Skip to content

Commit d5cc2a9

Browse files
committed
bugfix: correct start prefix match index for chunked prefill.
1 parent 0356c38 commit d5cc2a9

File tree

1 file changed

+1
-1
lines changed

1 file changed

+1
-1
lines changed

xllm/core/framework/prefix_cache/prefix_cache.cpp

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -54,7 +54,7 @@ std::vector<Block> PrefixCache::match(
5454
existed_shared_blocks.empty()
5555
? Murmur3Key{}
5656
: Murmur3Key{existed_shared_blocks.back().get_immutable_hash_value()};
57-
for (size_t i = 0; i < n_tokens; i += block_size_) {
57+
for (size_t i = start_index; i < n_tokens; i += block_size_) {
5858
if (i == 0) {
5959
murmur_hash3(
6060
nullptr, token_ids.slice(i, i + block_size_), token_hash_key.data);

0 commit comments

Comments
 (0)