OzzyGT HF Staff commited on
Commit
bc6fbf0
·
verified ·
1 Parent(s): 73cea30

Upload 23 files

Browse files
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer/tokenizer.json filter=lfs diff=lfs merge=lfs -text
LICENSE ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ MIT License
2
+
3
+ Copyright (c) 2026 inclusionAI
4
+ Copyright (c) 2026 The HuggingFace Team
5
+
6
+ Permission is hereby granted, free of charge, to any person obtaining a copy
7
+ of this software and associated documentation files (the "Software"), to deal
8
+ in the Software without restriction, including without limitation the rights
9
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10
+ copies of the Software, and to permit persons to whom the Software is
11
+ furnished to do so, subject to the following conditions:
12
+
13
+ The above copyright notice and this permission notice shall be included in all
14
+ copies or substantial portions of the Software.
15
+
16
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22
+ SOFTWARE.
README.md ADDED
@@ -0,0 +1,114 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - en
4
+ - zh
5
+ license: mit
6
+ base_model:
7
+ - inclusionAI/Ming-Image-0.1-Design
8
+ tags:
9
+ - text-to-image
10
+ - ming-image
11
+ pipeline_tag: text-to-image
12
+ library_name: diffusers
13
+ ---
14
+ # Ming-Image-0.1-Design Diffusers
15
+
16
+ Diffusers version of [inclusionAI/Ming-Image-0.1-Design](https://huggingface.co/inclusionAI/Ming-Image-0.1-Design): the transformer, the text encoder, the connector, the VAE, the tokenizer and the scheduler. Run it with the blocks in [OzzyGT/ming_image_custom_blocks](https://huggingface.co/OzzyGT/ming_image_custom_blocks).
17
+
18
+ Note: This model needs diffusers installed from source, `pip install git+https://github.com/huggingface/diffusers`. The example below uses about 19 GB of VRAM.
19
+
20
+ ## Sample image
21
+
22
+ ![A fitness app home screen generated with the code below](https://huggingface.co/datasets/OzzyGT/diffusers-examples/resolve/main/ming-image/image_phone_app_seed2024.png)
23
+
24
+ This image was generated with the following code, seed `2024`:
25
+
26
+ ```python
27
+ import torch
28
+
29
+ from diffusers import ModularPipelineBlocks
30
+
31
+
32
+ blocks = ModularPipelineBlocks.from_pretrained(
33
+ "OzzyGT/ming_image_custom_blocks",
34
+ trust_remote_code=True,
35
+ components_repo="OzzyGT/ming_image_0_1_design_diffusers",
36
+ trust_components_code=True,
37
+ )
38
+ pipe = blocks.init_pipeline()
39
+ pipe.load_components(dtype=torch.bfloat16)
40
+ for name in ("text_encoder", "transformer"):
41
+ getattr(pipe, name).enable_group_offload(
42
+ onload_device="cuda", offload_type="block_level", num_blocks_per_group=1, use_stream=True
43
+ )
44
+ pipe.connector.to("cuda")
45
+ pipe.vae.to("cuda")
46
+
47
+ prompt = r"""{
48
+ "canvas_settings": {
49
+ "aspect_ratio": "9:16, 1440 × 2560 px",
50
+ "ambient_lighting": "Flat, even digital lighting with soft card shadows only.",
51
+ "image_style": "High-fidelity mobile app screen design exported from Figma, full-bleed with no device frame. Modern dark theme with a near-black background, rounded 24 px cards in dark charcoal, vivid gradient accents in lime green, electric blue, and coral, a clean geometric sans-serif with bold numerals, simple line icons, and crisp, perfectly legible text."
52
+ },
53
+ "layers": [
54
+ {
55
+ "description": "Near-black app background with a faint blue radial glow behind the top third.",
56
+ "coordinates": "cx: 0.500, cy: 0.500, w: 1.000, h: 1.000",
57
+ "hierarchy_and_relation": "Bottom-most layer behind every other element.",
58
+ "color_specs": ["#0D0F14", "#141A2A"]
59
+ },
60
+ {
61
+ "description": "System status bar: on the left, the exact time \"9:41\" in white; on the right, signal, Wi-Fi, and battery icons in white.",
62
+ "coordinates": "cx: 0.500, cy: 0.025, w: 0.900, h: 0.030",
63
+ "hierarchy_and_relation": "Very top of the screen, above the greeting header.",
64
+ "color_specs": ["#FFFFFF"]
65
+ },
66
+ {
67
+ "description": "Greeting header: on the left, the exact small text \"Thursday, 2 October\" in gray above the exact large bold text \"Good morning, Sam\" in white; on the right, a circular profile avatar showing a friendly illustrated face.",
68
+ "coordinates": "cx: 0.500, cy: 0.090, w: 0.880, h: 0.070",
69
+ "hierarchy_and_relation": "Below the status bar, above the activity card; text left-aligned, avatar right-aligned.",
70
+ "color_specs": ["#FFFFFF", "#8A90A0", "#F2A65A"]
71
+ },
72
+ {
73
+ "description": "Activity card: a large dark charcoal rounded card. On the left, three concentric progress rings: an outer lime-green ring about 80% full, a middle electric-blue ring about 60% full, and an inner coral ring about 90% full. On the right, a vertical legend of three rows: a lime dot with the exact text \"Move 420 / 520 kcal\", a blue dot with the exact text \"Exercise 18 / 30 min\", and a coral dot with the exact text \"Stand 11 / 12 h\", all in white with the numbers in bold.",
74
+ "coordinates": "cx: 0.500, cy: 0.250, w: 0.880, h: 0.200",
75
+ "hierarchy_and_relation": "Below the greeting header, full content width; contains the rings on the left and the legend on the right.",
76
+ "color_specs": ["#1C1F27", "#B6F23A", "#3A8DFF", "#FF6B5A", "#FFFFFF"]
77
+ },
78
+ {
79
+ "description": "Stat card 1 of 2 in a row, left column: a dark charcoal rounded card with a small lime footprints icon, the exact label \"Steps\" in gray, and the exact large bold number \"8,342\" in white.",
80
+ "coordinates": "cx: 0.275, cy: 0.440, w: 0.430, h: 0.120",
81
+ "hierarchy_and_relation": "Left card of a two-card row below the activity card; same size and top alignment as the right card.",
82
+ "color_specs": ["#1C1F27", "#B6F23A", "#8A90A0", "#FFFFFF"]
83
+ },
84
+ {
85
+ "description": "Stat card 2 of 2 in a row, right column: a dark charcoal rounded card with a small coral heart icon, the exact label \"Heart rate\" in gray, and the exact large bold number \"72 bpm\" in white.",
86
+ "coordinates": "cx: 0.725, cy: 0.440, w: 0.430, h: 0.120",
87
+ "hierarchy_and_relation": "Right card of the two-card row below the activity card; same size and top alignment as the left card.",
88
+ "color_specs": ["#1C1F27", "#FF6B5A", "#8A90A0", "#FFFFFF"]
89
+ },
90
+ {
91
+ "description": "Weekly chart card: a dark charcoal rounded card with the exact title \"This week\" in bold white at the top left. Below it, seven vertical rounded bars in electric blue with varying heights, each labeled underneath in gray with one weekday letter, in order: \"M\", \"T\", \"W\", \"T\", \"F\", \"S\", \"S\". The fourth bar is the tallest and highlighted in lime green.",
92
+ "coordinates": "cx: 0.500, cy: 0.640, w: 0.880, h: 0.230",
93
+ "hierarchy_and_relation": "Below the two stat cards, full content width; the seven bars are evenly spaced inside the card.",
94
+ "color_specs": ["#1C1F27", "#3A8DFF", "#B6F23A", "#8A90A0", "#FFFFFF"]
95
+ },
96
+ {
97
+ "description": "Primary action button: a full-width lime-green rounded pill with the exact label \"Start Workout\" in bold near-black text and a small play icon on its left.",
98
+ "coordinates": "cx: 0.500, cy: 0.825, w: 0.880, h: 0.055",
99
+ "hierarchy_and_relation": "Below the weekly chart card, above the tab bar.",
100
+ "color_specs": ["#B6F23A", "#0D0F14"]
101
+ },
102
+ {
103
+ "description": "Bottom tab bar: a dark charcoal bar with four evenly spaced tabs, each a line icon above a label: a house icon labeled \"Home\" highlighted in lime, a running-figure icon labeled \"Workouts\", a bar-chart icon labeled \"Stats\", and a person icon labeled \"Profile\", the last three in gray. A thin white home indicator line is centered at the very bottom.",
104
+ "coordinates": "cx: 0.500, cy: 0.950, w: 1.000, h: 0.100",
105
+ "hierarchy_and_relation": "Full-width bar along the bottom edge of the screen, below the primary action button.",
106
+ "color_specs": ["#1C1F27", "#B6F23A", "#8A90A0", "#FFFFFF"]
107
+ }
108
+ ]
109
+ }
110
+ """
111
+
112
+ image = pipe(prompt=prompt, height=2560, width=1440, generator=torch.Generator().manual_seed(2024), output="images")[0]
113
+ image.save("phone_app.png")
114
+ ```
connector/config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "MingImageConnector",
3
+ "_diffusers_version": "0.41.0.dev0",
4
+ "auto_map": {
5
+ "AutoModel": "connector_ming_image.MingImageConnector"
6
+ },
7
+ "cond_dim": 2560,
8
+ "hidden_state_layers": [
9
+ 5,
10
+ 12,
11
+ 20
12
+ ],
13
+ "inner_dim": 3840,
14
+ "qwen2_config": {
15
+ "attention_dropout": 0.0,
16
+ "hidden_act": "silu",
17
+ "hidden_size": 1536,
18
+ "intermediate_size": 8960,
19
+ "max_position_embeddings": 32768,
20
+ "num_attention_heads": 12,
21
+ "num_hidden_layers": 28,
22
+ "num_key_value_heads": 2,
23
+ "rms_norm_eps": 1e-06,
24
+ "rope_theta": 1000000.0
25
+ },
26
+ "text_hidden_size": 2048
27
+ }
connector/connector_ming_image.py ADDED
@@ -0,0 +1,97 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright 2026 inclusionAI and The HuggingFace Team. All rights reserved.
