[Bug]: sometimes tool calling is not correctly parsed but remains in the plain content for qwen3 coder

[Co1lin]

Your current environment

The output of python collect_env.py

==============================
        System Info
==============================
OS                           : Ubuntu 22.04.5 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version                : Could not collect
CMake version                : version 4.0.3
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.7.1+cu128
Is debug build               : False
CUDA used to build PyTorch   : 12.8
ROCM used to build PyTorch   : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.11 (main, Jun  4 2025, 08:56:18) [GCC 11.4.0] (64-bit runtime)
Python platform              : Linux-6.8.0-1031-aws-x86_64-with-glibc2.35

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 12.8.93
CUDA_MODULE_LOADING set to   : LAZY
GPU models and configuration :
GPU 0: NVIDIA B200
GPU 1: NVIDIA B200
GPU 2: NVIDIA B200
GPU 3: NVIDIA B200
GPU 4: NVIDIA B200
GPU 5: NVIDIA B200
GPU 6: NVIDIA B200
GPU 7: NVIDIA B200

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 8559C
CPU family:                           6
Model:                                207
Thread(s) per core:                   2
Core(s) per socket:                   48
Socket(s):                            2
Stepping:                             2
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:                             640 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
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] numpy==2.2.6
[pip3] nvidia-cublas-cu12==12.8.3.14
[pip3] nvidia-cuda-cupti-cu12==12.8.57
[pip3] nvidia-cuda-nvrtc-cu12==12.8.61
[pip3] nvidia-cuda-runtime-cu12==12.8.57
[pip3] nvidia-cudnn-cu12==9.7.1.26
[pip3] nvidia-cufft-cu12==11.3.3.41
[pip3] nvidia-cufile-cu12==1.13.0.11
[pip3] nvidia-curand-cu12==10.3.9.55
[pip3] nvidia-cusolver-cu12==11.7.2.55
[pip3] nvidia-cusparse-cu12==12.5.7.53
[pip3] nvidia-cusparselt-cu12==0.6.3
[pip3] nvidia-ml-py==12.575.51
[pip3] nvidia-nccl-cu12==2.26.2
[pip3] nvidia-nvjitlink-cu12==12.8.61
[pip3] nvidia-nvshmem-cu12==3.3.9
[pip3] nvidia-nvtx-cu12==12.8.55
[pip3] pynvml==12.0.0
[pip3] pyzmq==27.0.0
[pip3] torch==2.7.1+cu128
[pip3] torchaudio==2.7.1+cu128
[pip3] torchvision==0.22.1+cu128
[pip3] transformers==4.54.1
[pip3] triton==3.3.1
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
Neuron SDK Version           : N/A
vLLM Version                 : 0.10.1.dev1+gbcc0a3cbe (git sha: bcc0a3cbe)
vLLM Build Flags:
  CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
        GPU0    GPU1    GPU2    GPU3    GPU4    GPU5    GPU6    GPU7    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      NV18    NV18    NV18    NV18    NV18    NV18    NV18    0-47,96-143     0               N/A
GPU1    NV18     X      NV18    NV18    NV18    NV18    NV18    NV18    0-47,96-143     0               N/A
GPU2    NV18    NV18     X      NV18    NV18    NV18    NV18    NV18    0-47,96-143     0               N/A
GPU3    NV18    NV18    NV18     X      NV18    NV18    NV18    NV18    0-47,96-143     0               N/A
GPU4    NV18    NV18    NV18    NV18     X      NV18    NV18    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    NV18    NV18     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
==============================
NVIDIA_VISIBLE_DEVICES=all
NVIDIA_REQUIRE_CUDA=cuda>=12.8 brand=unknown,driver>=470,driver<471 brand=grid,driver>=470,driver<471 brand=tesla,driver>=470,driver<471 brand=nvidia,driver>=470,driver<471 brand=quadro,driver>=470,driver<471 brand=quadrortx,driver>=470,driver<471 brand=nvidiartx,driver>=470,driver<471 brand=vapps,driver>=470,driver<471 brand=vpc,driver>=470,driver<471 brand=vcs,driver>=470,driver<471 brand=vws,driver>=470,driver<471 brand=cloudgaming,driver>=470,driver<471 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
NCCL_VERSION=2.25.1-1
NVIDIA_DRIVER_CAPABILITIES=compute,utility
NVIDIA_PRODUCT_NAME=CUDA
VLLM_USAGE_SOURCE=production-docker-image
CUDA_VERSION=12.8.1
LD_LIBRARY_PATH=/usr/local/cuda/lib64
NCCL_CUMEM_ENABLE=0
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY

🐛 Describe the bug

For some requests to qwen3 coder, I can get responses like this, where the tool calling is not correctly parsed as tool calls:

vllm/vllm-openai   latest      4d3b60090244   2 weeks ago   21.2GB

docker run --runtime nvidia --gpus all --shm-size 32g -it --rm \
    --network=host \
    --ipc=host \
    -v /ssd/share/huggingface:/root/.cache/huggingface \
    --name vllm_qwen3 \
    vllm/vllm-openai:latest \
    --model Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8 \
    --trust-remote-code \
    --host 0.0.0.0 --port 30008 \
    --tensor-parallel-size 4 --data-parallel-size 2 \
    --enable-auto-tool-choice \
    --tool-call-parser qwen3_coder \
    --enable-expert-parallel


ModelResponse(id='chatcmpl-66e83c6c9b2047519d4815483d85dbda', created=1755245218, model='Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8', object='chat.completion', system_fingerprint=None, choices=[Choices(finish_reason='stop', index=0, message=Message(content="Let me check if there might be a base.py file that's not showing up. Since we know it's being imported, let me try to grep for the implementation of BaseService.\n\n<tool_call>\n<function=GrepTool>\n<parameter=pattern>\nclass BaseService\\(\n</parameter>\n<parameter=path>\n.\n</parameter>\n</function>\n</tool_call>", role='assistant', tool_calls=None, function_call=None, provider_specific_fields={'refusal': None}), provider_specific_fields={'stop_reason': None})], usage=Usage(completion_tokens=69, prompt_tokens=9894, total_tokens=9963, completion_tokens_details=None, prompt_tokens_details=None), service_tier=None, prompt_logprobs=None, kv_transfer_params=None)

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.

[VivekMalipatel]

I have the same issue.

I am runnign Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8 on vLLM with

    vllm serve Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8 \
      --host 0.0.0.0 \
      --port 8001 \
      --served-model-name o3 \
      --dtype float16 \
      --enable-auto-tool-choice \
      --tool-call-parser qwen3_coder \
      --enable-expert-parallel \
      --enable-chunked-prefill \
      --enable-prefix-caching \
      --max-model-len 65536 \
      --trust-remote-code \
      --pipeline-parallel-size 2 \
      --gpu-memory-utilization 0.90 \
      --api-key test-key

But when I use it for a tool call, it outputs

I will use the knowledge_search_agent tool to retrieve a list of all documents in your knowledge base.
<function=knowledge_search_agent>
<parameter=prompt>
List all the documents in the user's knowledge base.
</parameter>
</function>
</tool_call>

Its calling the tool but vLLM is not able to pick parse that.

any fix for this?

[avi-nference]

I'm facing the same issue when using gptoss-20B model.

[Co1lin]

I think I also met with the issue for Qwen3-8B (not coder). It outputs the following content in the "content" field of a message. I think it is a valid tool calling.

<tool_call>\n{"name": "GrepTool", "arguments": {"pattern": "Plot\\\\.add\\\\(", "path": "seaborn/_core"}}\n</tool_call>

[killershrimp]

could you show some of the prompts + tool descriptions being passed into the engine?

[Co1lin]

Sorry I am not able to share the prompts at this time... In my understanding, given the ModelResponse, someone should be able to debug the parser.

[sempervictus]

This seems to be a problem with all Qwen3 MoE models themselves and its not specific to VLLM - Ollama and Candle-VLLM are experiencing similar things. From what i've gathered so far, because these models use markup language tags internally, they have been trained for specialized processing around <tag>/</tag> markers with the latter sometimes coming back as </tag} breaking whatever parser is receiving that. Probably some fun to be had around injections there but its a bloody mess for standardized tools and even the Qwen3-Coder agent users are experiencing similar things.

I've a gory MiTM patch for candle-vllm in testing right now which can correct some of this in the stream/response objects but the structural part of making the model aware of tool calls in its schema might take a little bit more work than that since even machinations with chat templates dont quite seem to get this thing resolved.

Curious to know if you folks come up with a reasonable solution as there's a bit more attention/momentum on this side of the house than in Rust. Will post a link to the PR if testing proves this in any way worth-while.

