Issue Station
Sign In
Tools hereToolbox

vllm - 💡(How to fix) Fix [Bug]: DeepSeek V4 intermittently leaks DSML fragments in auto + streaming mode, causing unstable tool-call parsing [1 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
vllm-project/vllm#40800•Fetched 2026-04-25 06:04:00
View on GitHub
Comments
0
Participants
1
Timeline
2
Reactions
0
Author
WindswithyouWindswithyou
Participants
WindswithyouWindswithyou
Timeline (top)
closed ×1labeled ×1

Root Cause

Initial Root Cause Hypothesis

  1. Parser emits plain text too early before confirming whether current chunk tail is part of a tool-start marker.
  2. DeepSeek DSML start tags are long and chunk-splitting makes this edge case likely.
  3. required follows a different streaming branch, which explains the behavioral difference.

Fix Action

Fix / Workaround

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

Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Address sizes: 46 bits physical, 57 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 8558P CPU family: 6 Model: 207 Thread(s) per core: 2 Core(s) per socket: 48 Socket(s): 2 Stepping: 2 Frequency boost: enabled CPU max MHz: 2701.0000 CPU min MHz: 800.0000 BogoMIPS: 5400.00 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cat_l2 cdp_l3 invpcid_single intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities Virtualization: VT-x L1d cache: 4.5 MiB (96 instances) L1i cache: 3 MiB (96 instances) L2 cache: 192 MiB (96 instances) L3 cache: 520 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 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 and seccomp Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS; IBPB conditional; RSB filling; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S Vulnerability Srbds: Not affected Vulnerability Tsx async abort: Not affected

RAW_BUFFERClick to expand / collapse

Your current environment

============================== System Info

OS : Ubuntu 22.04.5 LTS (x86_64) GCC version : (Ubuntu 11.4.0-1ubuntu1~22.04.3) 11.4.0 Clang version : Could not collect CMake version : Could not collect Libc version : glibc-2.35

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

PyTorch version : 2.11.0+cu129 Is debug build : False CUDA used to build PyTorch : 12.9 ROCM used to build PyTorch : N/A XPU used to build PyTorch : N/A

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

Python version : 3.12.13 (main, Mar 4 2026, 09:23:07) [GCC 11.4.0] (64-bit runtime) Python platform : Linux-5.15.0-119-generic-x86_64-with-glibc2.35

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

Is CUDA available : True CUDA runtime version : 12.9.86 CUDA_MODULE_LOADING set to : GPU models and configuration : GPU 0: NVIDIA H20-3e GPU 1: NVIDIA H20-3e GPU 2: NVIDIA H20-3e GPU 3: NVIDIA H20-3e GPU 4: NVIDIA H20-3e GPU 5: NVIDIA H20-3e GPU 6: NVIDIA H20-3e GPU 7: NVIDIA H20-3e

