Your current environment

vLLM 0.11.x , on H200s

--2025-10-20 22:19:16-- https://raw.githubusercontent.com/vllm-project/vllm/main/vllm/collect_env.py
Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.111.133, 185.199.108.133, 185.199.109.133, ...
Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.111.133|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 28050 (27K) [text/plain]
Saving to: ‘collect_env.py’

collect_env. 0% 0 --.-KB/s collect_env. 100% 27.39K --.-KB/s in 0s

2025-10-20 22:19:17 (185 MB/s) - ‘collect_env.py’ saved [28050/28050]

Collecting environment information...

    System Info

==============================
OS : Ubuntu 24.04.2 LTS (x86_64)
GCC version : (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
Clang version : Could not collect
CMake version : version 4.1.0
Libc version : glibc-2.39

==============================
PyTorch Info

PyTorch version : 2.8.0+cu128
Is debug build : False
CUDA used to build PyTorch : 12.8
ROCM used to build PyTorch : N/A

==============================
Python Environment

Python version : 3.12.3 (main, Aug 14 2025, 17:47:21) [GCC 13.3.0] (64-bit runtime)
Python platform : Linux-6.14.0-1014-aws-x86_64-with-glibc2.39

==============================
CUDA / GPU Info

Is CUDA available : True
CUDA runtime version : 12.9.86
CUDA_MODULE_LOADING set to : LAZY
GPU models and configuration :
GPU 0: NVIDIA H200
GPU 1: NVIDIA H200
GPU 2: NVIDIA H200
GPU 3: NVIDIA H200
GPU 4: NVIDIA H200
GPU 5: NVIDIA H200
GPU 6: NVIDIA H200
GPU 7: NVIDIA H200

Nvidia driver version : 570.172.08
cuDNN version : Could not collect
HIP runtime version : N/A
MIOpen runtime version : N/A
Is XNNPACK available : True

==============================
CPU Info

Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 46 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 192
On-line CPU(s) list: 0-191
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) Platinum 8488C
CPU family: 6
Model: 143
Thread(s) per core: 2
Core(s) per socket: 48
Socket(s): 2
Stepping: 8
BogoMIPS: 4800.00
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd ida arat avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq rdpid cldemote movdiri movdir64b md_clear serialize amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 4.5 MiB (96 instances)
L1i cache: 3 MiB (96 instances)
L2 cache: 192 MiB (96 instances)
L3 cache: 210 MiB (2 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0-47,96-143
NUMA node1 CPU(s): 48-95,144-191
Vulnerability Gather data sampling: Not affected
Vulnerability Ghostwrite: Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected

==============================
Versions of relevant libraries

[pip3] flashinfer-python==0.4.1
[pip3] numpy==2.2.6
[pip3] nvidia-cublas-cu12==12.8.4.1
[pip3] nvidia-cuda-cupti-cu12==12.8.90
[pip3] nvidia-cuda-nvrtc-cu12==12.8.93
[pip3] nvidia-cuda-runtime-cu12==12.8.90
[pip3] nvidia-cudnn-cu12==9.10.2.21
[pip3] nvidia-cudnn-frontend==1.13.0
[pip3] nvidia-cufft-cu12==11.3.3.83
[pip3] nvidia-cufile-cu12==1.13.1.3
[pip3] nvidia-curand-cu12==10.3.9.90
[pip3] nvidia-cusolver-cu12==11.7.3.90
[pip3] nvidia-cusparse-cu12==12.5.8.93
[pip3] nvidia-cusparselt-cu12==0.7.1
[pip3] nvidia-cutlass-dsl==4.2.1
[pip3] nvidia-ml-py==12.575.51
[pip3] nvidia-nccl-cu12==2.27.3
[pip3] nvidia-nvjitlink-cu12==12.8.93
[pip3] nvidia-nvshmem-cu12==3.3.9
[pip3] nvidia-nvtx-cu12==12.8.90
[pip3] pynvml==12.0.0
[pip3] pytorch-triton==3.4.0+git11ec6354
[pip3] pyzmq==27.1.0
[pip3] torch==2.8.0
[pip3] torchaudio==2.8.0
[pip3] torchvision==0.23.0
[pip3] transformers==4.56.1
[pip3] triton==3.4.0
[conda] Could not collect

