ADIL ECO · DEPLOYMENT
Plan Eco for your hardware.
“A model loads” is not the same as “the system is qualified.” We evaluate model, Thynkr-developed runtime environment, memory, platform and workload together.
Tested does not mean universally supported.
An Eco candidate has been exercised in internal tests on Apple M1/macOS ARM64. These observations do not establish broad support across Windows, Android, other Apple devices or consumer Linux hardware.
- macOSApple Silicon · Initial development target
- Linuxx86-64 · Qualification pending
- Windowsx86-64 · Qualification pending
- Androidarm64 · Qualification pending
Understanding model memory.
A model's file size is not its full operating requirement. Working memory, context state, application buffers and the operating system all consume resources.
- Runtime memory
- 867 MB process footprint · one Apple M1 · 4K context
- Model file
- Model file: 484 MB on disk (Q4_K_M).
- Status
- Measured on one Apple M1 test host. Other hosts are unverified; this is not a minimum requirement.
Local does not always mean disconnected.
Inference may operate locally on a qualified setup. Developer-enabled tools, updates, accounts, data connectors and optional remote services may still communicate over a network. Your application controls those integrations.
Before planning a deployment.
Check the model family, supported hardware, available package, context budget, tool permissions and maintenance needs. As of the current release record, no public package is available for general installation.