biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[BugFix] Fix Llama4 - Index Error When Single Request Near Max Context (#16209)

Browse Source 
Signed-off-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>
This commit is contained in:
Lucas Wilkinson
2025-04-08 14:56:51 -04:00
committed by GitHub
parent 0115ccd5c0
commit e1a2c699dd
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/v1/attention/backends/flash_attn.py
View File

@@ -264,7 +264,7 @@ def make_local_attention_virtual_batches(
np.arange(pages_per_local_batch, dtype=np.int32), np.arange(pages_per_local_batch, dtype=np.int32),
(virtual_batches, pages_per_local_batch)) \ (virtual_batches, pages_per_local_batch)) \
+ np.expand_dims(block_starts, axis=1) + np.expand_dims(block_starts, axis=1)
block_indices = block_indices.flatten() block_indices = block_indices.flatten().clip(max=block_table.shape[1] - 1)
batch_indices = np.repeat(np.arange(actual_batch_size, dtype=np.int32), batch_indices = np.repeat(np.arange(actual_batch_size, dtype=np.int32),
local_blocks * pages_per_local_batch) local_blocks * pages_per_local_batch)
block_table_local = block_table[batch_indices, block_indices]\ block_table_local = block_table[batch_indices, block_indices]\