In brief: Troubleshooting begins by identifying the failed layer: Python environment, package import, authentication, unsupported model ID or actual inference capacity.
Package and Python problems
Run python -m pip show esm in the active virtual environment and verify imports. Multiple Python installations can make pip install to the wrong environment. Match your platform and PyTorch setup to official requirements.
Authentication and access problems
Verify that ESM_API_KEY is set and the account has access to the requested Biohub model. Avoid writing real tokens into logs. Keep the service endpoint exactly as documented for the installed SDK version.
GPU memory and throughput
Large model weights plus activations can exhaust memory. Diagnose with device telemetry and a short, documented test. Switch to an appropriate smaller checkpoint or managed inference rather than assuming a generic --batch_size CLI flag exists.
Different model classes and outputs
ESM3, ESMC and ESMFold2 do not necessarily share identical model-client methods, prompt objects or outputs. Compare your snippet with the official cookbook for that family. A Python call that executes successfully still needs scientific interpretation.
Before opening a support issue
Record the installed package version, operating system, model ID, redacted traceback, minimal non-sensitive input and exact step that failed. Never publish private sequences, credentials or confidential laboratory data by accident.
Primary sources and further reading
Editorial review: October 2026. This guide is educational. Model availability and APIs may change; confirm exact versions in the linked documentation. Generated structures and sequences require independent scientific validation.