biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Simplify (and fix) passing of guided decoding backend options (#17008)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-04-29 20:02:23 +01:00
committed by GitHub
parent 2fa2a50bf9
commit a6977dbd15
17 changed files with 309 additions and 217 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/v1/structured_output/__init__.py
View File

@@ -45,17 +45,17 @@ class StructuredOutputManager:
# NOTE: We only support a single backend. We do NOT support different
# backends on a per-request basis in V1 (for now, anyway...).
if self.backend is None:
backend_name = request.sampling_params.guided_decoding.backend_name
if backend_name == "xgrammar":
backend = request.sampling_params.guided_decoding.backend
if backend == "xgrammar":
from vllm.v1.structured_output.backend_xgrammar import (
XgrammarBackend)
self.backend = XgrammarBackend(self.vllm_config)
elif backend_name == "guidance":
elif backend == "guidance":
self.backend = GuidanceBackend(self.vllm_config)
else:
raise ValueError(
f"Unsupported structured output backend: {backend_name}")
f"Unsupported structured output backend: {backend}")
grammar = self.executor.submit(self._async_create_grammar, request)
request.structured_output_request.grammar = grammar # type: ignore[assignment]