From ee561f0a9d5ae1f1168021d7dc79be09bbeac2e7 Mon Sep 17 00:00:00 2001
From: Asher Feldman <59994+asher@users.noreply.github.com>
Date: Sun, 13 Sep 2026 15:53:49 -0700
Subject: [PATCH] chore(typing): from __future__ import annotations in the last
 six annotated modules

---
 gmlx/__main__.py          | 2 ++
 gmlx/cache/kv_policy.py   | 2 ++
 gmlx/serve/batch_rows.py  | 2 ++
 gmlx/serve/kv_policy.py   | 2 ++
 gmlx/serve/step_timing.py | 2 ++
 gmlx/spec/helpers.py      | 2 ++
 6 files changed, 12 insertions(+)

diff --git a/gmlx/__main__.py b/gmlx/__main__.py
index a5ffe68b..39a8b991 100644
--- a/gmlx/__main__.py
+++ b/gmlx/__main__.py
@@ -2,6 +2,8 @@
 script). A lazy import keeps ``-m gmlx <verb>`` light; this is also the form the
 background / launchd re-exec uses, so it must route to the umbrella (note that
 ``-m gmlx.commands.cli`` runs ``run``'s ``main`` instead)."""
+from __future__ import annotations
+
 import sys
 
 
diff --git a/gmlx/cache/kv_policy.py b/gmlx/cache/kv_policy.py
index 762a56b8..3958e6f1 100644
--- a/gmlx/cache/kv_policy.py
+++ b/gmlx/cache/kv_policy.py
@@ -6,6 +6,8 @@
 logging, and memory pricing all read the same object.
 """
 
+from __future__ import annotations
+
 from dataclasses import dataclass
 
 VALID_BITS = (2, 3, 4, 6, 8)
diff --git a/gmlx/serve/batch_rows.py b/gmlx/serve/batch_rows.py
index 71531109..a16f826d 100644
--- a/gmlx/serve/batch_rows.py
+++ b/gmlx/serve/batch_rows.py
@@ -16,6 +16,8 @@
 of the batch as it currently is.
 """
 
+from __future__ import annotations
+
 
 def batch_rows(gen) -> int:
     """Active rows in ``gen._generation_batch``; 0 when there is none.
diff --git a/gmlx/serve/kv_policy.py b/gmlx/serve/kv_policy.py
index 9354a00a..f9b1ffba 100644
--- a/gmlx/serve/kv_policy.py
+++ b/gmlx/serve/kv_policy.py
@@ -6,6 +6,8 @@
 object. An error verdict fails residency with the reason.
 """
 
+from __future__ import annotations
+
 import logging
 import math
 import os
diff --git a/gmlx/serve/step_timing.py b/gmlx/serve/step_timing.py
index e3a84332..e8ebb50d 100644
--- a/gmlx/serve/step_timing.py
+++ b/gmlx/serve/step_timing.py
@@ -18,6 +18,8 @@
 (above the batch generator). No-op unless the env var is set.
 """
 
+from __future__ import annotations
+
 import atexit
 import os
 import sys
diff --git a/gmlx/spec/helpers.py b/gmlx/spec/helpers.py
index c23a621d..5a44d386 100644
--- a/gmlx/spec/helpers.py
+++ b/gmlx/spec/helpers.py
@@ -9,6 +9,8 @@
 Logic is a faithful copy (acceptance must stay token-identical to the validated
 path); keep it in sync when mlx-vlm's round changes in a way we want to track.
 """
+from __future__ import annotations
+
 import logging
 import os
 from dataclasses import dataclass