Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 01:04:32 +02:00
Code Issues Packages Projects Releases Wiki Activity

Cleanup

Browse Source
This commit is contained in:
Disty0
2025-06-05 13:38:04 +03:00
parent e25890bb1d
commit ad2a4ad616
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/model_quant_sdnq.py
+1 -1
View File
@@ -405,7 +405,7 @@ def quantize_fp8_matmul_input_tensorwise(input: torch.FloatTensor, scale: torch.
return input, scale
def quantize_int8_matmul_input(input: torch.FloatTensor, scale: torch.FloatTensor, flatten: bool = True) -> Tuple[torch.ByteTensor, torch.FloatTensor]:
def quantize_int8_matmul_input(input: torch.FloatTensor, scale: torch.FloatTensor) -> Tuple[torch.ByteTensor, torch.FloatTensor]:
input = input.flatten(0,-2).contiguous()
input_scale = torch.div(input.abs().amax(dim=-1, keepdims=True), 127)
input = torch.div(input, input_scale).round_().clamp_(-128, 127).to(torch.int8)