Hugging Face's logo

changwangss
/
FLUX.2-dev-MXFP8-RTN-FullEncoder-AutoRound

Text-to-Image
Diffusers
Safetensors
Flux2Pipeline
auto-round
mxfp8
rtn
multimodal
Model card Files
xet
 Community
FLUX.2-dev-MXFP8-RTN-FullEncoder-AutoRound
File size: 1,736 Bytes
6ec783c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
{
  "_class_name": "Flux2Transformer2DModel",
  "_diffusers_version": "0.36.0.dev0",
  "attention_head_dim": 128,
  "axes_dims_rope": [
    32,
    32,
    32,
    32
  ],
  "eps": 1e-06,
  "in_channels": 128,
  "joint_attention_dim": 15360,
  "mlp_ratio": 3.0,
  "num_attention_heads": 48,
  "num_layers": 8,
  "num_single_layers": 48,
  "out_channels": null,
  "patch_size": 1,
  "rope_theta": 2000,
  "timestep_guidance_channels": 256,
  "quantization_config": {
    "quant_method": "auto-round",
    "packing_format": "auto_round:llm_compressor",
    "bits": 8,
    "group_size": 32,
    "sym": true,
    "data_type": "mx_fp",
    "iters": 0,
    "model_free": true,
    "autoround_version": "0.17.0",
    "enable_quanted_input": false,
    "block_name_to_quantize": "transformer_blocks,single_transformer_blocks",
    "act_bits": 8,
    "act_data_type": "mx_fp",
    "act_dynamic": true,
    "act_group_size": 32,
    "act_sym": true,
    "extra_config": {
      "double_stream_modulation_img.linear": {
        "bits": 16,
        "data_type": "float",
        "act_bits": 16,
        "act_data_type": "float"
      },
      "double_stream_modulation_txt.linear": {
        "bits": 16,
        "data_type": "float",
        "act_bits": 16,
        "act_data_type": "float"
      },
      "single_stream_modulation.linear": {
        "bits": 16,
        "data_type": "float",
        "act_bits": 16,
        "act_data_type": "float"
      },
      "norm_out.linear": {
        "bits": 16,
        "data_type": "float",
        "act_bits": 16,
        "act_data_type": "float"
      },
      "proj_out": {
        "bits": 16,
        "data_type": "float",
        "act_bits": 16,
        "act_data_type": "float"
      }
    }
  }
}