[mondaylord]

Same error. The test is fine as below
{"id":"chatcmpl-37f36888418342ee9a32fb6b0620bb04","object":"chat.completion","created":1755764628,"model":"qwen/qwen3-coder","choices":[{"index":0,"message":{"role":"assistant","content":null,"refusal":null,"annotations":null,"audio":null,"function_call":null,"tool_calls":[{"id":"chatcmpl-tool-692e2aabebb14e658fe48f2c4dba5887","type":"function","function":{"name":"math_calculator","arguments":"{\"operation\": \"add\", \"operand1\": 15, \"operand2\": 27}"}}],"reasoning_content":null},"logprobs":null,"finish_reason":"tool_calls","stop_reason":null}],"service_tier":null,"system_fingerprint":null,"usage":{"prompt_tokens":354,"total_tokens":402,"completion_tokens":48,"prompt_tokens_details":null},"prompt_logprobs":null,"kv_transfer_params":null}

But sometimes it just returns
<tool_call> <function=get_current_weather> <parameter=location> Boston </parameter> <parameter=unit> fahrenheit </parameter> </function> </tool_call>
Seems the parser is not suitable for some outputs?

[mondaylord]

I think this is related to streaming output, if the stream is false, then tool-call is fine, but it returns <tool_call> tag when streaming is true

[bbartels]

I did just see that the tool call parser got updated in the qwen3 repo, might come with some fixes?
https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/commit/9d90cf8fca1bf7b7acca42d3fc9ae694a2194069

[Ithanil]

I did just see that the tool call parser got updated in the qwen3 repo, might come with some fixes? https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/commit/9d90cf8fca1bf7b7acca42d3fc9ae694a2194069

parser + chat template

Definitely works, exactly how much better I can't say.

[mondaylord]

I did just see that the tool call parser got updated in the qwen3 repo, might come with some fixes? https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/commit/9d90cf8fca1bf7b7acca42d3fc9ae694a2194069

parser + chat template

Definitely works, exactly how much better I can't say.

Still failed for streaming case, here is the content and result

{
    "model": "qwen3-coder",
    "stream": true,
    "stream_options": {
      "include_usage": true
    },
    "messages": [
      {
        "role": "user",
        "content": "What is the weather like in Boston in fahrenheit?"
      }
    ],
    "tools": [
      {
        "type": "function",
        "function": {
          "name": "get_current_weather",
          "description": "Get the current weather in a given location",
          "strict": true,
          "parameters": {
            "type": "object",
            "properties": {
              "location": {
                "type": "string",
                "description": "The city and state, e.g. San Francisco, CA"
              },
              "unit": {
                "type": "string",
                "enum": [
                  "celsius",
                  "fahrenheit"
                ]
              }
            },
            "additionalProperties": false,
            "required": [
              "location",
              "unit"
            ]
          }
        }
      }
    ],
    "max_tokens": 235808,
    "temperature": 1,
    "top_p": 1,
    "frequency_penalty": 0,
    "presence_penalty": 0,
    "seed": null,
    "min_p": 0,
    "repetition_penalty": 1
  }

Result:

<tool_call>
<function=get_current_weather>
<parameter=location>
Boston
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
</tool_call>

[mondaylord]

I did just see that the tool call parser got updated in the qwen3 repo, might come with some fixes? https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/commit/9d90cf8fca1bf7b7acca42d3fc9ae694a2194069

parser + chat template
Definitely works, exactly how much better I can't say.

Still failed for streaming case, here is the content and result

{
    "model": "qwen3-coder",
    "stream": true,
    "stream_options": {
      "include_usage": true
    },
    "messages": [
      {
        "role": "user",
        "content": "What is the weather like in Boston in fahrenheit?"
      }
    ],
    "tools": [
      {
        "type": "function",
        "function": {
          "name": "get_current_weather",
          "description": "Get the current weather in a given location",
          "strict": true,
          "parameters": {
            "type": "object",
            "properties": {
              "location": {
                "type": "string",
                "description": "The city and state, e.g. San Francisco, CA"
              },
              "unit": {
                "type": "string",
                "enum": [
                  "celsius",
                  "fahrenheit"
                ]
              }
            },
            "additionalProperties": false,
            "required": [
              "location",
              "unit"
            ]
          }
        }
      }
    ],
    "max_tokens": 235808,
    "temperature": 1,
    "top_p": 1,
    "frequency_penalty": 0,
    "presence_penalty": 0,
    "seed": null,
    "min_p": 0,
    "repetition_penalty": 1
  }

Result:

<tool_call>
<function=get_current_weather>
<parameter=location>
Boston
</parameter>
<parameter=unit>
fahrenheit
</parameter>
</function>
</tool_call>

Found the reason, no reasoning-parser in deployment stage, then everything works fine

[dotmobo]

same bug for me

[mdierolf]

If there is a way to make it work fine, please explain.

Found the reason, no reasoning-parser in deployment stage, then everything works fine