2
+ #
3
+ # Licensed under the MIT License. See the LICENSE file in this repository.
4
+
5
+ import torch
6
+ import torch.nn as nn
7
+ from transformers import Qwen2Config
8
+ from transformers.models.qwen2.modeling_qwen2 import Qwen2DecoderLayer, Qwen2RMSNorm, Qwen2RotaryEmbedding
9
+
10
+ from diffusers.configuration_utils import ConfigMixin, register_to_config
11
+ from diffusers.models.modeling_utils import ModelMixin
12
+ from diffusers.models.normalization import RMSNorm
13
+
14
+
15
+ class MingImageConnector(ModelMixin, ConfigMixin):
16
+ r"""
17
+ Turns Ming-Image text-encoder hidden states into the two conditioning streams of the transformer.
18
+
19
+ The query-token hidden states go through `proj_in`, a bidirectional Qwen2 stack and `proj_out` to give
20
+ `prompt_embeds`. The text-token hidden states of the selected encoder layers, concatenated along the feature axis,
21
+ go through `proj_directvlm` to give `prompt_embeds_2`.
22
+
23
+ Args:
24
+ text_hidden_size (`int`, defaults to `2048`): Hidden size of the text encoder.
25
+ hidden_state_layers (`list[int]`, defaults to `[5, 12, 20]`):
26
+ Indices into the text encoder's hidden states (0 = input embeddings) whose text-token states are
27
+ concatenated into the text stream.
28
+ cond_dim (`int`, defaults to `2560`): Width of `prompt_embeds`.
29
+ inner_dim (`int`, defaults to `3840`): Width of `prompt_embeds_2` (the transformer's model width).
30
+ qwen2_config (`dict`): Config of the Qwen2 stack, as accepted by `transformers.Qwen2Config`.
31
+ """
32
+
33
+ _no_split_modules = ["Qwen2DecoderLayer"]
34
+
35
+ @register_to_config
36
+ def __init__(
37
+ self,
38
+ text_hidden_size: int = 2048,
39
+ hidden_state_layers: list[int] = [5, 12, 20],
40
+ cond_dim: int = 2560,
41
+ inner_dim: int = 3840,
42
+ qwen2_config: dict | None = None,
43
+ ):
44
+ super().__init__()
45
+ qwen2_config = Qwen2Config(**qwen2_config)
46
+ qwen2_config._attn_implementation = "sdpa"
47
+
48
+ self.proj_in = nn.Linear(text_hidden_size, qwen2_config.hidden_size)
49
+ self.layers = nn.ModuleList(
50
+ [Qwen2DecoderLayer(qwen2_config, layer_idx) for layer_idx in range(qwen2_config.num_hidden_layers)]
51
+ )
52
+ for layer in self.layers:
53
+ # the connector attends bidirectionally over the query tokens
54
+ layer.self_attn.is_causal = False
55
+ self.norm = Qwen2RMSNorm(qwen2_config.hidden_size, eps=qwen2_config.rms_norm_eps)
56
+ self.rotary_emb = Qwen2RotaryEmbedding(config=qwen2_config)
57
+ self.proj_out = nn.Linear(qwen2_config.hidden_size, cond_dim)
58
+
59
+ directvlm_dim = text_hidden_size * len(hidden_state_layers)
60
+ self.proj_directvlm = nn.Sequential(RMSNorm(directvlm_dim, eps=1e-5), nn.Linear(directvlm_dim, inner_dim))
61
+
62
+ def forward(
63
+ self, query_hidden_states: torch.Tensor, text_hidden_states: torch.Tensor
64
+ ) -> tuple[torch.Tensor, torch.Tensor]:
65
+ r"""
66
+ Args:
67
+ query_hidden_states (`torch.Tensor`):
68
+ Last-layer text-encoder hidden states at the query-token positions, of shape
69
+ `(batch_size, num_query_tokens, text_hidden_size)`.
70
+ text_hidden_states (`torch.Tensor`):
71
+ Text-encoder hidden states at the text positions for each selected layer, concatenated along the last
72
+ axis, of shape `(batch_size, text_seq_len, text_hidden_size * len(hidden_state_layers))`.
73
+
74
+ Returns:
75
+ `tuple[torch.Tensor, torch.Tensor]`:
76
+ `prompt_embeds` of shape `(batch_size, num_query_tokens, cond_dim)` and `prompt_embeds_2` of shape
77
+ `(batch_size, text_seq_len, inner_dim)`.
78
+ """
79
+ hidden_states = self.proj_in(query_hidden_states)
80
+ batch_size, seq_len, _ = hidden_states.shape
81
+ position_ids = torch.arange(seq_len, device=hidden_states.device)[None]
82
+ position_embeddings = self.rotary_emb(hidden_states, position_ids)
83
+ # The original model adds an all-ones mask to the attention scores. It leaves the softmax unchanged
84
+ # mathematically but changes its rounding, so it is kept to reproduce the original outputs exactly.
