Merge pull request #18 from neuralmagic/fix-bias-clone

Bugfix for bias cloning

Author: mgoin

auto_fp8/quantize.py

@@ -1,6 +1,7 @@
 import gc
 import re
 from typing import List, Tuple
+import copy
 
 import torch
 import tqdm
@@ -47,7 +48,7 @@ def per_tensor_quantize(tensor: torch.Tensor) -> Tuple[torch.Tensor, float]:
         )
     else:
         min_val, max_val = tensor.aminmax()
-    amax = min_val.abs().max(max_val.abs())
+    amax = torch.maximum(min_val.abs(), max_val.abs())
     scale = finfo.max / amax.clamp(min=1e-12)
     # scale and clamp the tensor to bring it to
     # the representative range of float8 data type
@@ -202,8 +203,8 @@ def quantize_weights(
             or name in quantize_config.ignored_layers
         ):
             continue
-        quant_weight, quant_scale = per_tensor_quantize(linear.weight.clone())
-        bias = linear.bias.clone() if linear.bias is not None else None
+        quant_weight, quant_scale = per_tensor_quantize(linear.weight)
+        bias = copy.deepcopy(linear.bias) if linear.bias is not None else None
         quant_linear = FP8DynamicLinear(quant_weight, quant_scale, bias)
         replace_module(model, name, quant_linear)
         del linear.weight