Nvidia driver version : 575.57.08 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, 57 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 8558P CPU family: 6 Model: 207 Thread(s) per core: 2 Core(s) per socket: 48 Socket(s): 2 Stepping: 2 Frequency boost: enabled CPU max MHz: 2701.0000 CPU min MHz: 800.0000 BogoMIPS: 5400.00 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cat_l2 cdp_l3 invpcid_single intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities Virtualization: VT-x L1d cache: 4.5 MiB (96 instances) L1i cache: 3 MiB (96 instances) L2 cache: 192 MiB (96 instances) L3 cache: 520 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 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 and seccomp Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS; IBPB conditional; RSB filling; 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==0.6.8.post1 [pip3] numpy==2.2.6 [pip3] nvidia-cublas-cu12==12.9.1.4 [pip3] nvidia-cuda-cupti-cu12==12.9.79 [pip3] nvidia-cuda-nvrtc-cu12==12.9.86 [pip3] nvidia-cuda-runtime-cu12==12.9.79 [pip3] nvidia-cudnn-cu12==9.17.1.4 [pip3] nvidia-cudnn-frontend==1.18.0 [pip3] nvidia-cufft-cu12==11.4.1.4 [pip3] nvidia-cufile-cu12==1.14.1.1 [pip3] nvidia-curand-cu12==10.3.10.19 [pip3] nvidia-cusolver-cu12==11.7.5.82 [pip3] nvidia-cusparse-cu12==12.5.10.65 [pip3] nvidia-cusparselt-cu12==0.7.1 [pip3] nvidia-cutlass-dsl==4.4.2 [pip3] nvidia-cutlass-dsl-libs-base==4.4.2 [pip3] nvidia-ml-py==13.595.45 [pip3] nvidia-nccl-cu12==2.28.9 [pip3] nvidia-nvjitlink-cu12==12.9.86 [pip3] nvidia-nvshmem-cu12==3.4.5 [pip3] nvidia-nvtx-cu12==12.9.79 [pip3] pyzmq==27.1.0 [pip3] torch==2.11.0+cu129 [pip3] torch_c_dlpack_ext==0.1.5 [pip3] torchaudio==2.11.0+cu129 [pip3] torchvision==0.26.0+cu129 [pip3] transformers==4.57.6 [pip3] triton==3.6.0 [conda] Could not collect

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

ROCM Version : Could not collect vLLM Version : 0.1.dev15830+g8d599d76a (git sha: 8d599d76a) vLLM Build Flags: CUDA Archs: 7.0 7.5 8.0 8.9 9.0 10.0 12.0; ROCm: Disabled; XPU: Disabled GPU Topology: GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7 NIC0 NIC1 NIC2 NIC3 NIC4 NIC5 CPU Affinity NUMA Affinity GPU NUMA ID GPU0 X NV18 NV18 NV18 NV18 NV18 NV18 NV18 PIX PIX NODE NODE NODE SYS 0-47,96-143 0 N/A GPU1 NV18 X NV18 NV18 NV18 NV18 NV18 NV18 PIX PIX NODE NODE NODE SYS 0-47,96-143 0 N/A GPU2 NV18 NV18 X NV18 NV18 NV18 NV18 NV18 NODE NODE NODE NODE NODE SYS 0-47,96-143 0 N/A GPU3 NV18 NV18 NV18 X NV18 NV18 NV18 NV18 NODE NODE NODE NODE NODE SYS 0-47,96-143 0 N/A GPU4 NV18 NV18 NV18 NV18 X NV18 NV18 NV18 SYS SYS SYS SYS SYS NODE 48-95,144-191 1 N/A GPU5 NV18 NV18 NV18 NV18 NV18 X NV18 NV18 SYS SYS SYS SYS SYS NODE 48-95,144-191 1 N/A GPU6 NV18 NV18 NV18 NV18 NV18 NV18 X NV18 SYS SYS SYS SYS SYS PIX 48-95,144-191 1 N/A GPU7 NV18 NV18 NV18 NV18 NV18 NV18 NV18 X SYS SYS SYS SYS SYS PIX 48-95,144-191 1 N/A NIC0 PIX PIX NODE NODE SYS SYS SYS SYS X PIX NODE NODE NODE SYS
NIC1 PIX PIX NODE NODE SYS SYS SYS SYS PIX X NODE NODE NODE SYS
NIC2 NODE NODE NODE NODE SYS SYS SYS SYS NODE NODE X PIX NODE SYS
NIC3 NODE NODE NODE NODE SYS SYS SYS SYS NODE NODE PIX X NODE SYS
NIC4 NODE NODE NODE NODE SYS SYS SYS SYS NODE NODE NODE NODE X SYS
NIC5 SYS SYS SYS SYS NODE NODE PIX PIX SYS SYS SYS SYS SYS X

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

NIC Legend:

