[Bug]: Hermes tool call parser: pops empty list

[RobertFischer]

Your current environment

4xH100-80GiB

CLI Args:

        # Inference
        - --model
        - Qwen/Qwen3-235B-A22B-FP8
        - --gpu-memory-utilization
        - "0.90"
        - --disable-custom-all-reduce

        - --rope-scaling.rope_type
        - "yarn"
        - --rope-scaling.factor
        - 4
        - --rope-scaling.original_max_position_embeddings
        - 32768
        - --max-model-len
        - "131072"
        - --tensor-parallel-size
        - "4"

        # Function calling
        - --enable-auto-tool-choice
        - --tool-call-parser
        - hermes

        # Server
        - --host
        - "0.0.0.0"
        - --disable-log-requests

🐛 Describe the bug

Pops from an empty list in the Hermes tool call parser.

IndexError: pop from empty list
  File "/usr/local/lib/python3.12/dist-packages/partial_json_parser/core/myelin.py", line 50, in fix_fast
    _i, _char = stack.pop()

  File "/usr/local/lib/python3.12/dist-packages/partial_json_parser/core/api.py", line 22, in ensure_json
    head, tail = fix_fast(json_string, allow_partial)

  File "/usr/local/lib/python3.12/dist-packages/partial_json_parser/core/api.py", line 15, in parse_json
    return parser(ensure_json(json_string, allow_partial, use_fast_fix))

  File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/tool_parsers/hermes_tool_parser.py", line 241, in extract_tool_calls
    current_tool_call = partial_json_parser.loads(...)

Unfortunately, I don't have the client request for you.

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.

[RobertFischer]

Looks like a bug in partial_json_parser....unless we're somehow calling it in violation of its contract. (In which case, it should throw one of the errors vLLM is already catching.)

[hmellor]

Hi @RobertFischer does #20474 solve this issue for you?