85
+ attention_mask = hidden_states.new_ones(batch_size, 1, seq_len, seq_len)
86
+ for layer in self.layers:
87
+ hidden_states = layer(
88
+ hidden_states,
89
+ attention_mask=attention_mask,
90
+ position_ids=position_ids,
91
+ position_embeddings=position_embeddings,
92
+ )
93
+ hidden_states = self.norm(hidden_states)
94
+ prompt_embeds = self.proj_out(hidden_states)
95
+
96
+ prompt_embeds_2 = self.proj_directvlm(text_hidden_states)
97
+ return prompt_embeds, prompt_embeds_2
connector/diffusion_pytorch_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:019f7dd8c1c1c7464442d6e4a5690f70a6c4fc33bf8769c7af0918026da3d867
3
+ size 5364138856
scheduler/scheduler_config.json ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "FlowMatchEulerDiscreteScheduler",
3
+ "_diffusers_version": "0.41.0.dev0",
4
+ "base_image_seq_len": 256,
5
+ "base_shift": 0.5,
6
+ "invert_sigmas": false,
7
+ "max_image_seq_len": 4096,
8
+ "max_shift": 1.15,
9
+ "num_train_timesteps": 1000,
10
+ "shift": 6.0,
11
+ "shift_terminal": null,
12
+ "stochastic_sampling": false,
13
+ "time_shift_type": "exponential",
14
+ "use_beta_sigmas": false,
15
+ "use_dynamic_shifting": false,
16
+ "use_exponential_sigmas": false,
17
+ "use_karras_sigmas": false
18
+ }
text_encoder/config.json ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "MingImageTextEncoder",
3
+ "_diffusers_version": "0.41.0.dev0",
4
+ "auto_map": {
5
+ "AutoModel": "text_encoder_ming_image.MingImageTextEncoder"
6
+ },
7
+ "first_k_dense_replace": 1,
8
+ "head_dim": 128,
9
+ "hidden_size": 2048,
10
+ "image_end_token": 157159,
11
+ "image_patch_token": 157157,
12
+ "image_start_token": 157158,
13
+ "intermediate_size": 5120,
14
+ "moe_intermediate_size": 512,
15
+ "mrope_section": [
16
+ 8,
17
+ 12,
18
+ 12
19
+ ],
20
+ "n_group": 8,
21
+ "num_attention_heads": 16,
22
+ "num_experts": 256,
23
+ "num_experts_per_tok": 8,
24
+ "num_hidden_layers": 20,
25
+ "num_key_value_heads": 4,
26
+ "num_query_tokens": 256,
27
+ "num_shared_experts": 1,
28
+ "pad_token_id": 156892,
29
+ "partial_rotary_factor": 0.5,
30
+ "rms_norm_eps": 1e-06,
31
+ "rope_theta": 600000,
32
+ "rope_time_scale": 2.0,
33
+ "routed_scaling_factor": 2.5,
34
+ "spatial_merge_size": 2,
35
+ "topk_group": 4,
36
+ "vocab_size": 157184
37
+ }
text_encoder/diffusion_pytorch_model-00001-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1a6d7e79f73cb6631ed0996485d1869718d6cd1a7a928d434137333edb10e94a
3
+ size 9999799456
text_encoder/diffusion_pytorch_model-00002-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:242d52d967b6b5e922c1602a764c1909ad439b350c35676a49fbe741cbd45acf
3
+ size 9999841200
text_encoder/diffusion_pytorch_model-00003-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6f58ae41919d7e0d222940b137a4f5c402415aab41001fbf555250aa75206a9e
3
+ size 9999843792
text_encoder/diffusion_pytorch_model-00004-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8b6aed50c3c56dd2d3288ae62c358ac18b2763a8bb99f66b0ce59cd1d2a694dc
3
+ size 1891759504
text_encoder/diffusion_pytorch_model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
text_encoder/text_encoder_ming_image.py ADDED
@@ -0,0 +1,403 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright 2026 inclusionAI and The HuggingFace Team. All rights reserved.
2
+ #
3
+ # Licensed under the MIT License. See the LICENSE file in this repository.
4
+
5
+ import math
6
+
7
+ import torch
8
+ import torch.nn as nn
9
+ import torch.nn.functional as F
10
+
11
+ from diffusers.configuration_utils import ConfigMixin, register_to_config
12
+ from diffusers.models.modeling_utils import ModelMixin
13
+
14
+
15
+ class MingImageRMSNorm(nn.Module):
16
+ def __init__(self, hidden_size: int, eps: float = 1e-6):
17
+ super().__init__()
18
+ self.weight = nn.Parameter(torch.ones(hidden_size))
19
+ self.variance_epsilon = eps
20
+
21
+ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
22
+ input_dtype = hidden_states.dtype
23
+ hidden_states = hidden_states.to(torch.float32)
24
+ variance = hidden_states.pow(2).mean(-1, keepdim=True)
25
+ hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
26
+ return self.weight * hidden_states.to(input_dtype)
27
+
28
+
29
+ def rotate_half(x: torch.Tensor) -> torch.Tensor:
30
+ x1 = x[..., : x.shape[-1] // 2]
31
+ x2 = x[..., x.shape[-1] // 2 :]
32
+ return torch.cat((-x2, x1), dim=-1)
33
+
34
+
35
+ def video_rope_cos_sin(cos: torch.Tensor, sin: torch.Tensor, mrope_section: list[int]):
36
+ """Interleave the (frame, height, width) rotary tables into the "video_rope" channel layout.
37
+
38
+ `cos` / `sin` are `(3, batch, seq_len, rotary_dim)` for the frame, height and width axes. Height and width
39
+ alternate channel by channel, and the frame axis takes the last channels of each half.
40
+ """
41
+ sections = [mrope_section[0], mrope_section[1] + mrope_section[2]] * 2
42
+ index = 0
43
+ result_cos, result_sin = [], []
44
+ for i, section in enumerate(sections[::-1]):
45
+ if i % 2 == 0:
46
+ for j in range(section):
47
+ row = 1 if j % 2 == 0 else 2
48
+ result_cos.append(cos[row, ..., index : index + 1])
49
+ result_sin.append(sin[row, ..., index : index + 1])
50
+ index += 1
51
+ else:
52
+ result_cos.append(cos[0, ..., index : index + section])
53
+ result_sin.append(sin[0, ..., index : index + section])
54
+ index += section
55
+ return torch.cat(result_cos, dim=-1), torch.cat(result_sin, dim=-1)
56
+
57
+
58
+ class MingImageRotaryEmbedding(nn.Module):
59
+ def __init__(self, rotary_dim: int, theta: float):
60
+ super().__init__()
61
+ inv_freq = 1.0 / (theta ** (torch.arange(0, rotary_dim, 2, dtype=torch.int64).float() / rotary_dim))
62
+ self.register_buffer("inv_freq", inv_freq, persistent=False)
63
+
64
+ @torch.no_grad()
65
+ def forward(self, x: torch.Tensor, position_ids: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
66
+ # position_ids: (3, batch, seq_len) for the frame, height and width axes
67
+ inv_freq = self.inv_freq[None, None, :, None].float().expand(3, position_ids.shape[1], -1, 1).to(x.device)
68
+ position_ids = position_ids[:, :, None, :].float()
69
+ with torch.autocast(device_type=x.device.type if x.device.type != "mps" else "cpu", enabled=False):
70
+ freqs = (inv_freq @ position_ids).transpose(2, 3)
71
+ emb = torch.cat((freqs, freqs), dim=-1)
72
+ return emb.cos().to(x.dtype), emb.sin().to(x.dtype)
73
+
74
+
75
+ class MingImageMLP(nn.Module):
76
+ def __init__(self, hidden_size: int, intermediate_size: int):
77
+ super().__init__()
78
+ self.gate_proj = nn.Linear(hidden_size, intermediate_size, bias=False)
79
+ self.up_proj = nn.Linear(hidden_size, intermediate_size, bias=False)
80
+ self.down_proj = nn.Linear(intermediate_size, hidden_size, bias=False)
81
+
82
+ def forward(self, x: torch.Tensor) -> torch.Tensor:
83
+ return self.down_proj(F.silu(self.gate_proj(x)) * self.up_proj(x))
84
+
85
+
86
+ class MingImageGate(nn.Module):
87
+ """Sigmoid router with expert bias and group-limited top-k selection."""
88
+
89
+ def __init__(self, hidden_size, num_experts, top_k, n_group, topk_group, routed_scaling_factor):
90
+ super().__init__()
91
+ self.top_k = top_k
92
+ self.num_experts = num_experts
93
+ self.n_group = n_group
94
+ self.topk_group = topk_group
95
+ self.routed_scaling_factor = routed_scaling_factor
96
+ self.weight = nn.Parameter(torch.zeros((num_experts, hidden_size)))
97
+ self.expert_bias = nn.Parameter(torch.zeros(num_experts), requires_grad=False)
98
+
99
+ def forward(self, hidden_states: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
100
+ hidden_states = hidden_states.view(-1, hidden_states.shape[-1])
101
+ logits = F.linear(hidden_states.float(), self.weight.float())
102
+ scores = torch.sigmoid(logits)
103
+
104
+ # pick the top `topk_group` groups by the sum of their two best (bias-corrected) scores, then the top-k
105
+ # experts inside those groups
106
+ scores_for_routing = scores + self.expert_bias
107
+ num_tokens = scores_for_routing.shape[0]
108
+ group_scores = scores_for_routing.view(num_tokens, self.n_group, -1).topk(2, dim=-1)[0].sum(dim=-1)
109
+ group_idx = torch.topk(group_scores, k=self.topk_group, dim=-1, sorted=False)[1]
110
+ group_mask = torch.zeros_like(group_scores).scatter_(1, group_idx, 1)
111
+ score_mask = (
112
+ group_mask.unsqueeze(-1)
113
+ .expand(num_tokens, self.n_group, self.num_experts // self.n_group)
114
+ .reshape(num_tokens, -1)
115
+ )
116
+ masked_scores = scores_for_routing.masked_fill(~score_mask.bool(), float("-inf"))
117
+ topk_idx = torch.topk(masked_scores, k=self.top_k, dim=-1, sorted=False)[1]
118
+
119
+ topk_weight = torch.gather(scores, dim=1, index=topk_idx).type_as(logits)
120
+ topk_weight = topk_weight / (topk_weight.sum(dim=-1, keepdim=True) + 1e-20)
121
+ return topk_idx, topk_weight * self.routed_scaling_factor
122
+
123
+
124
+ class MingImageSparseMoeBlock(nn.Module):
125
+ """Routed experts plus a shared expert. Tokens at image positions are routed by `image_gate`."""