NIC0: mlx5_0 NIC1: mlx5_1 NIC2: mlx5_2 NIC3: mlx5_3 NIC4: mlx5_4 NIC5: mlx5_5

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

NVIDIA_VISIBLE_DEVICES=all NVIDIA_REQUIRE_CUDA=cuda>=12.9 brand=unknown,driver>=535,driver<536 brand=grid,driver>=535,driver<536 brand=tesla,driver>=535,driver<536 brand=nvidia,driver>=535,driver<536 brand=quadro,driver>=535,driver<536 brand=quadrortx,driver>=535,driver<536 brand=nvidiartx,driver>=535,driver<536 brand=vapps,driver>=535,driver<536 brand=vpc,driver>=535,driver<536 brand=vcs,driver>=535,driver<536 brand=vws,driver>=535,driver<536 brand=cloudgaming,driver>=535,driver<536 brand=unknown,driver>=550,driver<551 brand=grid,driver>=550,driver<551 brand=tesla,driver>=550,driver<551 brand=nvidia,driver>=550,driver<551 brand=quadro,driver>=550,driver<551 brand=quadrortx,driver>=550,driver<551 brand=nvidiartx,driver>=550,driver<551 brand=vapps,driver>=550,driver<551 brand=vpc,driver>=550,driver<551 brand=vcs,driver>=550,driver<551 brand=vws,driver>=550,driver<551 brand=cloudgaming,driver>=550,driver<551 brand=unknown,driver>=560,driver<561 brand=grid,driver>=560,driver<561 brand=tesla,driver>=560,driver<561 brand=nvidia,driver>=560,driver<561 brand=quadro,driver>=560,driver<561 brand=quadrortx,driver>=560,driver<561 brand=nvidiartx,driver>=560,driver<561 brand=vapps,driver>=560,driver<561 brand=vpc,driver>=560,driver<561 brand=vcs,driver>=560,driver<561 brand=vws,driver>=560,driver<561 brand=cloudgaming,driver>=560,driver<561 brand=unknown,driver>=565,driver<566 brand=grid,driver>=565,driver<566 brand=tesla,driver>=565,driver<566 brand=nvidia,driver>=565,driver<566 brand=quadro,driver>=565,driver<566 brand=quadrortx,driver>=565,driver<566 brand=nvidiartx,driver>=565,driver<566 brand=vapps,driver>=565,driver<566 brand=vpc,driver>=565,driver<566 brand=vcs,driver>=565,driver<566 brand=vws,driver>=565,driver<566 brand=cloudgaming,driver>=565,driver<566 brand=unknown,driver>=570,driver<571 brand=grid,driver>=570,driver<571 brand=tesla,driver>=570,driver<571 brand=nvidia,driver>=570,driver<571 brand=quadro,driver>=570,driver<571 brand=quadrortx,driver>=570,driver<571 brand=nvidiartx,driver>=570,driver<571 brand=vapps,driver>=570,driver<571 brand=vpc,driver>=570,driver<571 brand=vcs,driver>=570,driver<571 brand=vws,driver>=570,driver<571 brand=cloudgaming,driver>=570,driver<571 TORCH_CUDA_ARCH_LIST=7.0 7.5 8.0 8.9 9.0 10.0 12.0 NVIDIA_DRIVER_CAPABILITIES=compute,utility VLLM_USAGE_SOURCE=production-docker-image CUDA_VERSION=12.9.1 VLLM_ENABLE_CUDA_COMPATIBILITY=0 LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/cuda/lib64:/usr/local/cuda/lib64 PYTORCH_NVML_BASED_CUDA_CHECK=1 TORCHINDUCTOR_COMPILE_THREADS=1 TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_root

🐛 Describe the bug

Title: DeepSeek V4 intermittently leaks DSML fragments in auto + streaming mode, causing unstable tool-call parsing

Description vllm docker image: [deepseekv4-cu129] When serving DeepSeek V4 through vLLM OpenAI-compatible APIs, tool calling behaves inconsistently across modes:

  1. tool_choice=required is generally stable and returns structured tool_calls.
  2. tool_choice=auto with stream=true can intermittently leak DSML marker fragments into normal content.
  3. After one leaked turn, subsequent turns are more likely to fail (conversation-state contamination risk).

