From 11366842c9ab6b3657738112c39943df16471060 Mon Sep 17 00:00:00 2001
From: Mr-Neutr0n <64578610+Mr-Neutr0n@users.noreply.github.com>
Date: Fri, 6 Feb 2026 23:57:37 +0530
Subject: [PATCH] Add LARS to str2optimizer32bit dictionary

LARS optimizer was missing from str2optimizer32bit, causing KeyError
when using LARS32bit optimizer. LARS uses momentum-based kernels since
it's essentially SGD with momentum plus layerwise adaptive learning rates.

Fixes #1810
---
 bitsandbytes/backends/cuda/ops.py | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/bitsandbytes/backends/cuda/ops.py b/bitsandbytes/backends/cuda/ops.py
index cb2796b1f..ecafbdf25 100644
--- a/bitsandbytes/backends/cuda/ops.py
+++ b/bitsandbytes/backends/cuda/ops.py
@@ -574,6 +574,10 @@ def _gemv_4bit_impl(
         lib.cademamix32bit_grad_fp16,
         lib.cademamix32bit_grad_bf16,
     ),
+    "lars": (
+        lib.cmomentum32bit_grad_32,
+        lib.cmomentum32bit_grad_16,
+    ),
 }
 
 str2optimizer8bit_blockwise = {