126
+
127
+ def __init__(self, config):
128
+ super().__init__()
129
+ gate_kwargs = {
130
+ "hidden_size": config.hidden_size,
131
+ "num_experts": config.num_experts,
132
+ "top_k": config.num_experts_per_tok,
133
+ "n_group": config.n_group,
134
+ "topk_group": config.topk_group,
135
+ "routed_scaling_factor": config.routed_scaling_factor,
136
+ }
137
+ self.gate = MingImageGate(**gate_kwargs)
138
+ self.image_gate = MingImageGate(**gate_kwargs)
139
+ self.experts = nn.ModuleList(
140
+ [MingImageMLP(config.hidden_size, config.moe_intermediate_size) for _ in range(config.num_experts)]
141
+ )
142
+ self.shared_experts = MingImageMLP(
143
+ config.hidden_size, config.moe_intermediate_size * config.num_shared_experts
144
+ )
145
+
146
+ def forward(self, hidden_states: torch.Tensor, image_mask: torch.Tensor) -> torch.Tensor:
147
+ batch_size, seq_len, hidden_size = hidden_states.shape
148
+ topk_idx, topk_weight = self.gate(hidden_states)
149
+ image_topk_idx, image_topk_weight = self.image_gate(hidden_states)
150
+ image_mask = image_mask.view(-1, 1)
151
+ topk_idx = torch.where(image_mask, image_topk_idx, topk_idx)
152
+ topk_weight = torch.where(image_mask, image_topk_weight, topk_weight)
153
+
154
+ routed = self.moe_infer(hidden_states.view(-1, hidden_size), topk_idx, topk_weight)
155
+ return routed.view(batch_size, seq_len, hidden_size) + self.shared_experts(hidden_states)
156
+
157
+ def moe_infer(self, x: torch.Tensor, topk_ids: torch.Tensor, topk_weight: torch.Tensor) -> torch.Tensor:
158
+ tokens_per_expert = torch.bincount(topk_ids.view(-1), minlength=len(self.experts)).tolist()
159
+ idxs = topk_ids.view(-1).argsort()
160
+ sorted_tokens = x[idxs // topk_ids.shape[1]]
161
+ outputs = []
162
+ start_idx = 0
163
+ for expert, num_tokens in zip(self.experts, tokens_per_expert):
164
+ if num_tokens == 0:
165
+ continue
166
+ outputs.append(expert(sorted_tokens[start_idx : start_idx + num_tokens]))
167
+ start_idx += num_tokens
168
+
169
+ outs = torch.cat(outputs, dim=0)
170
+ new_x = torch.empty_like(outs)
171
+ new_x[idxs] = outs
172
+ return (
173
+ new_x.view(*topk_ids.shape, -1)
174
+ .type(topk_weight.dtype)
175
+ .mul_(topk_weight.unsqueeze(dim=-1))
176
+ .sum(dim=1)
177
+ .type(new_x.dtype)
178
+ )
179
+
180
+
181
+ class MingImageTextAttention(nn.Module):
182
+ def __init__(self, config):
183
+ super().__init__()
184
+ self.num_heads = config.num_attention_heads
185
+ self.num_kv_heads = config.num_key_value_heads
186
+ self.head_dim = config.head_dim
187
+ self.mrope_section = config.mrope_section
188
+
189
+ self.query_key_value = nn.Linear(
190
+ config.hidden_size, (self.num_heads + 2 * self.num_kv_heads) * self.head_dim, bias=False
191
+ )
192
+ self.q_norm = MingImageRMSNorm(self.head_dim, eps=config.rms_norm_eps)
193
+ self.k_norm = MingImageRMSNorm(self.head_dim, eps=config.rms_norm_eps)
194
+ self.dense = nn.Linear(self.num_heads * self.head_dim, config.hidden_size, bias=False)
195
+
196
+ def forward(
197
+ self,
198
+ hidden_states: torch.Tensor,
199
+ attention_mask: torch.Tensor,
200
+ position_embeddings: tuple[torch.Tensor, torch.Tensor],
201
+ ) -> torch.Tensor:
202
+ batch_size, seq_len, _ = hidden_states.shape
203
+ qkv = self.query_key_value(hidden_states).view(
204
+ batch_size, seq_len, self.num_heads + 2 * self.num_kv_heads, self.head_dim
205
+ )
206
+ query, key, value = qkv.split([self.num_heads, self.num_kv_heads, self.num_kv_heads], dim=2)
207
+ query = self.q_norm(query)
208
+ key = self.k_norm(key)
209
+
210
+ # partial rotary: only the first `rotary_dim` channels of each head are rotated
211
+ cos, sin = video_rope_cos_sin(*position_embeddings, self.mrope_section)
212
+ cos, sin = cos.unsqueeze(2), sin.unsqueeze(2)
213
+ rotary_dim = cos.shape[-1]
214
+ q_rot, q_pass = query[..., :rotary_dim], query[..., rotary_dim:]
215
+ k_rot, k_pass = key[..., :rotary_dim], key[..., rotary_dim:]
216
+ query = torch.cat([q_rot * cos + rotate_half(q_rot) * sin, q_pass], dim=-1)
217
+ key = torch.cat([k_rot * cos + rotate_half(k_rot) * sin, k_pass], dim=-1)
218
+
219
+ # (batch, heads, seq_len, head_dim), with the key/value heads repeated for grouped-query attention
220
+ query = query.transpose(1, 2)
221
+ key = key.repeat_interleave(self.num_heads // self.num_kv_heads, dim=2).transpose(1, 2)
222
+ value = value.repeat_interleave(self.num_heads // self.num_kv_heads, dim=2).transpose(1, 2)
223
+
224
+ # Explicit attention matching the original model's default ("eager") implementation operation for operation.
225
+ # A fused kernel changes the rounding, which is enough to flip some near-tied expert choices in the MoE
226
+ # layers; at the text encoder's sequence lengths the explicit form costs next to nothing.
227
+ attn_weights = torch.matmul(query / math.sqrt(self.head_dim), key.transpose(2, 3)) + attention_mask
228
+ attn_weights = torch.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
229
+ hidden_states = torch.matmul(attn_weights, value).transpose(1, 2)
230
+ return self.dense(hidden_states.flatten(2, 3))
231
+
232
+
233
+ class MingImageTextDecoderLayer(nn.Module):
234
+ def __init__(self, config, layer_idx: int):
235
+ super().__init__()
236
+ self.attention = MingImageTextAttention(config)
237
+ self.is_moe = layer_idx >= config.first_k_dense_replace
238
+ self.mlp = (
239
+ MingImageSparseMoeBlock(config)
240
+ if self.is_moe
241
+ else MingImageMLP(config.hidden_size, config.intermediate_size)
242
+ )
243
+ self.input_layernorm = MingImageRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
244
+ self.post_attention_layernorm = MingImageRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
245
+
246
+ def forward(self, hidden_states, attention_mask, position_embeddings, image_mask) -> torch.Tensor:
247
+ hidden_states = hidden_states + self.attention(
248
+ self.input_layernorm(hidden_states), attention_mask, position_embeddings
249
+ )
250
+ residual = hidden_states
251
+ hidden_states = self.post_attention_layernorm(hidden_states)
252
+ hidden_states = self.mlp(hidden_states, image_mask) if self.is_moe else self.mlp(hidden_states)
253
+ return residual + hidden_states
254
+
255
+
256
+ class MingImageTextEncoder(ModelMixin, ConfigMixin):
257
+ r"""
258
+ The Ming-Image text encoder: the language model of the Ming multimodal LLM (a BailingMoeV2 mixture-of-experts
259
+ decoder with a separate router for image positions and 3D "video_rope" positions), together with the learnable
260
+ image-generation query tokens it is prompted with.