Impact

  1. Upstream agents may treat DSML fragments as normal assistant text.
  2. Tool-state machines can desynchronize (tool branch expected, text branch taken).
  3. The issue is intermittent in streaming mode, making it hard to detect and debug.

Reproduction Conditions

  1. Model: DeepSeek V4 family
  2. Server: vLLM OpenAI-compatible endpoint
  3. Server flags include:
    • --enable-auto-tool-choice
    • --tool-call-parser deepseek_v4
  4. Request parameters:
    • non-empty tools
    • tool_choice=auto
    • stream=true
  5. Repeating requests eventually shows DSML fragments in streamed content.

Observed Comparison

  1. Switching to tool_choice=required significantly reduces or removes the issue.
  2. Switching to stream=false significantly reduces or removes the issue.
  3. This strongly suggests the problem is in the auto streaming parse path.

Expected Behavior

  1. DSML markers should never be surfaced in streamed content, even when start markers span multiple chunks.
  2. Parser should consistently output structured tool_calls.
  3. One bad turn should not increase failure probability of later turns.

Actual Behavior

  1. In auto + stream, if DSML start markers are split across chunks, partial marker text may be emitted as plain content before tool mode is recognized.
  2. If clients append this leaked content into history, later turns become less stable.

Initial Root Cause Hypothesis

  1. Parser emits plain text too early before confirming whether current chunk tail is part of a tool-start marker.
  2. DeepSeek DSML start tags are long and chunk-splitting makes this edge case likely.
  3. required follows a different streaming branch, which explains the behavioral difference.

Suggested Fix Direction

  1. Add a prefix buffer before entering tool-call region.
  2. Emit only content that is guaranteed not to be part of a split start marker.
  3. As soon as a full start marker is recognized, switch to tool parsing immediately.
  4. Reset buffer/state per request to prevent cross-request contamination.

Compatibility and Performance Notes

  1. Scope: primarily affects auto + stream before tool region is confirmed.
  2. Runtime overhead: small constant-time string prefix checks per chunk.
  3. Memory overhead: tiny short-lived buffer (bounded by marker length).

Recommended Regression Tests

  1. Start marker split across chunks at multiple boundaries.
  2. Mixed plain-text prefix followed by tool call block.
  3. Multi-turn requests to verify state reset.
  4. Auto+stream vs required+stream comparison tests.

Environment

  1. Date observed: 2026-04-24
  2. Deployment: Linux server (vLLM), issue observed from agent/client side
  3. API: OpenAI-compatible chat completions

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.

extent analysis

TL;DR

The issue can be fixed by modifying the parser to buffer content before entering a tool-call region and only emit content that is guaranteed not to be part of a split start marker.

Guidance

  • Review the current implementation of the tool-call-parser in deepseek_v4 to identify where the premature emission of plain text is occurring.
  • Implement a prefix buffer to store content before confirming whether the current chunk is part of a tool-start marker.
  • Modify the parser to switch to tool parsing immediately after recognizing a full start marker.
  • Ensure the buffer and state are reset per request to prevent cross-request contamination.

Example

No specific code example can be provided without the actual implementation details of the tool-call-parser. However, the general approach would involve adding a buffer to store incoming content and checking each chunk against known start markers before emitting any content.

Notes

The suggested fix direction provided in the issue description seems plausible and should be explored further. The compatibility and performance notes suggest that the fix should have minimal overhead.

Recommendation

Apply the suggested fix direction by implementing a prefix buffer and modifying the parser to handle split start markers correctly, as this approach addresses the root cause hypothesis and has minimal expected overhead.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…
#api#ssr#retriever error#indexing error#environment variable

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING

vllm - 💡(How to fix) Fix [Bug]: DeepSeek V4 intermittently leaks DSML fragments in auto + streaming mode, causing unstable tool-call parsing [1 participants]