Skip to content

[UC-8] Efficient sequence models / spiking transformers #19

Description

@w4ffl35

see plans/use_case_spiking_transformers.md — goal, reference architecture (encode → spiking sequence session → head → spikeforge-serve control → clients → throughput/memory /metrics; NIR export gate), rate/delta coding, sequence_attn (simulation-only) + sequence_mlp control, acceptance (quality within 5% of dense baseline or honest statement; throughput/memory reported; typed UnsupportedStageError), MVP phases P0–P4, dependencies PT-W1/W2/W3/W6.

Spec: plans/use_case_spiking_transformers.md

Refs #12 (umbrella tracking the ten chip-less production use cases).

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    architectureArchitectural design or cross-cutting structural changeenhancementNew feature or request

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions