core-labs/core
1
0
Fork 0
Code Issues Pull requests 1 Projects Releases Packages Wiki Activity Actions

Avoid identity stalls in generation loop

Browse source
This commit is contained in:
Shay 2026-05-13 13:16:48 -07:00
parent defd7a9a53
commit d781ba71db
1 changed files with 17 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

22
generate/stream.py
View file

@ -1,7 +1,7 @@
"""
Generation loop — token streaming from the versor manifold.
Every token: nearest word to current F via CGA inner product.
Every token: nearest non-current word to current F via CGA inner product.
Every step: F <- versor_apply(V, F) where V = word_transition_rotor(A, B).
Architectural boundaries enforced here:
@ -12,7 +12,7 @@ Architectural boundaries enforced here:
structurally by versor_apply() and the closed algebra.
No confidence gates. No IDK fallback. No attractor clamping.
F is always on the manifold. nearest() is always exact.
F is always on the manifold. nearest() is exact.
"""
from __future__ import annotations
@ -22,6 +22,18 @@ from algebra.rotor import word_transition_rotor
from generate.result import GenerationResult
def _nearest_next(vocab, F_voiced, current_node: int) -> tuple[str, int]:
"""
Select the nearest non-current vocabulary point when possible.
Allowing the current node to win makes V = transition(A, A), which is an
identity-like transition and can stall generation forever on one token.
VocabManifold already exposes exclude_idx for this exact seam.
"""
exclude_idx = current_node if len(vocab) > 1 else -1
return vocab.nearest(F_voiced, exclude_idx=exclude_idx)
def generate(
state: FieldState,
vocab,
@ -34,7 +46,7 @@ def generate(
Loop:
1. Apply persona motor to current field
2. Find nearest vocab node via CGA inner product
2. Find nearest non-current vocab node via CGA inner product
3. Emit token
4. Build transition rotor: V = word_transition_rotor(A, B)
where A = versor at current node, B = versor at nearest node
@ -50,7 +62,7 @@ def generate(
for _ in range(max_tokens):
F_voiced = persona.apply(current.F)
word, word_idx = vocab.nearest(F_voiced)
word, word_idx = _nearest_next(vocab, F_voiced, current.node)
tokens.append(word)
if record_trajectory:
@ -89,7 +101,7 @@ async def agenerate(
current = state
for _ in range(max_tokens):
F_voiced = persona.apply(current.F)
word, word_idx = vocab.nearest(F_voiced)
word, word_idx = _nearest_next(vocab, F_voiced, current.node)
yield word
A = vocab.get_versor_at(current.node)