==============================
vLLM Info

ROCM Version : Could not collect
vLLM Version : 0.11.1rc2.dev165+g6cce4698e (git sha: 6cce469)
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled
GPU Topology:
GPU0 GPU1 GPU2 GPU3 GPU4 GPU5GPU6 GPU7 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X NV18 NV18 NV18 NV18 NV18NV18 NV18 0-47,96-143 0 N/A
GPU1 NV18 X NV18 NV18 NV18 NV18NV18 NV18 0-47,96-143 0 N/A
GPU2 NV18 NV18 X NV18 NV18 NV18NV18 NV18 0-47,96-143 0 N/A
GPU3 NV18 NV18 NV18 X NV18 NV18NV18 NV18 0-47,96-143 0 N/A
GPU4 NV18 NV18 NV18 NV18 X NV18NV18 NV18 48-95,144-191 1 N/A
GPU5 NV18 NV18 NV18 NV18 NV18 X NV18 NV18 48-95,144-191 1 N/A
GPU6 NV18 NV18 NV18 NV18 NV18 NV18 X NV18 48-95,144-191 1 N/A
GPU7 NV18 NV18 NV18 NV18 NV18 NV18NV18 X 48-95,144-191 1 N/A

Legend:

X = Self
SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
PXB = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
PIX = Connection traversing at most a single PCIe bridge
NV# = Connection traversing a bonded set of # NVLinks

==============================
Environment Variables

LD_LIBRARY_PATH=/usr/local/cuda-12.9/lib64:/usr/local/cuda-12.8/lib:/usr/local/cuda-12.8/lib64:/usr/local/cuda-12.8:/usr/local/cuda-12.8/targets/x86_64-linux/lib/:/usr/local/cuda-12.8/extras/CUPTI/lib64:/opt/amazon/efa/lib:/opt/amazon/openmpi/lib:/opt/amazon/ofi-nccl/lib/x86_64-linux-gnu:/usr/local/lib:/usr/lib
CUDA_HOME=/usr/local/cuda-12.9
CUDA_HOME=/usr/local/cuda-12.9
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY

🐛 Describe the bug

Issue

When using Eagle speculative decoding with structured output (tool calling), vLLM crashes with AssertionError when the FSM rejects a spec token that is present in the scheduled_spec_decode_tokens list.

Full Stack Trace

[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [backend_xgrammar.py:159] Failed to advance FSM for request chatcmpl-e85b68e4c00e4fb1bbeddfd122f5b3de for tokens 497. Please file an issue.

[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795] Traceback (most recent call last):
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]   File "/home/ubuntu/vllm/vllm/v1/engine/core.py", line 786, in run_engine_core
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]     engine_core.run_busy_loop()
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]   File "/home/ubuntu/vllm/vllm/v1/engine/core.py", line 813, in run_busy_loop
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]     self._process_engine_step()
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]   File "/home/ubuntu/vllm/vllm/v1/engine/core.py", line 842, in _process_engine_step
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]     outputs, model_executed = self.step_fn()
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]                               ^^^^^^^^^^^^^^
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]   File "/home/ubuntu/vllm/vllm/v1/engine/core.py", line 320, in step
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]     scheduler_output = self.scheduler.schedule()
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]                        ^^^^^^^^^^^^^^^^^^^^^^^^^
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]   File "/home/ubuntu/vllm/vllm/v1/core/sched/scheduler.py", line 615, in schedule
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]     structured_output_request_ids, grammar_bitmask = self.get_grammar_bitmask(
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]                                                      ^^^^^^^^^^^^^^^^^^^^^^^^^
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]   File "/home/ubuntu/vllm/vllm/v1/core/sched/scheduler.py", line 897, in get_grammar_bitmask
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]     bitmask = self.structured_output_manager.grammar_bitmask(
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]   File "/home/ubuntu/vllm/vllm/v1/structured_output/__init__.py", line 263, in grammar_bitmask
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]     assert structured_output_request.grammar.accept_tokens(
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[EngineCore_DP0 pid=2404644] ERROR 10-20 15:33:04 [core.py:795] AssertionError

[APIServer pid=2404466] ERROR 10-20 15:33:04 [async_llm.py:533] AsyncLLM output_handler failed.
[APIServer pid=2404466] ERROR 10-20 15:33:04 [async_llm.py:533] Traceback (most recent call last):
[APIServer pid=2404466] ERROR 10-20 15:33:04 [async_llm.py:533]   File "/home/ubuntu/vllm/vllm/v1/engine/async_llm.py", line 487, in output_handler
[APIServer pid=2404466] ERROR 10-20 15:33:04 [async_llm.py:533]     outputs = await engine_core.get_output_async()
[APIServer pid=2404466] ERROR 10-20 15:33:04 [async_llm.py:533]               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[APIServer pid=2404466] ERROR 10-20 15:33:04 [async_llm.py:533]   File "/home/ubuntu/vllm/vllm/v1/engine/core_client.py", line 882, in get_output_async
[APIServer pid=2404466] ERROR 10-20 15:33:04 [async_llm.py:533]     raise self._format_exception(outputs) from None
[APIServer pid=2404466] ERROR 10-20 15:33:04 [async_llm.py:533] vllm.v1.engine.exceptions.EngineDeadError: EngineCore encountered an issue. See stack trace (above) for the root cause.
[APIServer pid=2404466] INFO:     127.0.0.1:48250 - "POST /v1/chat/completions HTTP/1.1" 500 Internal Server Error

