From f8b28b00241f69fcbcd47892dec7adc5f0f7ba6c Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Sat, 4 Jul 2026 12:57:50 +0000 Subject: [PATCH] Tighten comments --- unsloth/models/loader.py | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/unsloth/models/loader.py b/unsloth/models/loader.py index 9c485af1b5..d60a3f02bd 100644 --- a/unsloth/models/loader.py +++ b/unsloth/models/loader.py @@ -207,10 +207,8 @@ DISABLE_COMPILE_MODEL_NAMES = [ ] # Architectures with gated-deltanet (linear attention) layers. Unsloth bundles the -# flash-linear-attention (fla) Triton kernels (unsloth_zoo/_vendored/fla) and injects -# them automatically, so no `pip install flash-linear-attention` is needed. Transformers -# only falls back to the several-times-slower pure PyTorch path when those bundled kernels -# cannot be enabled on the current setup. +# flash-linear-attention Triton kernels (unsloth_zoo/_vendored/fla), so no install is +# needed; transformers uses the much slower pure PyTorch path only when they can't be enabled. FLA_MODEL_TYPE_PREFIXES = ("qwen3_next", "qwen3_5", "kimi_linear") _fla_advised = False