From 74d9dfa89e4fcb2c8830207fa9964f7d12a00229 Mon Sep 17 00:00:00 2001
From: chenxichen <chenxichen@cvte.com>
Date: Wed, 27 Dec 2023 03:25:47 +0000
Subject: [PATCH] Fix incorrect use of bias in awq

---
 server/text_generation_server/utils/awq/quantize/qmodule.py | 5 +----
 server/text_generation_server/utils/layers.py               | 2 +-
 2 files changed, 2 insertions(+), 5 deletions(-)

diff --git a/server/text_generation_server/utils/awq/quantize/qmodule.py b/server/text_generation_server/utils/awq/quantize/qmodule.py
index ca8caf50..31ac7463 100644
--- a/server/text_generation_server/utils/awq/quantize/qmodule.py
+++ b/server/text_generation_server/utils/awq/quantize/qmodule.py
@@ -35,10 +35,7 @@ class WQLinear(nn.Module):
         self.qweight = qweight
         self.qzeros = qzeros
         self.scales = scales
-        if bias:
-            self.bias = bias
-        else:
-            self.bias = None
+        self.bias = bias
 
     @torch.no_grad()
     def forward(self, x):
diff --git a/server/text_generation_server/utils/layers.py b/server/text_generation_server/utils/layers.py
index f38f130e..8f0965ff 100644
--- a/server/text_generation_server/utils/layers.py
+++ b/server/text_generation_server/utils/layers.py
@@ -335,7 +335,7 @@ def get_linear(weight, bias, quantize):
             qweight=qweight,
             qzeros=qzeros,
             scales=scales,
-            bias=bias is not None,
+            bias=bias,
         )
     else:
         raise NotImplementedError(f"Quantization `{quantize}` is not implemented yet.")