Troubleshoot#
No engine: check
lllm2 engines listand the configured search roots; runlllm2 engines install cudaif needed.No GPU: check
nvidia-smiand the selected engine’s device probe.Download certificate errors: engine and model downloads automatically use RHEL’s system CA bundle when present. Other systems use Python’s default trust settings. Explicit
SSL_CERT_FILEorSSL_CERT_DIRsettings take precedence; check those settings if verification still fails. Site certificates should be installed in the system trust store.Startup fails: read the engine log, reduce context or GPU layers, and disable unsupported acceleration options. Automatic placement needs an engine with memory-fitting support.
Client exceeds context: increase context or reduce slots, then restart the model and client. Available memory still limits the allocation.