = self.blocksparse_params.block_size
|
||||||
self.head_sliding_step = self.blocksparse_params.head_sliding_step
|
self.head_sliding_step = self.blocksparse_params.head_sliding_step
|
||||||
|
|
||||||
suppored_head_sizes = PagedAttention.get_supported_head_sizes()
|
supported_head_sizes = PagedAttention.get_supported_head_sizes()
|
||||||
if head_size not in suppored_head_sizes:
|
if head_size not in supported_head_sizes:
|
||||||
raise ValueError(
|
raise ValueError(
|
||||||
f"Head size {head_size} is not supported by PagedAttention. "
|
f"Head size {head_size} is not supported by PagedAttention. "
|
||||||
f"Supported head sizes are: {suppored_head_sizes}.")
|
f"Supported head sizes are: {supported_head_sizes}.")
|
||||||
|
|
||||||
self.tp_size = get_tensor_model_parallel_world_size()
|
self.tp_size = get_tensor_model_parallel_world_size()
|
||||||
self.tp_rank = get_tensor_model_parallel_rank()
|
self.tp_rank = get_tensor_model_parallel_rank()
|
||||||
|
|||||||
Reference in New Issue
Block a user