Prepare these inputs
- Pinned attention architecture, layers, KV heads and head dimensions where available.
- Input/output token distribution, concurrent sequences and cache dtype/policy.
- Runtime support for sliding windows, offload or quantized cache if proposed.
Decisions and sequence
- Test realistic long inputs and output lengths, including the prefill phase.
- Measure peak memory and latency as concurrency increases; keep model limit and service limit distinct.
- Document cache persistence and isolation assumptions, especially where prefix reuse is enabled.
The first useful deliverable
A tested context-and-concurrency envelope with cache strategy, request limits and overload behavior.
Failure modes to catch
A nominal context maximum is not a promise that every request at that size fits. Do not apply a standard attention-cache formula to recurrent, hybrid or multimodal models without checking the implementation.
How this follows the paper
Read The Independent Institution, sections 5. The role and sequence above are practical preparation derived from those sections; any newly proposed commercial arrangement still needs its own confirmed terms.
Computraq participation status
Computraq is in prelaunch. This guide interprets the proposed operating model in The Independent Institution; it does not establish current hosting, paid booking, provider integration, partnership or a guarantee of service. Use the paper, original slides and expression-of-interest form to prepare a concrete contribution.