[Bugfix][Frontend] Seed derender detokenization from the prompt (#59046)

Signed-off-by: Martin Hickey <martin.hickey@ie.ibm.com>
This commit is contained in:
Martin Hickey
2026-10-02 03:17:56 +00:00
committed by GitHub
parent 8faacd707f
commit fcfacc3499
9 changed files with 760 additions and 163 deletions
+2 -2
View File
@@ -63,7 +63,7 @@ _CACHED_MARKER_KEY = "_vllm_space_marker_cache"
_NOT_CACHED = "__not_computed__"
def _get_leading_space_marker(tokenizer: TokenizerLike) -> str | None:
def get_leading_space_marker(tokenizer: TokenizerLike) -> str | None:
"""Read the space marker from the tokenizer's pre_tokenizer config.
Only Metaspace pre_tokenizers (used by SentencePiece-based models like
@@ -160,7 +160,7 @@ def convert_ids_list_to_tokens(
"""
if not token_ids:
return []
marker = _get_leading_space_marker(tokenizer)
marker = get_leading_space_marker(tokenizer)
if marker is None:
return [tokenizer.decode([tid]) or "" for tid in token_ids]
raw_tokens = tokenizer.convert_ids_to_tokens(token_ids)