Navigate

Quick find

Navigate NeuroForge

Esc

Start typing to search public pages and evidence.

    ERAIS technical claim

    production.local_full_weight_training_viability.v1

    Can all model weights be trained on the local host?

    Evidence maturity: BlockedSource-pinned record

    Reviewed · neuroforge-io/erais

    The wording and state below belong to this source snapshot. Related current context is separate from the recorded claim.

    Approved wording · 12 July 2026

    What the record says.

    Full-weight local training remains blocked pending a safe measured envelope.

    Required evidence

    • A fresh isolated memory plan and guarded dry run must fit model, gradients, optimizer, activations, checkpoint staging, host reserve, and bounded swap without OOM or thrashing.

    Do not infer

    • Do not claim viability merely because large swap exists or because adapter training passed.

    Continue in context

    Related work.

    Explore the current context. It does not update this recorded claim.

    Inspect the training evidence

    Read the state together with its scope and evidence requirements. Commercial availability and technical maturity are separate decisions.

    Return to the complete register