From 2ffe847358ebe983bf09152defd8b6e5cdf1c3e4 Mon Sep 17 00:00:00 2001
From: Asher Feldman <59994+asher@users.noreply.github.com>
Date: Sun, 18 Oct 2026 16:44:34 -0700
Subject: [PATCH] release: 0.5.1

---
 .github/workflows/test.yml      | 2 +-
 CHANGELOG.md                    | 7 +++++--
 gmlx/comfyui_pack/version.py    | 2 +-
 gmlx/upstream/versions.py       | 2 +-
 pyproject.toml                  | 4 ++--
 tests/upstream/test_versions.py | 2 +-
 6 files changed, 11 insertions(+), 8 deletions(-)

diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index fd7357eca..df4004c5e 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -48,7 +48,7 @@ jobs:
           pip install "mlx==0.32.3"
           # Wheel, not sdist. Must satisfy the pyproject floor or the
           # editable install upgrades it; tests/test_ci_pins.py gates that.
-          pip install "mlx-kquant==0.5.1"
+          pip install "mlx-kquant==0.5.2"
           # `all` brings the chat, voice and assistant extras, so their tests
           # run instead of skipping themselves away.
           pip install -e ".[all]" pytest pytest-xdist httpx build
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 652061b73..4c68358d7 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -6,6 +6,8 @@ adhere to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased]
 
+## [0.5.1] - 2026-10-18
+
 ### Added
 
 - `gmlx pull`, `image generate`, `serve` and the ComfyUI pack run MOSS-Image
@@ -49,8 +51,9 @@ adhere to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
   its oldest requests instead of decoding plain. `GMLX_MTP_MIXED=0` reverts.
 - `gmlx serve` with MTP on Qwen3.5 to 3.8 decodes faster when it batches
   several requests.
-- Requires mlx-kquant 0.5.1. With it, speculative decoding on Qwen3.5 to 3.8
-  runs faster on NAX GPUs (M5).
+- Requires mlx-kquant 0.5.2. With it, Nemotron 3.5 Bolt decodes about 1.75
+  times as fast, Q8_0 and IQ1 to IQ3 models decode faster, and speculative
+  decoding on Qwen3.5 to 3.8 runs faster on NAX GPUs (M5).
 - With `--stream-experts`, a prompt that uses most of a layer's experts
   prefills faster. `GMLX_ARENA_SPLIT_MAX_SHARE` tunes the choice.
 - `gmlx pull` of an image model reuses a VAE GGUF on disk packed for its
diff --git a/gmlx/comfyui_pack/version.py b/gmlx/comfyui_pack/version.py
index 62634594c..b29a91062 100644
--- a/gmlx/comfyui_pack/version.py
+++ b/gmlx/comfyui_pack/version.py
@@ -6,7 +6,7 @@
 differ.
 """
 
-__version__ = "0.5.0"
+__version__ = "0.5.1"
 
 # The oldest ComfyUI the pack loads on. `gmlx comfyui install` refuses an
 # older one.
diff --git a/gmlx/upstream/versions.py b/gmlx/upstream/versions.py
index 78e54294d..b1599b052 100644
--- a/gmlx/upstream/versions.py
+++ b/gmlx/upstream/versions.py
@@ -18,7 +18,7 @@
 # gmlx's mlx-kquant requirement. Each mlx-kquant release pins the mlx its
 # kernels were built against, so mlx moves with it. The test holds it equal
 # to the pyproject dependency.
-KQUANT_REQUIREMENT = "mlx-kquant>=0.5.1,<0.6"
+KQUANT_REQUIREMENT = "mlx-kquant>=0.5.2,<0.6"
 
 _COPY_DIR = os.path.join(os.path.dirname(os.path.dirname(
     os.path.abspath(__file__))), "_vendor")
diff --git a/pyproject.toml b/pyproject.toml
index 42c3b102e..1e5eab881 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "gmlx"
-version = "0.5.0"
+version = "0.5.1"
 description = "A local inference platform for Apple Silicon: run, chat with, serve, and fine-tune the GGUF ecosystem's quantized models natively on MLX, straight off the file."
 readme = "README.md"
 requires-python = ">=3.11"  # 3.10 EOLs 2026-10; mlx-kquant's lower floor is a library floor
@@ -38,7 +38,7 @@ dependencies = [
     # 0.5.0 on pins mlx 0.32.3, above the copied mlx-vlm's floor of 0.32.2
     # (tests/test_ci_pins.py checks the CI pin); <0.6 guards the next
     # kquant breaking change.
-    "mlx-kquant>=0.5.1,<0.6",
+    "mlx-kquant>=0.5.2,<0.6",
     # Direct imports of both throughout (mx.*, np.*).
     "mlx",
     "numpy",
diff --git a/tests/upstream/test_versions.py b/tests/upstream/test_versions.py
index 87e4bc49d..97420f02b 100644
--- a/tests/upstream/test_versions.py
+++ b/tests/upstream/test_versions.py
@@ -36,7 +36,7 @@ def test_mlx_below_the_floor_raises(monkeypatch):
 
 @pytest.mark.parametrize("route, hint", [
     ("pip", "upgrade mlx-kquant, which brings the mlx it pins "
-            "(pip install -U 'mlx-kquant>=0.5.1,<0.6'), or reinstall gmlx "
+            "(pip install -U 'mlx-kquant>=0.5.2,<0.6'), or reinstall gmlx "
             "into a fresh venv"),
     ("homebrew", "reinstall gmlx to upgrade it "
                  "(brew reinstall asher/gmlx/gmlx)"),