Critical crash point:

File "/home/ubuntu/vllm/vllm/v1/structured_output/__init__.py", line 263, in grammar_bitmask
    assert structured_output_request.grammar.accept_tokens(
AssertionError

Error

ERROR: Failed to advance FSM for request ... for tokens XXX. Please file an issue.
AssertionError at vllm/v1/structured_output/__init__.py line 263

Potential Root Cause

Potential inconsistency in xgrammar's GrammarMatcher.rollback() that fails to reset the terminated flag when rolling back past a token that terminated the FSM. This causes the FSM to remain in a "terminated" state even after rolling back to a non-terminated state.

Debugging logs

(I added some debug statements after incorporating the defensive fallback logic below to inspect the problematic tokens)

[16:01:19] /project/cpp/grammar_matcher.cc:371: Warning: The matcher has terminated after accepting the stop token, but is trying to accept new token with id 198.

[APIServer pid=2417014] INFO 10-20 16:01:19 [metrics.py:100] SpecDecoding metrics: Mean acceptance length: 2.22, Accepted throughput: 474.87 tokens/s, Drafted throughput: 1939.16 tokens/s, Accepted: 4749 tokens, Drafted: 19393 tokens, Per-position acceptance rate: 0.577, 0.302, 0.171, 0.107, 0.064, Avg Draft acceptance rate: 24.5%

[EngineCore_DP0 pid=2417186] ERROR 10-20 16:01:20 [backend_xgrammar.py:159] Failed to advance FSM for request chatcmpl-83d721aea8a34274a568c73f7fa2040f for tokens 330. Please file an issue.

[EngineCore_DP0 pid=2417186] WARNING 10-20 16:01:20 [eagle_structured_output_patch.py:189] [FSM DEBUG] Request chatcmpl-83d721aea8a34274a568c73f7fa2040f: Token 330 REJECTED in grammar_bitmask! FSM State: terminated_before=False, terminated_after=False, processed_before=6, processed_after=6. Token Context: scheduled_tokens=[330], token_index=0/1, state_advancements_so_far=0. This indicates an FSM state mismatch in Eagle + structured output. Continuing without advancing grammar state to avoid crash.

Analysis

1. 16:01:19: xgrammar warning shows FSM terminated after accepting stop token, then tried to accept token 198
2. 16:01:20: Token 330 rejected even though FSM reports terminated=False and has processed=6 tokens

It looks like:

• FSM terminated during earlier validation (accepted stop token)
• xgrammar's rollback() was called to rewind FSM state
• Bug: rollback correctly rewound state to position 6, but failed to reset internal terminated_ flag
• Result: xgrammar internally thinks FSM is terminated, vLLM wrapper thinks it's not
• When trying to accept token 330, xgrammar's accept_token() checks terminated flag first, sees true, returns false

Reproduction

Setup:

• Model: Qwen3-Coder-30B with FP8 quantization
• Eagle speculative decoding enabled (num_speculative_tokens: 5)
• Structured output enabled (--enable-auto-tool-choice)
• Load: Multiple concurrent requests with tool calling

Trigger:

• Use Eagle + structured output under load
• Wait for FSM to hit a terminating token during validation
• Next accept_tokens() call fails on a valid token that passed earlier validation

Proposed Fix

Replace assertion with defensive conditional check in vllm/v1/structured_output/__init__.py:

# ORIGINAL (crashes):
assert structured_output_request.grammar.accept_tokens(req_id, [token])
state_advancements += 1

# POTENTIAL FIX (defensive approach):
if structured_output_request.grammar.accept_tokens(req_id, [token]):
    state_advancements += 1
else:
    # Token rejected by FSM even though it's in scheduled list
    # Root cause: xgrammar rollback bug with terminated flag
    logger.debug(
        f"Grammar rejected spec token {token} for request {req_id}. "
        "FSM state mismatch in Eagle + structured output. "
        "Continuing without advancing grammar state."
    )
    # Continue to fill bitmasks for all tokens (required by downstream code)

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.