wassname/vllm
1
0 Fork 0
mirror of https://github.com/wassname/vllm.git synced 2026-10-01 12:42:39 +08:00
Code Issues Packages Projects Releases Wiki Activity

Consider pempty tensor

Browse Source
This commit is contained in:
Woosuk Kwon committed 2023-02-23 20:20:33 +00:00
1 parent 86f9eb6d39
commit 7f985166f7
1 file changed
+4 -1
Download Patch File Download Diff File Expand all files Collapse all files
cacheflow/models/input_metadata.py
+4 -1
View File
@@ -23,6 +23,9 @@ class InputMetadata:
self.num_prompts = len(prompt_lens)
self.num_generation_tokens = context_lens.shape[0]
self.max_num_blocks_per_seq = block_tables.shape[1]
if block_tables.numel() > 0:
self.max_num_blocks_per_seq = block_tables.shape[1]
else:
self.max_num_blocks_per_seq = 0
assert self.num_generation_tokens == block_tables.shape[0]
assert self.num_prompts + self.num_generation_tokens == len(seq_ids)