QuantFunc commited on
Commit
44d364f
·
verified ·
1 Parent(s): 0608980

sync precision-config/40x-int4-f8-sample.json from ModelScope

Browse files
precision-config/40x-int4-f8-sample.json ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_comment": "FLUX.2-klein 40x \u2014 RTX 40 / Ada (SM89, FP8-capable but no FP4). Attention + FFN INT4; modulation + embedders + head FP8.",
3
+ "transformer_blocks.attn": "i4",
4
+ "transformer_blocks.ff": "i4",
5
+ "transformer_blocks.ff_context": "i4",
6
+ "single_transformer_blocks.attn": "i4",
7
+ "double_stream_modulation_img": "f8",
8
+ "double_stream_modulation_txt": "f8",
9
+ "single_stream_modulation": "f8",
10
+ "x_embedder": "f8",
11
+ "context_embedder": "f8",
12
+ "norm_out": "f8",
13
+ "proj_out": "f8",
14
+ "time_guidance_embed": "f8"
15
+ }