A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 19:45 UTC
fix some typos : supported_head_sizes (#14627)
Yang.Tao committed
Mar 12, 2025 at 01:38 UTC
53056731fdf8aa8f960cd3473b01a36d5cb0281d
1 file changed
+3
-3
vllm/attention/backends/blocksparse_attn.py
+3
-3
@@ -335,11 +335,11 @@ class BlocksparseFlashAttentionImpl(AttentionImpl):
335
self.sparse_block_size = self.blocksparse_params.block_size
336
self.head_sliding_step = self.blocksparse_params.head_sliding_step
337
338
- suppored_head_sizes = PagedAttention.get_supported_head_sizes()
339
- if head_size not in suppored_head_sizes:
338
+ supported_head_sizes = PagedAttention.get_supported_head_sizes()
339
+ if head_size not in supported_head_sizes:
340
raise ValueError(
341
f"Head size {head_size} is not supported by PagedAttention. "
342
- f"Supported head sizes are: {suppored_head_sizes}.")
342
+ f"Supported head sizes are: {supported_head_sizes}.")
343
344
self.tp_size = get_tensor_model_parallel_world_size()
345
self.tp_rank = get_tensor_model_parallel_rank()