Fix: Correctly filter special tokens in benchmark_prefix_caching (vllm-project#28615)

dw2761 · web-flow · commit e63fd445605b · 2025-11-13T10:57:44.000Z
Signed-off-by: Di Wu &lt;dw2761@nyu.edu&gt;
diff --git a/benchmarks/benchmark_prefix_caching.py b/benchmarks/benchmark_prefix_caching.py
@@ -69,7 +69,7 @@ def sample_tokens(tokenizer: PreTrainedTokenizerBase, length: int) -> list[int]:
 
     # Remove the special tokens.
     return random.choices(
-        [v for k, v in vocab.items() if k not in all_special_ids],
+        [v for v in vocab.values() if v not in all_special_ids],
         k=length,
     )
 

Original file line number	Diff line number	Diff line change
`@@ -69,7 +69,7 @@ def sample_tokens(tokenizer: PreTrainedTokenizerBase, length: int) -> list[int]:`
`69`	`69`
`70`	`70`	`# Remove the special tokens.`
`71`	`71`	`return random.choices(`
`72`		`- [v for k, v in vocab.items() if k not in all_special_ids],`
	`72`	`+ [v for v in vocab.values() if v not in all_special_ids],`
`73`	`73`	`k=length,`
`74`	`74`	`)`
`75`	`75`