261
+
262
+ The query tokens are placed in `input_ids` as a block of `image_patch_token` ids between `image_start_token` and
263
+ `image_end_token`. They are treated as a `1 x num_query_tokens` image: they are routed by the image router and get
264
+ image positions.
265
+
266
+ Args:
267
+ vocab_size (`int`): Vocabulary size.
268
+ hidden_size (`int`): Hidden size.
269
+ intermediate_size (`int`): FFN width of the dense layers.
270
+ moe_intermediate_size (`int`): FFN width of each expert.
271
+ num_hidden_layers (`int`): Number of decoder layers.
272
+ num_attention_heads (`int`): Number of query heads.
273
+ num_key_value_heads (`int`): Number of key/value heads.
274
+ head_dim (`int`): Head dimension.
275
+ rms_norm_eps (`float`): RMSNorm epsilon.
276
+ rope_theta (`float`): RoPE base.
277
+ partial_rotary_factor (`float`): Fraction of each head's channels that are rotated.
278
+ mrope_section (`list[int]`): Rotary channels of the frame, height and width axes.
279
+ rope_time_scale (`float`): Frame-axis position step of image tokens.
280
+ spatial_merge_size (`int`): Merge factor between an image grid and its tokens.
281
+ num_experts (`int`): Number of routed experts.
282
+ num_experts_per_tok (`int`): Experts per token.
283
+ n_group (`int`): Number of expert groups for group-limited routing.
284
+ topk_group (`int`): Number of groups a token may route to.
285
+ routed_scaling_factor (`float`): Scale applied to the routing weights.
286
+ num_shared_experts (`int`): Width multiplier of the shared expert.
287
+ first_k_dense_replace (`int`): Number of leading layers with a dense FFN instead of experts.
288
+ pad_token_id (`int`): Padding token id.
289
+ image_patch_token (`int`): Placeholder id of image (and query) tokens.
290
+ image_start_token (`int`): Id that opens an image block.
291
+ image_end_token (`int`): Id that closes an image block.
292
+ num_query_tokens (`int`): Number of image-generation query tokens.
293
+ """
294
+
295
+ _no_split_modules = ["MingImageTextDecoderLayer"]
296
+
297
+ @register_to_config
298
+ def __init__(
299
+ self,
300
+ vocab_size: int = 157184,
301
+ hidden_size: int = 2048,
302
+ intermediate_size: int = 5120,
303
+ moe_intermediate_size: int = 512,
304
+ num_hidden_layers: int = 20,
305
+ num_attention_heads: int = 16,
306
+ num_key_value_heads: int = 4,
307
+ head_dim: int = 128,
308
+ rms_norm_eps: float = 1e-6,
309
+ rope_theta: float = 600000.0,
310
+ partial_rotary_factor: float = 0.5,
311
+ mrope_section: list[int] = [8, 12, 12],
312
+ rope_time_scale: float = 2.0,
313
+ spatial_merge_size: int = 2,
314
+ num_experts: int = 256,
315
+ num_experts_per_tok: int = 8,
316
+ n_group: int = 8,
317
+ topk_group: int = 4,
318
+ routed_scaling_factor: float = 2.5,
319
+ num_shared_experts: int = 1,
320
+ first_k_dense_replace: int = 1,
321
+ pad_token_id: int = 156892,
322
+ image_patch_token: int = 157157,
323
+ image_start_token: int = 157158,
324
+ image_end_token: int = 157159,
325
+ num_query_tokens: int = 256,
326
+ ):
327
+ super().__init__()
328
+ self.word_embeddings = nn.Embedding(vocab_size, hidden_size, pad_token_id)
329
+ self.query_tokens = nn.Parameter(torch.zeros(num_query_tokens, hidden_size))
330
+ self.layers = nn.ModuleList(
331
+ [MingImageTextDecoderLayer(self.config, layer_idx) for layer_idx in range(num_hidden_layers)]
332
+ )
333
+ self.norm = MingImageRMSNorm(hidden_size, eps=rms_norm_eps)
334
+ self.rotary_emb = MingImageRotaryEmbedding(int(head_dim * partial_rotary_factor), rope_theta)
335
+
336
+ def get_position_ids(self, input_ids: torch.Tensor, attention_mask: torch.Tensor) -> torch.Tensor:
337
+ """3D (frame, height, width) positions: text tokens advance all three axes together; each query-token block
338
+ is laid out as a `1 x num_query_tokens` image grid, centred and offset by the text before it."""
339
+ position_ids = torch.ones(3, *input_ids.shape, dtype=input_ids.dtype, device=input_ids.device)
340
+ query_grid_w = self.config.num_query_tokens
341
+ for i in range(input_ids.shape[0]):
342
+ tokens = input_ids[i][attention_mask[i] == 1].tolist()
343
+ num_images = sum(
344
+ 1
345
+ for idx, token in enumerate(tokens[:-1])
346
+ if token == self.config.image_start_token and tokens[idx + 1] == self.config.image_patch_token
347
+ )
348
+ positions = []
349
+ start = 0
350
+ for _ in range(num_images):
351
+ end = tokens.index(self.config.image_patch_token, start)
352
+ text_len = end - start
353
+ st_idx = positions[-1][0].max() + 1 if positions else 0
354
+ positions.append(torch.arange(text_len).view(1, -1).expand(3, -1) + st_idx)
355
+
356
+ t_index = torch.zeros(query_grid_w, dtype=torch.long) * self.config.rope_time_scale + text_len + st_idx
357
+ h_index = torch.zeros(query_grid_w, dtype=torch.long) + t_index
358
+ w_index = torch.arange(query_grid_w) - (query_grid_w - 1) // 2 + t_index
359
+ positions.append(torch.stack([t_index, h_index, w_index]))
360
+ start = end + query_grid_w
361
+
362
+ if start < len(tokens):
363
+ st_idx = positions[-1][0].max() + 1 if positions else 0
364
+ positions.append(torch.arange(len(tokens) - start).view(1, -1).expand(3, -1) + st_idx)
365
+
366
+ position_ids[..., i, attention_mask[i] == 1] = torch.cat(positions, dim=1).to(position_ids)
367
+ return position_ids
368
+
369
+ def forward(self, input_ids: torch.Tensor, attention_mask: torch.Tensor) -> tuple[torch.Tensor, ...]:
370
+ r"""
371
+ Args:
372
+ input_ids (`torch.Tensor`):
373
+ Token ids of shape `(batch_size, seq_len)`, with each item's query-token block already inserted.
374
+ attention_mask (`torch.Tensor`):
375
+ Padding mask of shape `(batch_size, seq_len)`, 1 for real tokens.
376
+
377
+ Returns:
378
+ `tuple[torch.Tensor, ...]`:
379
+ `num_hidden_layers + 1` hidden states of shape `(batch_size, seq_len, hidden_size)`: the input
380
+ embeddings, the outputs of layers 1 to `num_hidden_layers - 1`, and the final normalized output.
