[Bug]: GPT-OSS Tool Calls Fail in Stream Mode

[Yuyz0112]

Your current environment

The output of python collect_env.py

Collecting environment information...                                                                   ==============================                                                                                  System Info                                                                                     ==============================
OS                           : Ubuntu 22.04.5 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04.2) 11.4.0
Clang version                : Could not collect                                                        CMake version                : version 3.22.1
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.8.0+cu129
Is debug build               : False
CUDA used to build PyTorch   : 12.9
ROCM used to build PyTorch   : N/A

==============================
      Python Environment
==============================
Python version               : 3.10.12 (main, Aug 15 2025, 14:32:43) [GCC 11.4.0] (64-bit runtime)
Python platform              : Linux-5.15.0-153-generic-x86_64-with-glibc2.35

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 12.4.131
CUDA_MODULE_LOADING set to   : LAZY
GPU models and configuration :                                                                          GPU 0: NVIDIA L20
GPU 1: NVIDIA L20
GPU 2: NVIDIA L20
GPU 3: NVIDIA L20

Nvidia driver version        : 550.163.01
cuDNN version                : Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.9.7
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.9.7
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.9.7
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.9.7
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.9.7
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.9.7
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.9.7
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):                                  32
On-line CPU(s) list:                     0-31
Vendor ID:                               GenuineIntel
Model name:                              Intel(R) Xeon(R) Silver 4309Y CPU @ 2.80GHz
CPU family:                              6
Model:                                   106
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               2
Stepping:                                6
BogoMIPS:                                5600.00
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 pni pclmulqdq dtes64 monitor 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 invpcid_single 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 wbnoinvd dtherm ida arat pln pts hwp_epp avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid fsrm md_clear pconfig flush_l1d arch_capabilities
Virtualization:                          VT-x
L1d cache:                               768 KiB (16 instances)
L1i cache:                               512 KiB (16 instances)
L2 cache:                                20 MiB (16 instances)
L3 cache:                                24 MiB (2 instances)
NUMA node(s):                            2
NUMA node0 CPU(s):                       0-7,16-23
NUMA node1 CPU(s):                       8-15,24-31
Vulnerability Gather data sampling:      Mitigation; Microcode
Vulnerability Indirect target selection: Mitigation; Aligned branch/return thunks
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Mitigation; Clear CPU buffers; SMT vulnerable
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 disabled; PBRSB-eIBRS SW sequence; BHI SW loop, KVM SW loop
Vulnerability Srbds:                     Not affected
Vulnerability Tsx async abort:           Not affected

==============================
Versions of relevant libraries
==============================
[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==9.10.2.21
[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-ml-py==13.580.65
[pip3] nvidia-nccl-cu12==2.27.3
[pip3] nvidia-nvjitlink-cu12==12.9.86
[pip3] nvidia-nvtx-cu12==12.9.79
[pip3] pyzmq==27.1.0
[pip3] torch==2.8.0+cu129
[pip3] torchaudio==2.8.0+cu129
[pip3] torchvision==0.23.0+cu129
[pip3] transformers==4.56.2
[pip3] triton==3.4.0
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
vLLM Version                 : 0.10.2
vLLM Build Flags:
  CUDA Archs: Not Set; ROCm: Disabled
GPU Topology:
        GPU0    GPU1    GPU2    GPU3    NIC0    NIC1    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      NODE    SYS     SYS     SYS     SYS     0-7,16-23       0               N/A
GPU1    NODE     X      SYS     SYS     SYS     SYS     0-7,16-23       0               N/A
GPU2    SYS     SYS      X      NODE    NODE    NODE    8-15,24-31      1               N/A
GPU3    SYS     SYS     NODE     X      NODE    NODE    8-15,24-31      1               N/A
NIC0    SYS     SYS     NODE    NODE     X      PIX
NIC1    SYS     SYS     NODE    NODE    PIX      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

==============================
     Environment Variables
==============================
LD_LIBRARY_PATH=/usr/local/cuda/lib64:
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY

🐛 Describe the bug

Tool calls work correctly in non-stream mode but fail in stream mode with GPT-OSS-20B. In stream mode: Tool call tokens are incorrectly identified as reasoning content.

Root Cause

The GPT-OSS-20B model does not consistently output tool calls with the commentary channel. Instead, it uses the analysis channel with a recipient specified via to=functions.XXX(mabye a model bug?):

<|channel|>analysis to=functions.grep code<|message|>{...}<|call|>

The stream mode logic in serving_chat.py:697-745 checks channel type before checking recipient, causing tool calls in the analysis channel to be misidentified as reasoning content:

# Current (incorrect) logic:
elif cur_channel == "analysis":
    # Treats as reasoning, even if cur_recipient = "functions.grep"
    delta_message = DeltaMessage(reasoning_content=delta_text)
elif cur_channel == "commentary" and cur_recipient.startswith("functions."):
    # Never reaches here for analysis channel

In contrast, non-stream mode correctly identifies tool calls by checking recipient only, regardless of channel:

# openai_tool_parser.py:43
if msg.recipient and msg.recipient.startswith("functions."):
    tool_calls.append(...)  # Correct

Example Token Sequences

Non-stream (73 tokens, works correctly):

<|channel|>analysis<|message|>User asks: "check the docs, Is it ok to add VMs without virtual volumes to a snapshot plan?" Need to search guides for snapshot plan and VMs without virtual volumes. Use grep.<|end|><|start|>assistant<|channel|>analysis to=functions.grep code<|message|>{"pattern":"snapshot plan","glob":"**/*.md","output_mode":"content"}<|call|>

Stream (66 tokens, fails):

<|channel|>analysis<|message|>We need to search guides for the question. Likely in guides about Snapshot plans. Use grep.<|end|><|start|>assistant<|channel|>analysis to=functions.grep code<|message|>{
  "pattern": "snapshot plan",
  "glob": "guides/**/*.md",
  "output_mode": "files_with_matches"
}<|call|>

Both use <|channel|>analysis to=functions.grep for tool calls, but stream mode mishandles this format.

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.

[Yuyz0112]

A reference from OpenAI's harmony repo: openai/harmony#40 (comment)

The recipient can appear in either part of the header. It's mentioned in the guide as The recipient might be defined in the role or channel section of the header.. The parser can handle either order.

[bbrowning]

I believe this is fixed by #24768 which merged to main 8 hours ago. The reason the model was outputting to the analysis channel instead of the commentary channel was because we were not properly activating the commentary channel in the our system prompt.

[Yuyz0112]

@bbrowning Thanks for the update! The fix patch looks reasonable to me, going to check the latest build.

[alew3]

@bbrowning using the nightly docker seems to have fixed tool calling with streaming.

[bbrowning]

Great, thanks for the feedback! I cannot actually close this issue myself, but it sounds like this is resolved for the general cases at least.

[bbrowning]

@@yeqcharlotte This gpt-oss issue appears fixed, and can be closed 😄