vllm - 💡(How to fix) Fix BUG: ValueError too many values to unpack in internlm2_tool_parser when output contains multiple tool-call markers

What breaks

When a model using the InternLM2 tool parser produces output that contains more than one <|action_start|><|plugin|> marker (e.g. a response that repeats the tool-call header, or genuinely attempts parallel tool calls), both `extract_tool_calls` (line 205) and `extract_tool_calls_streaming` (line 80) crash with:

``` ValueError: too many values to unpack (expected 2) ```

How to trigger

```python model_output = ( "Some text" "< |action_start|><|plugin|>{"name": "tool1", "parameters": {}}<|action_end|>" "<|action_start|><|plugin|>{"name": "tool2", "parameters": {}}<|action_end|>" ) text, action = model_output.split("<|action_start|><|plugin|>") # crashes ```

Root cause

Both lines use `str.split(delimiter)` without a `maxsplit` argument. When the delimiter appears more than once, `split()` returns 3+ items, but the code unpacks into exactly 2 variables.

Fix

Change both calls to `split("<|action_start|><|plugin|>", 1)` so only the first occurrence is used as the split point.

Affected file: `vllm/tool_parsers/internlm2_tool_parser.py` lines 80 and 205.