[AdaAibaby]

Purpose

Internlm2ToolParser.extract_tool_calls_streaming splits the accumulated text on the tool-call marker with an unbounded str.split and unpacks the result into exactly two values:

new_delta = current_text[last_pos:]
text, action = new_delta.split("<|action_start|><|plugin|>")

self.position is only advanced past text that precedes the first marker, and self.current_tool_id never exceeds 0, so the if self.current_tool_id > 0 early-return never fires on the tool-call path. Once the model streams a second <|action_start|><|plugin|> block, current_text[last_pos:] contains two markers, split returns three parts, and the unpack raises:

ValueError: too many values to unpack (expected 2)

This line runs before the method's try/except (which starts a few lines later), so the exception propagates and aborts the streaming request rather than being swallowed.

Trigger

A model emitting two consecutive tool-call blocks, e.g. accumulated text:

<|action_start|><|plugin|>{"name": "f", "parameters": {}}<|action_end|><|action_start|><|plugin|>{"name": "g", "parameters": {}}<|action_end|>

InternLM2 is documented as not supporting parallel tool calls, but that is a limitation, not a guard — nothing constrains the decoder from emitting a second block, and a misprompted/hallucinating model realistically can.

Fix

Split only on the first marker with str.partition:

text, _, action = new_delta.partition("<|action_start|><|plugin|>")

The subsequent action.split("<|action_end|>")[0] already isolates the first tool call, so single-tool-call behavior is unchanged. Verified that for single-block input the extracted (text, action) is identical to the old code.

Test

Added test_streaming_two_action_blocks_no_unpack_error in tests/tool_parsers/test_internlm2_tool_parser.py, which streams two action blocks and asserts the extractor does not raise. It fails on main (ValueError) and passes with this change.

pytest tests/tool_parsers/test_internlm2_tool_parser.py -k two_action_blocks

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @AdaAibaby.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork