From 7d0b3a6ba82ec8d1fd0775114e78b8812488d5b2 Mon Sep 17 00:00:00 2001 From: Russell Bryant Date: Fri, 15 Aug 2025 14:59:27 +0000 Subject: [PATCH 1/2] Bump xgrammar to 0.1.23 https://github.com/mlc-ai/xgrammar/releases/tag/v0.1.23 Signed-off-by: Russell Bryant --- requirements/common.txt | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/requirements/common.txt b/requirements/common.txt index 1a8fea0dd7d9..73a0158e22ef 100644 --- a/requirements/common.txt +++ b/requirements/common.txt @@ -24,7 +24,7 @@ outlines_core == 0.2.10 # required for outlines backend disk cache diskcache == 5.6.3 lark == 1.2.2 -xgrammar == 0.1.21; platform_machine == "x86_64" or platform_machine == "aarch64" or platform_machine == "arm64" +xgrammar == 0.1.23; platform_machine == "x86_64" or platform_machine == "aarch64" or platform_machine == "arm64" typing_extensions >= 4.10 filelock >= 3.16.1 # need to contain https://github.com/tox-dev/filelock/pull/317 partial-json-parser # used for parsing partial JSON outputs From 3801b7977ef7b14fb0228e7b97e36b2c8ed816da Mon Sep 17 00:00:00 2001 From: Russell Bryant Date: Fri, 15 Aug 2025 15:03:33 +0000 Subject: [PATCH 2/2] Revert "[V1] Resolve failed concurrent structured output requests (#19565)" This reverts commit c57bb199b3eff82b4bdd5ffb089502936a5b0c9a. Signed-off-by: Russell Bryant --- vllm/v1/worker/gpu_model_runner.py | 9 +-------- 1 file changed, 1 insertion(+), 8 deletions(-) diff --git a/vllm/v1/worker/gpu_model_runner.py b/vllm/v1/worker/gpu_model_runner.py index 9460d91c5832..01b479a68564 100644 --- a/vllm/v1/worker/gpu_model_runner.py +++ b/vllm/v1/worker/gpu_model_runner.py @@ -87,15 +87,11 @@ if TYPE_CHECKING: import xgrammar as xgr - import xgrammar.kernels.apply_token_bitmask_inplace_torch_compile as xgr_torch_compile # noqa: E501 from vllm.model_executor.model_loader.tensorizer import TensorizerConfig from vllm.v1.core.sched.output import SchedulerOutput else: xgr = LazyLoader("xgr", globals(), "xgrammar") - xgr_torch_compile = LazyLoader( - "xgr_torch_compile", globals(), - "xgrammar.kernels.apply_token_bitmask_inplace_torch_compile") logger = init_logger(__name__) @@ -1363,10 +1359,7 @@ def apply_grammar_bitmask( # so we receive it in that format. grammar_bitmask = torch.from_numpy(grammar_bitmask).contiguous() - # Force use of the torch.compile implementation from xgrammar to work - # around issues with the Triton kernel in concurrent structured output - # scenarios. See PR #19565 and issues #19493, #18376 for details. - xgr_torch_compile.apply_token_bitmask_inplace_torch_compile( + xgr.apply_token_bitmask_inplace( logits, grammar_bitmask.to(self.device, non_blocking=True), indices=out_indices if not skip_out_indices else None,