Foundation Models issues manifest as context window exceeded errors, guardrail violations, slow generation, availability failures, and unexpected output. Core principle 80% of Foundation Models problems stem from misunderstanding model capabilities (3B parameter device-scale model, not world knowledge), context limits (4096 tokens), or availability requirements—not framework bugs.
If you see ANY of these, suspect a Foundation Models misunderstanding, not framework breakage:
Critical distinction Foundation Models is a device-scale model (3B parameters) optimized for summarization, extraction, classification—NOT world knowledge or complex reasoning. Using it for the wrong task guarantees poor results.
Úselo al depurar problemas de Foundation Models: contexto excedido, violaciones de las barreras de seguridad, generación lenta, problemas de disponibilidad, lenguaje no compatible o resultados inesperados. Diagnóstico sistemático con defensa de crisis de producción. Fuente: megastep/codex-skills.