Core Objective:
Develop an extension for the ail specification that allows the runner to influence Mixture-of-Experts (MoE) routing at the inference layer, as theorized in arXiv:2509.09660 (SteerMoE). This moves ail from "Top-Down" prompt orchestration to "Direct Neuromodulation" of the model’s internal reasoning experts.
Technical Context for the Issue:
- The Problem: Prompting alone is a "leaky" abstraction for controlling model behavior. MoE models often "flicker" between experts, leading to inconsistent reasoning or "alignment faking."
- The Solution: Implement a
steering block in the ail YAML that allows the runner to apply logit-biasing to specific experts during the forward pass.
- Safety & Compliance: Frame this as a Faithfulness & Determinism feature. Steering should be used to reinforce "Logic" and "Verification" experts, ensuring the model remains grounded in the provided context.
Requested Issue Sections:
-
Proposed Spec Syntax:
- Introduce a
steering block nested within steps.
- Include fields for
activate, suppress, and intensity.
- Define a
requirement_level (e.g., strict, flexible) to handle non-MoE or unsupported models.
-
Expert Mapping Strategy:
- Propose a sidecar
expert_map.json standard. Since Expert IDs are model-specific, ail needs a translation layer that maps human-readable intents (e.g., reasoning, code_generation) to specific expert indices.
-
Runner Requirements (Rust/C++):
- Outline the need for hooks in the inference engine (e.g.,
llama.cpp or candle) to intercept the Gate/Router logits before the top-k selection.
-
Error Handling & Validation:
- What happens if a
strict steering requirement is met with a dense model?
- Define the "Degraded Mode" where the runner falls back to standard prompting if steering is unavailable.
-
Safety Guardrails:
- Explicitly mention that steering can be used to enforce safety experts in high-stakes loops, providing a verifiable "Neuro-Audit" log of the model's internal state.
Core Objective:
Develop an extension for the
ailspecification that allows the runner to influence Mixture-of-Experts (MoE) routing at the inference layer, as theorized in arXiv:2509.09660 (SteerMoE). This movesailfrom "Top-Down" prompt orchestration to "Direct Neuromodulation" of the model’s internal reasoning experts.Technical Context for the Issue:
steeringblock in theailYAML that allows the runner to apply logit-biasing to specific experts during the forward pass.Requested Issue Sections:
Proposed Spec Syntax:
steeringblock nested withinsteps.activate,suppress, andintensity.requirement_level(e.g.,strict,flexible) to handle non-MoE or unsupported models.Expert Mapping Strategy:
expert_map.jsonstandard. Since Expert IDs are model-specific,ailneeds a translation layer that maps human-readable intents (e.g.,reasoning,code_generation) to specific expert indices.Runner Requirements (Rust/C++):
llama.cpporcandle) to intercept the Gate/Router logits before the top-k selection.Error Handling & Validation:
strictsteering requirement is met with a dense model?Safety Guardrails: