20231088/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[V1][Structured Output] Minor modification to _validate_structured_output() (#16748)

Browse Source 
Signed-off-by: shen-shanshan <467638484@qq.com>
This commit is contained in:
Shanshan Shen 2025-04-18 13:12:54 +08:00 committed by GitHub
parent 7a4a5de729
commit 30ed81b7ca
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 10 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
vllm/v1/engine/processor.py
View File

@ -149,6 +149,7 @@ class Processor:
"xgrammar", "xgrammar:disable-any-whitespace", "guidance", "xgrammar", "xgrammar:disable-any-whitespace", "guidance",
"guidance:disable-any-whitespace", "auto" "guidance:disable-any-whitespace", "auto"
] ]
engine_level_backend = self.decoding_config.guided_decoding_backend engine_level_backend = self.decoding_config.guided_decoding_backend
if engine_level_backend not in supported_backends: if engine_level_backend not in supported_backends:
raise ValueError(f"Only {supported_backends} structured output is " raise ValueError(f"Only {supported_backends} structured output is "
@ -169,8 +170,15 @@ class Processor:
if engine_level_backend.startswith("xgrammar"): if engine_level_backend.startswith("xgrammar"):
# xgrammar with no fallback # xgrammar with no fallback
validate_xgrammar_grammar(params) validate_xgrammar_grammar(params)
params.guided_decoding.backend = engine_level_backend elif engine_level_backend.startswith("guidance"):
elif engine_level_backend == "auto": # TODO: ideally we would have the LLTokenizer here as Lark syntax
# allows <|special_token|> and similar, see
# https://github.com/guidance-ai/llguidance/blob/main/docs/syntax.md#special-tokens
# Without tokenizer these are disallowed in grammars.
validate_guidance_grammar(params, tokenizer=None)
else:
# NOTE: engine_level_backend must be "auto" here, because we have
# checked supported_backends above.
# "auto" is an opt-in to opinionated behavior where we try to # "auto" is an opt-in to opinionated behavior where we try to
# choose a backend based on request contents. This is not the # choose a backend based on request contents. This is not the
# default as it is less predictable and subject to change # default as it is less predictable and subject to change
@ -183,14 +191,6 @@ class Processor:
# are not supported in xgrammar. Fall back to guidance. # are not supported in xgrammar. Fall back to guidance.
params.guided_decoding.backend = "guidance" params.guided_decoding.backend = "guidance"
if engine_level_backend.startswith("guidance"):
# TODO ideally we would have the LLTokenizer here as Lark syntax
# allows <|special_token|> and similar, see
# https://github.com/guidance-ai/llguidance/blob/main/docs/syntax.md#special-tokens
# Without tokenizer these are disallowed in grammars.
validate_guidance_grammar(params, tokenizer=None)
params.guided_decoding.backend = engine_level_backend
def process_inputs( def process_inputs(
self, self,
request_id: str, request_id: str,