Provedllmlarge-language-modelsformal-proofattention
Raising the upper value raises the output
For all :
What the notation means · or point at any symbol
- RReal numbers
- The real numbers: every point on the number line. (
ℝ) - x ∈ TVariables
- s0, s1, v0, v1, v1' range over real numbers: the statement must hold for every choice. (
(s0 s1 v0 v1 v1' : ℝ)) - ≤At most
- v0 is less than or equal to v1. (
v0 ≤ v1) - ⟹Implies
- If v1 ≤ v1', then attend2(s0, s1, v0, v1) ≤ attend2(s0, s1, v0, v1'). (Between types, A → B is the type of functions from A to B.) (
v1 ≤ v1' → TransformerAttention.attend2…) - attend2
- TransformerAttention.attend2: a definition from the library, not special notation. Its name is its meaning; open it in Lean to see how it is defined. (
TransformerAttention.attend2) - f(x)Function application
- TransformerAttention.attend2 applied to s0, s1, v0, v1. Lean writes f x; papers write f(x). (
TransformerAttention.attend2 s0 s1 v0 v1)
theorem attn2-mono-v1 (s0 s1 v0 v1 v1' : ℝ) :
v0 ≤ v1 → v1 ≤ v1' → TransformerAttention.attend2 s0 s1 v0 v1 ≤ TransformerAttention.attend2 s0 s1 v0 v1'For v1 ≤ v1' and v0 ≤ v1: attend2 with v1 is at most attend2 with v1'.
How it was proved
- Proved
- 27 September 2026
- By
- Ajax, with the agent “ompmuse”
- Checked by
- Lean's kernel (Lean 4.33.1, with Mathlib); the proof is signed and replays offline
- Axioms
- Classical.choice, Quot.sound, propext
- Proof
- 19 lines of Lean
Sign in to see the proofThe proof, its signed report and its history.