From a76cac2c3bb9c0c9edf36980f86d1fcac4c2c56a Mon Sep 17 00:00:00 2001
From: Sunny-bot1 <592045536@qq.com>
Date: Fri, 26 Sep 2025 16:46:13 +0800
Subject: [PATCH] fix machete pre quant

---
 fastdeploy/model_executor/layers/linear.py                   | 1 +
 fastdeploy/model_executor/layers/quantization/weight_only.py | 5 ++++-
 2 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/fastdeploy/model_executor/layers/linear.py b/fastdeploy/model_executor/layers/linear.py
index d17ab1be34..b8f65fb492 100644
--- a/fastdeploy/model_executor/layers/linear.py
+++ b/fastdeploy/model_executor/layers/linear.py
@@ -128,6 +128,7 @@ def __init__(
         self.with_bias = with_bias
         self.add_bias = add_bias
         self.prefix = prefix
+        self.is_quantized = fd_config.model_config.is_quantized
         # key
         if weight_key:
             self.weight_key = f"{prefix}.{weight_key}"
diff --git a/fastdeploy/model_executor/layers/quantization/weight_only.py b/fastdeploy/model_executor/layers/quantization/weight_only.py
index 070d0fbf41..8d18b86075 100644
--- a/fastdeploy/model_executor/layers/quantization/weight_only.py
+++ b/fastdeploy/model_executor/layers/quantization/weight_only.py
@@ -20,6 +20,7 @@
 
 import paddle
 from paddle.nn.quant import weight_quantize
+from paddleformers.utils.log import logger
 
 from fastdeploy import envs
 from fastdeploy.model_executor.layers.linear import (
@@ -159,9 +160,11 @@ def get_quant_method(self, layer) -> Optional[QuantMethodBase]:
                 if (
                     _ENABLE_MACHETE
                     and envs.FD_USE_MACHETE == "1"
+                    and not layer.is_quantized
                     and layer.weight_shape[1]
                     and layer.weight_shape[1] % 128 == 0
                 ):
+                    logger.info("Using Machete kernel for WeightOnlyLinearMethod")
                     return MacheteWeightOnlyLinearMethod(self)
                 return GPUWeightOnlyLinearMethod(self)
 
@@ -399,7 +402,7 @@ def __init__(
         super().__init__(quant_config)
 
     def process_prequanted_weights(self, layer, state_dict) -> None:
-        pass
+        raise NotImplementedError("Machete kernel doesn't support prequant. Please set FD_USE_MACHETE to 0.")
 
     def process_loaded_weights(self, layer, weight) -> None:
         from fastdeploy.model_executor.layers.quantization.ops import (