From fa5b05b70b15de86c63a210bca6e55f47875629d Mon Sep 17 00:00:00 2001
From: Giuseppe Franco <giuseppefranco4@gmail.com>
Date: Fri, 14 Nov 2025 11:52:24 +0100
Subject: [PATCH] Support for out-of-source quantizers (#3534)

* Support for out-of-source quantizers

* Fix decorators and functions to be staticmethod

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
---
 unsloth/models/llama.py  | 8 +++++++-
 unsloth/models/loader.py | 7 ++++++-
 2 files changed, 13 insertions(+), 2 deletions(-)

diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index 9537dd63c3..caf259c0bd 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -2046,6 +2046,11 @@ def unsloth_fast_generate(
 
 
 class FastLlamaModel:
+    @staticmethod
+    def _prepare_for_qat(model, qat_scheme):
+        model = _prepare_model_for_qat(model, qat_scheme)
+        return model
+
     @staticmethod
     def pre_patch():
         init_name, function = patch_llama_rope_scaling(
@@ -3007,7 +3012,8 @@ class FastLlamaModel:
         # Apply QAT + LoRA if specified
         if qat_scheme is not None:
             print("Unsloth: Applying QAT to mitigate quantization degradation")
-            model = _prepare_model_for_qat(model, qat_scheme)
+            model = FastLlamaModel._prepare_for_qat(model, qat_scheme)
+
 
         model._saved_temp_tokenizer = _saved_temp_tokenizer
 
diff --git a/unsloth/models/loader.py b/unsloth/models/loader.py
index 15f01cf15d..9d8a2ce9fa 100644
--- a/unsloth/models/loader.py
+++ b/unsloth/models/loader.py
@@ -595,6 +595,11 @@ except:
 
 
 class FastModel(FastBaseModel):
+    @staticmethod
+    def _prepare_for_qat(model, qat_scheme):
+        model = _prepare_model_for_qat(model, qat_scheme)
+        return model
+
     @staticmethod
     def from_pretrained(
         model_name = "unsloth/Llama-3.2-11B-Vision-Instruct-bnb-4bit",
@@ -1144,7 +1149,7 @@ class FastModel(FastBaseModel):
         # Apply QAT if specified
         if qat_scheme is not None:
             print("Unsloth: Applying QAT to mitigate quantization degradation")
-            model = _prepare_model_for_qat(model, qat_scheme)
+            model = FastModel._prepare_for_qat(model, qat_scheme)
 
         # Patch Tiled MLP
         # to turn on set UNSLOTH_TILED_MLP to "arctic", "target", or "target:{GB}""