381
+ """
382
+ image_mask = input_ids == self.config.image_patch_token
383
+ inputs_embeds = self.word_embeddings(input_ids.clip(0, self.config.vocab_size - 1))
384
+ query_tokens = self.query_tokens.to(inputs_embeds.dtype).repeat(input_ids.shape[0], 1)
385
+ inputs_embeds = inputs_embeds.masked_scatter(image_mask.unsqueeze(-1), query_tokens)
386
+
387
+ position_ids = self.get_position_ids(input_ids, attention_mask)
388
+ position_embeddings = self.rotary_emb(inputs_embeds, position_ids)
389
+
390
+ # Additive causal mask that also excludes padded keys, (batch, 1, seq_len, seq_len)
391
+ seq_len = input_ids.shape[1]
392
+ attend = torch.ones(seq_len, seq_len, dtype=torch.bool, device=input_ids.device).tril()[None, None]
393
+ attend = attend & attention_mask.bool()[:, None, None, :]
394
+ mask = torch.zeros(attend.shape, dtype=inputs_embeds.dtype, device=input_ids.device)
395
+ mask = mask.masked_fill(~attend, torch.finfo(inputs_embeds.dtype).min)
396
+
397
+ hidden_states = inputs_embeds
398
+ all_hidden_states = ()
399
+ for layer in self.layers:
400
+ all_hidden_states += (hidden_states,)
401
+ hidden_states = layer(hidden_states, mask, position_embeddings, image_mask)
402
+ all_hidden_states += (self.norm(hidden_states),)
403
+ return all_hidden_states
tokenizer/special_tokens_map.json ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|arithmetic_end|>",
4
+ "<role>",
5
+ "<|number_end|>",
6
+ "<|arithmetic_start|>",
7
+ "</role>",
8
+ "<|number_start|>",
9
+ "<|role_end|>",
10
+ "<tool_call>",
11
+ "</tool_call>",
12
+ "<tool_response>",
13
+ "</tool_response>"
14
+ ],
15
+ "bos_token": {
16
+ "content": "<|startoftext|>",
17
+ "lstrip": false,
18
+ "normalized": false,
19
+ "rstrip": false,
20
+ "single_word": false
21
+ },
22
+ "cls_token": {
23
+ "content": "[CLS]",
24
+ "lstrip": false,
25
+ "normalized": false,
26
+ "rstrip": false,
27
+ "single_word": false
28
+ },
29
+ "eos_token": {
30
+ "content": "<|role_end|>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false
35
+ },
36
+ "gmask_token": {
37
+ "content": "[gMASK]",
38
+ "lstrip": false,
39
+ "normalized": false,
40
+ "rstrip": false,
41
+ "single_word": false
42
+ },
43
+ "pad_token": "<|role_end|>"
44
+ }
tokenizer/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e7ff01708d504f7bf4dbf7f5815adde57bab9a40e7f563ab6ad1acace4464917
3
+ size 12210709
tokenizer/tokenizer_config.json ADDED
@@ -0,0 +1,2334 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_eos_token": false,
4
+ "added_tokens_decoder": {
5
+ "156891": {
6
+ "content": "<|startoftext|>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "156892": {
14
+ "content": "<|endoftext|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "156893": {
22
+ "content": "[CLS]",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "156894": {
30
+ "content": "[gMASK]",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "156895": {
38
+ "content": "<|role_end|>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false,
43
+ "special": true
44
+ },
45
+ "156896": {
46
+ "content": "<tool_call>",
47
+ "lstrip": false,
48
+ "normalized": false,
49
+ "rstrip": false,
50
+ "single_word": false,
51
+ "special": true
52
+ },
53
+ "156897": {
54
+ "content": "</tool_call>",
55
+ "lstrip": false,
56
+ "normalized": false,
57
+ "rstrip": false,
58
+ "single_word": false,
59
+ "special": true
60
+ },
61
+ "156898": {
62
+ "content": "<tool_response>",
63
+ "lstrip": false,
64
+ "normalized": false,
65
+ "rstrip": false,
66
+ "single_word": false,
67
+ "special": true
68
+ },
69
+ "156899": {
70
+ "content": "</tool_response>",
71
+ "lstrip": false,
72
+ "normalized": false,
73
+ "rstrip": false,
74
+ "single_word": false,
75
+ "special": true
76
+ },
77
+ "156900": {
78
+ "content": "<|reserved_token_5|>",
79
+ "lstrip": false,
80
+ "normalized": false,
81
+ "rstrip": false,
82
+ "single_word": false,
83
+ "special": true
84
+ },
85
+ "156901": {
86
+ "content": "<|reserved_token_6|>",
87
+ "lstrip": false,
88
+ "normalized": false,
89
+ "rstrip": false,
90
+ "single_word": false,
91
+ "special": true
92
+ },
93
+ "156902": {
94
+ "content": "<|reserved_token_7|>",
95
+ "lstrip": false,
96
+ "normalized": false,
97
+ "rstrip": false,
98
+ "single_word": false,
99
+ "special": true
100
+ },
101
+ "156903": {
102
+ "content": "<|reserved_token_8|>",
103
+ "lstrip": false,
104
+ "normalized": false,
105
+ "rstrip": false,
106
+ "single_word": false,
107
+ "special": true
108
+ },
109
+ "156904": {
110
+ "content": "<|reserved_token_9|>",
111
+ "lstrip": false,
112
+ "normalized": false,
113
+ "rstrip": false,
114
+ "single_word": false,
115
+ "special": true
116
+ },
117
+ "156905": {
118
+ "content": "<|reserved_token_10|>",
119
+ "lstrip": false,
120
+ "normalized": false,
121
+ "rstrip": false,
122
+ "single_word": false,
123
+ "special": true
124
+ },
125
+ "156906": {
126
+ "content": "<|reserved_token_11|>",
127
+ "lstrip": false,
128
+ "normalized": false,
129
+ "rstrip": false,
130
+ "single_word": false,
131
+ "special": true
132
+ },
133
+ "156907": {
134
+ "content": "<|reserved_token_12|>",
135
+ "lstrip": false,
136
+ "normalized": false,
137
+ "rstrip": false,
138
+ "single_word": false,
139
+ "special": true
140
+ },
141
+ "156908": {
142
+ "content": "<|reserved_token_13|>",
143
+ "lstrip": false,
144
+ "normalized": false,
145
+ "rstrip": false,
146
+ "single_word": false,
147
+ "special": true
148
+ },
149
+ "156909": {
150
+ "content": "<|reserved_token_14|>",
151
+ "lstrip": false,
152
+ "normalized": false,
153
+ "rstrip": false,
154
+ "single_word": false,
155
+ "special": true
156
+ },
157
+ "156910": {
158
+ "content": "<|reserved_token_15|>",
159
+ "lstrip": false,
160
+ "normalized": false,
161
+ "rstrip": false,
162
+ "single_word": false,
163
+ "special": true
164
+ },
165
+ "156911": {
166
+ "content": "<|reserved_token_16|>",
167
+ "lstrip": false,
168
+ "normalized": false,
169
+ "rstrip": false,
170
+ "single_word": false,
171
+ "special": true
172
+ },
173
+ "156912": {
174
+ "content": "<|reserved_token_17|>",
175
+ "lstrip": false,
176
+ "normalized": false,
177
+ "rstrip": false,
178
+ "single_word": false,
179
+ "special": true
180
+ },
181
+ "156913": {
182
+ "content": "<|reserved_token_18|>",
183
+ "lstrip": false,
184
+ "normalized": false,
185
+ "rstrip": false,
186
+ "single_word": false,
187
+ "special": true
188
+ },
189
+ "156914": {
190
+ "content": "<|reserved_token_19|>",
191
+ "lstrip": false,
192
+ "normalized": false,
193
+ "rstrip": false,
194
+ "single_word": false,
195
+ "special": true
196
+ },
197
+ "156915": {
198
+ "content": "<|reserved_token_20|>",
199
+ "lstrip": false,
200
+ "normalized": false,
201
+ "rstrip": false,
202
+ "single_word": false,
203
+ "special": true
204
+ },
205
+ "156916": {
206
+ "content": "<|reserved_token_21|>",
207
+ "lstrip": false,
208
+ "normalized": false,
209
+ "rstrip": false,
210
+ "single_word": false,
211
+ "special": true
212
+ },
213
+ "156917": {
214
+ "content": "<|reserved_token_22|>",
215
+ "lstrip": false,
