lym00 commited on
Commit
f615c15
·
verified ·
1 Parent(s): 40477c1

Upload nvfp4-fast.yaml

Browse files
Files changed (1) hide show
  1. nvfp4-fast.yaml +44 -0
nvfp4-fast.yaml ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ quant:
2
+ smooth:
3
+ proj:
4
+ num_grids: 10
5
+ calib:
6
+ num_samples: 32
7
+ wgts:
8
+ dtype: sfp4_e2m1_all
9
+ group_shapes:
10
+ - - -1
11
+ - -1
12
+ - - 1
13
+ - 16
14
+ - 1
15
+ - 1
16
+ - 1
17
+ scale_dtypes:
18
+ - null
19
+ - sfp8_e4m3_nan
20
+ ipts:
21
+ static: false
22
+ dtype: sfp4_e2m1_all
23
+ group_shapes:
24
+ - - 1
25
+ - 16
26
+ - 1
27
+ - 1
28
+ - 1
29
+ scale_dtypes:
30
+ - sfp8_e4m3_nan
31
+ enable_extra_wgts: true
32
+ extra_wgts:
33
+ dtype: sint4
34
+ group_shapes:
35
+ - - 1
36
+ - 64
37
+ - 1
38
+ - 1
39
+ - 1
40
+ scale_dtypes:
41
+ - null
42
+ includes:
43
+ - transformer_norm
44
+ - transformer_add_norm