Skip to content

GitLab

Unverified Commit a2986b3e authored by kourosh hakhamaneshi's avatar kourosh hakhamaneshi Committed by GitHub
Browse files

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


Signed-off-by: default avatarKourosh Hakhamaneshi <Kourosh@anyscale.com>
parent 96b9aa5a
Hide whitespace changes
Inline Side-by-side
Showing with 4 additions and 3 deletions
vllm/benchmarks/datasets.py
View file @ a2986b3e
...@@ -2979,13 +2979,14 @@ class PrefixRepetitionRandomDataset(BenchmarkDataset): ...@@ -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)
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment