Figure 2.7
LLM Configuration Dimensions: Configuring an Lm Instance
Configuration Dimensions
λ
Scale
small mid large frontier
Capability ceiling. Determines cost and latency characteristics.
ρ
Reasoning Mode
standard extended-think chain-of-thought
How the Lm processes complex problems. Trading latency for reasoning depth.
τ
Modality
text multimodal code-specialized
What kinds of input the Lm can process. A hard constraint, not a preference.
κ
Context Window
bounded extended
How much context the Lm can hold in active attention.
φ
Provider
Anthropic OpenAI Google open-weight
Training origin. Determines capability profile, safety properties, cost structure.
Element
Lm
Large Language Model
Period 4 / Cognition
Only element requiring internal configuration specification
Molecular Notation
Full specification
Lm(lambda:frontier,
rho:think,
tau:multi,
phi:Anthropic)
Frontier-scale multimodal reasoning model. All four dimensions specified.
Minimal specification
Lm(lambda:small,
tau:code)
Small code-specialized model. Unspecified dimensions (rho, kappa, phi) are unconstrained by the design.
Substitution rule: Substituting an Lm instance with different configuration dimensions is not a performance degradation. It is a structural change that may invalidate the composition entirely. Configuration dimensions are structural specifications, not metadata.
Five configuration dimensions specify Lm configuration within the ASM. Highlighted values (orange borders) show a fully specified frontier-scale Anthropic model. Configuration dimensions appear in molecular formulas using parenthetical notation immediately following the Lm symbol. Unspecified dimensions indicate that the composition can function with any valid value for that dimension: a design choice, not an omission. The substitution rule at lower right states the structural significance: changing any configuration dimension is a structural modification, not merely a preference change.