Architecture decision
Use the selected model card for sizing. A memory estimate for dense 7B weights cannot be reused for a different Mistral release.
GLOBALATLAS AI · Infrastructure for Mistral
Infrastructure for Mistral
Mistral 7B and other Mistral releases differ in size and architecture. Select a specific model card before sizing hardware.
Match the Mistral architecture to the runtime
- 01
Exact model card and weight revision
- 02
Dense or mixture-of-experts architecture
- 03
Supported runtime, precision and context
Configuration requirements
A 7B dense model at 16 bits has roughly 14 GB decimal of weights alone. Leave memory for runtime and context; do not infer a user count from that estimate.