Skip to main content
Diagnose ML training or deployment failures using knowledge base evidence. Returns diagnosis, fix steps, and prevention advice.

​
When to Use

  • Error diagnosis: Root cause analysis of errors, crashes, or unexpected behavior
  • Config issues: Debugging configuration problems or dependency conflicts
  • Unexpected behavior: Understanding why a framework behaves differently than expected

​
Parameters

ParameterRequiredTypeDescription
symptomsYesstringDescription of the failure symptoms
logsYesstringRelevant log output or error messages

​
Example

Tool: diagnose_failure
Symptoms: "QLoRA training OOM on A100 40GB with 7B model, batch size 4"
Logs: "RuntimeError: CUDA out of memory. Tried to allocate 2.00 GiB..."
Returns: Diagnosis (likely cause), fix steps (e.g., enable gradient checkpointing, reduce batch size), and prevention advice.