216
+ "normalized": false,
217
+ "rstrip": false,
218
+ "single_word": false,
219
+ "special": true
220
+ },
221
+ "156918": {
222
+ "content": "<|reserved_token_23|>",
223
+ "lstrip": false,
224
+ "normalized": false,
225
+ "rstrip": false,
226
+ "single_word": false,
227
+ "special": true
228
+ },
229
+ "156919": {
230
+ "content": "<|reserved_token_24|>",
231
+ "lstrip": false,
232
+ "normalized": false,
233
+ "rstrip": false,
234
+ "single_word": false,
235
+ "special": true
236
+ },
237
+ "156920": {
238
+ "content": "<|reserved_token_25|>",
239
+ "lstrip": false,
240
+ "normalized": false,
241
+ "rstrip": false,
242
+ "single_word": false,
243
+ "special": true
244
+ },
245
+ "156921": {
246
+ "content": "<|reserved_token_26|>",
247
+ "lstrip": false,
248
+ "normalized": false,
249
+ "rstrip": false,
250
+ "single_word": false,
251
+ "special": true
252
+ },
253
+ "156922": {
254
+ "content": "<|reserved_token_27|>",
255
+ "lstrip": false,
256
+ "normalized": false,
257
+ "rstrip": false,
258
+ "single_word": false,
259
+ "special": true
260
+ },
261
+ "156923": {
262
+ "content": "<|reserved_token_28|>",
263
+ "lstrip": false,
264
+ "normalized": false,
265
+ "rstrip": false,
266
+ "single_word": false,
267
+ "special": true
268
+ },
269
+ "156924": {
270
+ "content": "<|reserved_token_29|>",
271
+ "lstrip": false,
272
+ "normalized": false,
273
+ "rstrip": false,
274
+ "single_word": false,
275
+ "special": true
276
+ },
277
+ "156925": {
278
+ "content": "<|reserved_token_30|>",
279
+ "lstrip": false,
280
+ "normalized": false,
281
+ "rstrip": false,
282
+ "single_word": false,
283
+ "special": true
284
+ },
285
+ "156926": {
286
+ "content": "<|reserved_token_31|>",
287
+ "lstrip": false,
288
+ "normalized": false,
289
+ "rstrip": false,
290
+ "single_word": false,
291
+ "special": true
292
+ },
293
+ "156927": {
294
+ "content": "<|reserved_token_32|>",
295
+ "lstrip": false,
296
+ "normalized": false,
297
+ "rstrip": false,
298
+ "single_word": false,
299
+ "special": true
300
+ },
301
+ "156928": {
302
+ "content": "<|reserved_token_33|>",
303
+ "lstrip": false,
304
+ "normalized": false,
305
+ "rstrip": false,
306
+ "single_word": false,
307
+ "special": true
308
+ },
309
+ "156929": {
310
+ "content": "<|reserved_token_34|>",
311
+ "lstrip": false,
312
+ "normalized": false,
313
+ "rstrip": false,
314
+ "single_word": false,
315
+ "special": true
316
+ },
317
+ "156930": {
318
+ "content": "<|reserved_token_35|>",
319
+ "lstrip": false,
320
+ "normalized": false,
321
+ "rstrip": false,
322
+ "single_word": false,
323
+ "special": true
324
+ },
325
+ "156931": {
326
+ "content": "<|reserved_token_36|>",
327
+ "lstrip": false,
328
+ "normalized": false,
329
+ "rstrip": false,
330
+ "single_word": false,
331
+ "special": true
332
+ },
333
+ "156932": {
334
+ "content": "<|reserved_token_37|>",
335
+ "lstrip": false,
336
+ "normalized": false,
337
+ "rstrip": false,
338
+ "single_word": false,
339
+ "special": true
340
+ },
341
+ "156933": {
342
+ "content": "<|reserved_token_38|>",
343
+ "lstrip": false,
344
+ "normalized": false,
345
+ "rstrip": false,
346
+ "single_word": false,
347
+ "special": true
348
+ },
349
+ "156934": {
350
+ "content": "<|reserved_token_39|>",
351
+ "lstrip": false,
352
+ "normalized": false,
353
+ "rstrip": false,
354
+ "single_word": false,
355
+ "special": true
356
+ },
357
+ "156935": {
358
+ "content": "<|reserved_token_40|>",
359
+ "lstrip": false,
360
+ "normalized": false,
361
+ "rstrip": false,
362
+ "single_word": false,
363
+ "special": true
364
+ },
365
+ "156936": {
366
+ "content": "<|reserved_token_41|>",
367
+ "lstrip": false,
368
+ "normalized": false,
369
+ "rstrip": false,
370
+ "single_word": false,
371
+ "special": true
372
+ },
373
+ "156937": {
374
+ "content": "<|reserved_token_42|>",
375
+ "lstrip": false,
376
+ "normalized": false,
377
+ "rstrip": false,
378
+ "single_word": false,
379
+ "special": true
380
+ },
381
+ "156938": {
382
+ "content": "<|reserved_token_43|>",
383
+ "lstrip": false,
384
+ "normalized": false,
385
+ "rstrip": false,
386
+ "single_word": false,
387
+ "special": true
388
+ },
389
+ "156939": {
390
+ "content": "<|reserved_token_44|>",
391
+ "lstrip": false,
392
+ "normalized": false,
393
+ "rstrip": false,
394
+ "single_word": false,
395
+ "special": true
396
+ },
397
+ "156940": {
398
+ "content": "<|reserved_token_45|>",
399
+ "lstrip": false,
400
+ "normalized": false,
401
+ "rstrip": false,
402
+ "single_word": false,
403
+ "special": true
404
+ },
405
+ "156941": {
406
+ "content": "<|reserved_token_46|>",
407
+ "lstrip": false,
408
+ "normalized": false,
409
+ "rstrip": false,
410
+ "single_word": false,
411
+ "special": true
412
+ },
413
+ "156942": {
414
+ "content": "<|reserved_token_47|>",
415
+ "lstrip": false,
416
+ "normalized": false,
417
+ "rstrip": false,
418
+ "single_word": false,
419
+ "special": true
420
+ },
421
+ "156943": {
422
+ "content": "<|reserved_token_48|>",
423
+ "lstrip": false,
424
+ "normalized": false,
425
+ "rstrip": false,
426
+ "single_word": false,
427
+ "special": true
428
+ },
429
+ "156944": {
430
+ "content": "<|reserved_token_49|>",
431
+ "lstrip": false,
432
+ "normalized": false,
433
+ "rstrip": false,
434
+ "single_word": false,
435
+ "special": true
436
+ },
437
+ "156945": {
438
+ "content": "<|reserved_token_50|>",
439
+ "lstrip": false,
440
+ "normalized": false,
441
+ "rstrip": false,
442
+ "single_word": false,
443
+ "special": true
444
+ },
445
+ "156946": {
446
+ "content": "<|reserved_token_51|>",
447
+ "lstrip": false,
448
+ "normalized": false,
449
+ "rstrip": false,
450
+ "single_word": false,
451
+ "special": true
452
+ },
453
+ "156947": {
454
+ "content": "<|reserved_token_52|>",
455
+ "lstrip": false,
456
+ "normalized": false,
457
+ "rstrip": false,
458
+ "single_word": false,
459
+ "special": true
460
+ },
461
+ "156948": {
462
+ "content": "<|reserved_token_53|>",
463
+ "lstrip": false,
464
+ "normalized": false,
465
+ "rstrip": false,
466
+ "single_word": false,
467
+ "special": true
468
+ },
469
+ "156949": {
470
+ "content": "<|reserved_token_54|>",
471
+ "lstrip": false,
472
+ "normalized": false,
473
+ "rstrip": false,
474
+ "single_word": false,
475
+ "special": true
476
+ },
477
+ "156950": {
478
+ "content": "<|reserved_token_55|>",
479
+ "lstrip": false,
480
+ "normalized": false,
481
+ "rstrip": false,
482
+ "single_word": false,
483
+ "special": true
484
+ },
485
+ "156951": {
486
+ "content": "<|reserved_token_56|>",
487
+ "lstrip": false,
488
+ "normalized": false,
489
+ "rstrip": false,
490
+ "single_word": false,
491
+ "special": true
492
+ },
493
+ "156952": {
494
+ "content": "<|reserved_token_57|>",
495
+ "lstrip": false,
496
+ "normalized": false,
497
+ "rstrip": false,
498
+ "single_word": false,
499
+ "special": true
500
+ },
501
+ "156953": {
502
+ "content": "<|reserved_token_58|>",
503
+ "lstrip": false,
504
+ "normalized": false,
505
+ "rstrip": false,
506
+ "single_word": false,
507
+ "special": true
508
+ },
509
+ "156954": {
510
