tool_call_malformed_jsonTier 1 · 70% confidence

ai-agents-tool-call-malformed--when-using-vllm-with-glm-5-or-glm-4-7-models-and-m-f78551f9

agent: ai_agents

When does this happen?

IF When using vLLM with GLM-5 or GLM-4.7 models and MTP (Multi-Token Prediction / speculative decoding) enabled, tool call JSON output is malformed, e.g., missing closing braces or truncated arguments, causing parse failures.

How others solved it

THEN Disable MTP decoding or downgrade vLLM to version 0.15.0 where tool calling works correctly. If using NVIDIA NIM, check the underlying vLLM version and apply the same workaround.

Related patterns

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics