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

Fixes assertion failure in prefix caching: the lora index mapping should respect prefix_len (#2688)

Browse Source 
Signed-off-by: Tao He <sighingnow@gmail.com>
This commit is contained in:
Tao He 2024-02-01 01:00:13 +08:00 committed by GitHub
parent 1af090b57d
commit d69ff0cbbb
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/worker/model_runner.py
View File

@ -142,10 +142,10 @@ class ModelRunner:
if lora_id > 0: if lora_id > 0:
lora_requests.add(seq_group_metadata.lora_request) lora_requests.add(seq_group_metadata.lora_request)
lora_index_mapping.append([lora_id] * prompt_len) lora_index_mapping.append([lora_id] * (prompt_len - prefix_len))
lora_prompt_mapping.extend( lora_prompt_mapping.extend(
[lora_id] * [lora_id] *
(prompt_len (prompt_len - prefix_len
if seq_group_metadata.sampling_params.prompt_logprobs else 1)) if seq_group_metadata.sampling_params.prompt_logprobs else 1))
if seq_group_metadata.block_tables is None: if seq_group_metadata.block_tables is None: