分类: AI 工具排错

Eval bug: Muse Glimmer “The model produced output that does not match the expected peg-native format” when running tool calls

Eval bug: Muse Glimmer "The model produced output that does not match the expected peg-native format" when running tool calls

该报错是 llama-server 在使用 Muse Glimmer 模型进行工具调用(tool calls)时,模型偶尔生成不符合 peg-native 格式的输出导致的解析失败,约每 5 次工具调用出现 1 次,属于间歇性错误而非必然崩溃。优先排查工具名称的命名方式,因为 Issue 中用户通过