fix: EETQLinear with bias in layers.py (#1176)

2023-10-19 18:15:05 +08:00 · 2023-10-19 18:15:05 +08:00 · 648ea06430
parent 9179605e1e
commit 648ea06430
1 changed files with 1 additions and 4 deletions
--- a/server/text_generation_server/utils/layers.py
+++ b/server/text_generation_server/utils/layers.py
@ -155,10 +155,7 @@ class EETQLinear(nn.Module):
        device = weight.device
        weight = torch.t(weight).contiguous().cpu()
        weight, scale = quant_weights(weight, torch.int8, False)
-        if bias:
-            bias = weights.get_tensor(f"{prefix}.bias")
-        else:
-            bias = None
+
        self.weight = weight.cuda(device)
        self.scale = scale.cuda(device)
        self.bias = bias.cuda(device) if bias is not None else None