xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-04-30 17:50:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fixes prefix-repetition benchmark script (#26828)

Browse Source 
Signed-off-by: Kourosh Hakhamaneshi <Kourosh@anyscale.com>
This commit is contained in:
kourosh hakhamaneshi 2025-10-14 19:54:43 -07:00 committed by GitHub
parent 96b9aa5aa0
commit a2986b3e33
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/benchmarks/datasets.py
View File

@ -2979,13 +2979,14 @@ class PrefixRepetitionRandomDataset(BenchmarkDataset):
requests = [] requests = []
token_mismatch_total = 0 token_mismatch_total = 0
for _ in range(num_prefixes): for _ in range(num_prefixes):
prefix_tokens = _generate_exact_length_tokens(prefix_len) prefix_tokens, prefix_mismatch = _generate_exact_length_tokens(prefix_len)
token_mismatch_total += prefix_mismatch
for _ in range(prompts_per_prefix): for _ in range(prompts_per_prefix):
suffix_tokens, token_mistmatch = _generate_exact_length_tokens( suffix_tokens, suffix_mismatch = _generate_exact_length_tokens(
suffix_len suffix_len
) )
token_mismatch_total += token_mistmatch token_mismatch_total += suffix_mismatch
combined_tokens = prefix_tokens + suffix_tokens combined_tokens = prefix_tokens + suffix_tokens
prompt = tokenizer.decode(combined_tokens) prompt = tokenizer.decode(combined_tokens)
prompt_len = len(combined_tokens) prompt_len = len(combined_tokens)