Hugging Face's logo

WaveCut
/
Anima-Preview-3-SDNQ-uint4

Text-to-Image
Diffusers
Safetensors
Cosmos
sdnq
transformer-component
anima
uint4
4bit
8-bit precision
Model card Files
xet
 Community
WaveCut commited on
Commit
2cb5ead
·
verified ·
1 Parent(s): 3b43ca3

Add SDNQ uint4 quantized Anima transformer

Browse files
Files changed (7) hide show
  1. README.md +17 -0
  2. config.json +73 -0
  3. diffusion_pytorch_model-00001-of-00003.safetensors +3 -0
  4. diffusion_pytorch_model-00002-of-00003.safetensors +3 -0
  5. diffusion_pytorch_model-00003-of-00003.safetensors +3 -0
  6. diffusion_pytorch_model.safetensors.index.json +0 -0
  7. quantization_config.json +29 -0
README.md ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ base_model: circlestone-labs/Anima
4
+ tags:
5
+ - sdnq
6
+ - uint4
7
+ - 4bit
8
+ - diffusion-single-file
9
+ - text-to-image
10
+ - cosmos
11
+ ---
12
+
13
+ # Anima SDNQ UINT4 Diffusers Transformer
14
+
15
+ SDNQ 4-bit (`uint4`) quantization of the `circlestone-labs/Anima` Preview3 diffusion transformer checkpoint.
16
+
17
+ Important: the upstream Anima checkpoint contains additional `llm_adapter.*` weights used by the native ComfyUI loader. Current `diffusers.CosmosTransformer3DModel.from_single_file` ignores those adapter weights, so this repo is a diffusers-compatible SDNQ transformer artifact, not a full ComfyUI-native split checkpoint replacement.
config.json ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "CosmosTransformer3DModel",
3
+ "_diffusers_version": "0.38.0",
4
+ "adaln_lora_dim": 256,
5
+ "attention_head_dim": 128,
6
+ "concat_padding_mask": true,
7
+ "controlnet_block_every_n": null,
8
+ "crossattn_proj_in_channels": 1024,
9
+ "encoder_hidden_states_channels": 1024,
10
+ "extra_pos_embed_type": null,
11
+ "img_context_dim_in": null,
12
+ "img_context_dim_out": 2048,
13
+ "img_context_num_tokens": 256,
14
+ "in_channels": 16,
15
+ "max_size": [
16
+ 128,
17
+ 240,
18
+ 240
19
+ ],
20
+ "mlp_ratio": 4.0,
21
+ "num_attention_heads": 16,
22
+ "num_layers": 28,
23
+ "out_channels": 16,
24
+ "patch_size": [
25
+ 1,
26
+ 2,
27
+ 2
28
+ ],
29
+ "quantization_config": {
30
+ "add_skip_keys": false,
31
+ "dequantize_fp32": false,
32
+ "dynamic_loss_threshold": null,
33
+ "group_size": 0,
34
+ "is_integer": true,
35
+ "is_training": false,
36
+ "modules_dtype_dict": {},
37
+ "modules_quant_config": {},
38
+ "modules_to_not_convert": [
39
+ "crossattn_proj",
40
+ "patch_embed",
41
+ "transformer_blocks.0.norm*",
42
+ "norm_out",
43
+ "proj_out",
44
+ "learnable_pos_embed",
45
+ "time_embed"
46
+ ],
47
+ "non_blocking": false,
48
+ "quant_conv": false,
49
+ "quant_embedding": false,
50
+ "quant_method": "sdnq",
51
+ "quantization_device": null,
52
+ "quantized_matmul_dtype": null,
53
+ "return_device": null,
54
+ "sdnq_version": "0.1.8",
55
+ "svd_rank": 32,
56
+ "svd_steps": 8,
57
+ "use_dynamic_quantization": false,
58
+ "use_grad_ckpt": true,
59
+ "use_quantized_matmul": true,
60
+ "use_quantized_matmul_conv": false,
61
+ "use_static_quantization": true,
62
+ "use_stochastic_rounding": false,
63
+ "use_svd": false,
64
+ "weights_dtype": "uint4"
65
+ },
66
+ "rope_scale": [
67
+ 1.0,
68
+ 4.0,
69
+ 4.0
70
+ ],
71
+ "text_embed_dim": 1024,
72
+ "use_crossattn_projection": false
73
+ }
diffusion_pytorch_model-00001-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:17539390f73da7346934e68e0dc48c01e49eef1fa61c3d0b324bcb45a7ba37e4
3
+ size 499618232
diffusion_pytorch_model-00002-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e4b10de93317e2bd7cd8ad0d1efb8bd264eecabad604d818bc45e9704230dc87
3
+ size 497771600
diffusion_pytorch_model-00003-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7ff2e0a2a931e2932e015bf492082470abcb5363f5745acd195195ac0f04d656
3
+ size 139089360
diffusion_pytorch_model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
quantization_config.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_skip_keys": false,
3
+ "dequantize_fp32": false,
4
+ "dynamic_loss_threshold": null,
5
+ "group_size": 0,
6
+ "is_integer": true,
7
+ "is_training": false,
8
+ "modules_dtype_dict": {},
9
+ "modules_quant_config": {},
10
+ "modules_to_not_convert": [],
11
+ "non_blocking": false,
12
+ "quant_conv": false,
13
+ "quant_embedding": false,
14
+ "quant_method": "sdnq",
15
+ "quantization_device": null,
16
+ "quantized_matmul_dtype": null,
17
+ "return_device": null,
18
+ "sdnq_version": "0.1.8",
19
+ "svd_rank": 32,
20
+ "svd_steps": 8,
21
+ "use_dynamic_quantization": false,
22
+ "use_grad_ckpt": true,
23
+ "use_quantized_matmul": true,
24
+ "use_quantized_matmul_conv": false,
25
+ "use_static_quantization": true,
26
+ "use_stochastic_rounding": false,
27
+ "use_svd": false,
28
+ "weights_dtype": "uint4"
29
+ }