+ "content": "<|reserved_token_59|>",
511
+ "lstrip": false,
512
+ "normalized": false,
513
+ "rstrip": false,
514
+ "single_word": false,
515
+ "special": true
516
+ },
517
+ "156955": {
518
+ "content": "<|reserved_token_60|>",
519
+ "lstrip": false,
520
+ "normalized": false,
521
+ "rstrip": false,
522
+ "single_word": false,
523
+ "special": true
524
+ },
525
+ "156956": {
526
+ "content": "<|reserved_token_61|>",
527
+ "lstrip": false,
528
+ "normalized": false,
529
+ "rstrip": false,
530
+ "single_word": false,
531
+ "special": true
532
+ },
533
+ "156957": {
534
+ "content": "<|reserved_token_62|>",
535
+ "lstrip": false,
536
+ "normalized": false,
537
+ "rstrip": false,
538
+ "single_word": false,
539
+ "special": true
540
+ },
541
+ "156958": {
542
+ "content": "<|reserved_token_63|>",
543
+ "lstrip": false,
544
+ "normalized": false,
545
+ "rstrip": false,
546
+ "single_word": false,
547
+ "special": true
548
+ },
549
+ "156959": {
550
+ "content": "<|reserved_token_64|>",
551
+ "lstrip": false,
552
+ "normalized": false,
553
+ "rstrip": false,
554
+ "single_word": false,
555
+ "special": true
556
+ },
557
+ "156960": {
558
+ "content": "<|reserved_token_65|>",
559
+ "lstrip": false,
560
+ "normalized": false,
561
+ "rstrip": false,
562
+ "single_word": false,
563
+ "special": true
564
+ },
565
+ "156961": {
566
+ "content": "<|reserved_token_66|>",
567
+ "lstrip": false,
568
+ "normalized": false,
569
+ "rstrip": false,
570
+ "single_word": false,
571
+ "special": true
572
+ },
573
+ "156962": {
574
+ "content": "<|reserved_token_67|>",
575
+ "lstrip": false,
576
+ "normalized": false,
577
+ "rstrip": false,
578
+ "single_word": false,
579
+ "special": true
580
+ },
581
+ "156963": {
582
+ "content": "<|reserved_token_68|>",
583
+ "lstrip": false,
584
+ "normalized": false,
585
+ "rstrip": false,
586
+ "single_word": false,
587
+ "special": true
588
+ },
589
+ "156964": {
590
+ "content": "<|reserved_token_69|>",
591
+ "lstrip": false,
592
+ "normalized": false,
593
+ "rstrip": false,
594
+ "single_word": false,
595
+ "special": true
596
+ },
597
+ "156965": {
598
+ "content": "<|reserved_token_70|>",
599
+ "lstrip": false,
600
+ "normalized": false,
601
+ "rstrip": false,
602
+ "single_word": false,
603
+ "special": true
604
+ },
605
+ "156966": {
606
+ "content": "<|reserved_token_71|>",
607
+ "lstrip": false,
608
+ "normalized": false,
609
+ "rstrip": false,
610
+ "single_word": false,
611
+ "special": true
612
+ },
613
+ "156967": {
614
+ "content": "<|reserved_token_72|>",
615
+ "lstrip": false,
616
+ "normalized": false,
617
+ "rstrip": false,
618
+ "single_word": false,
619
+ "special": true
620
+ },
621
+ "156968": {
622
+ "content": "<|reserved_token_73|>",
623
+ "lstrip": false,
624
+ "normalized": false,
625
+ "rstrip": false,
626
+ "single_word": false,
627
+ "special": true
628
+ },
629
+ "156969": {
630
+ "content": "<|reserved_token_74|>",
631
+ "lstrip": false,
632
+ "normalized": false,
633
+ "rstrip": false,
634
+ "single_word": false,
635
+ "special": true
636
+ },
637
+ "156970": {
638
+ "content": "<|reserved_token_75|>",
639
+ "lstrip": false,
640
+ "normalized": false,
641
+ "rstrip": false,
642
+ "single_word": false,
643
+ "special": true
644
+ },
645
+ "156971": {
646
+ "content": "<|reserved_token_76|>",
647
+ "lstrip": false,
648
+ "normalized": false,
649
+ "rstrip": false,
650
+ "single_word": false,
651
+ "special": true
652
+ },
653
+ "156972": {
654
+ "content": "<|reserved_token_77|>",
655
+ "lstrip": false,
656
+ "normalized": false,
657
+ "rstrip": false,
658
+ "single_word": false,
659
+ "special": true
660
+ },
661
+ "156973": {
662
+ "content": "<|reserved_token_78|>",
663
+ "lstrip": false,
664
+ "normalized": false,
665
+ "rstrip": false,
666
+ "single_word": false,
667
+ "special": true
668
+ },
669
+ "156974": {
670
+ "content": "<|reserved_token_79|>",
671
+ "lstrip": false,
672
+ "normalized": false,
673
+ "rstrip": false,
674
+ "single_word": false,
675
+ "special": true
676
+ },
677
+ "156975": {
678
+ "content": "<|reserved_token_80|>",
679
+ "lstrip": false,
680
+ "normalized": false,
681
+ "rstrip": false,
682
+ "single_word": false,
683
+ "special": true
684
+ },
685
+ "156976": {
686
+ "content": "<|reserved_token_81|>",
687
+ "lstrip": false,
688
+ "normalized": false,
689
+ "rstrip": false,
690
+ "single_word": false,
691
+ "special": true
692
+ },
693
+ "156977": {
694
+ "content": "<|reserved_token_82|>",
695
+ "lstrip": false,
696
+ "normalized": false,
697
+ "rstrip": false,
698
+ "single_word": false,
699
+ "special": true
700
+ },
701
+ "156978": {
702
+ "content": "<|reserved_token_83|>",
703
+ "lstrip": false,
704
+ "normalized": false,
705
+ "rstrip": false,
706
+ "single_word": false,
707
+ "special": true
708
+ },
709
+ "156979": {
710
+ "content": "<|reserved_token_84|>",
711
+ "lstrip": false,
712
+ "normalized": false,
713
+ "rstrip": false,
714
+ "single_word": false,
715
+ "special": true
716
+ },
717
+ "156980": {
718
+ "content": "<|reserved_token_85|>",
719
+ "lstrip": false,
720
+ "normalized": false,
721
+ "rstrip": false,
722
+ "single_word": false,
723
+ "special": true
724
+ },
725
+ "156981": {
726
+ "content": "<|reserved_token_86|>",
727
+ "lstrip": false,
728
+ "normalized": false,
729
+ "rstrip": false,
730
+ "single_word": false,
731
+ "special": true
732
+ },
733
+ "156982": {
734
+ "content": "<|reserved_token_87|>",
735
+ "lstrip": false,
736
+ "normalized": false,
737
+ "rstrip": false,
738
+ "single_word": false,
739
+ "special": true
740
+ },
741
+ "156983": {
742
+ "content": "<|reserved_token_88|>",
743
+ "lstrip": false,
744
+ "normalized": false,
745
+ "rstrip": false,
746
+ "single_word": false,
747
+ "special": true
748
+ },
749
+ "156984": {
750
+ "content": "<|reserved_token_89|>",
751
+ "lstrip": false,
752
+ "normalized": false,
753
+ "rstrip": false,
754
+ "single_word": false,
755
+ "special": true
756
+ },
757
+ "156985": {
758
+ "content": "<|reserved_token_90|>",
759
+ "lstrip": false,
760
+ "normalized": false,
761
+ "rstrip": false,
762
+ "single_word": false,
763
+ "special": true
764
+ },
765
+ "156986": {
766
+ "content": "<|reserved_token_91|>",
767
+ "lstrip": false,
768
+ "normalized": false,
769
+ "rstrip": false,
770
+ "single_word": false,
771
+ "special": true
772
+ },
773
+ "156987": {
774
+ "content": "<|reserved_token_92|>",
775
+ "lstrip": false,
776
+ "normalized": false,
777
+ "rstrip": false,
778
+ "single_word": false,
779
+ "special": true
780
+ },
781
+ "156988": {
782
+ "content": "<|reserved_token_93|>",
783
+ "lstrip": false,
784
+ "normalized": false,
785
+ "rstrip": false,
786
+ "single_word": false,
787
+ "special": true
788
+ },
789
+ "156989": {
790
+ "content": "<|reserved_token_94|>",
791
+ "lstrip": false,
792
+ "normalized": false,
793
+ "rstrip": false,
794
+ "single_word": false,
795
+ "special": true
796
+ },
797
+ "156990": {
798
+ "content": "<|reserved_token_95|>",
799
+ "lstrip": false,
800
+ "normalized": false,
801
+ "rstrip": false,
802
+ "single_word": false,
803
+ "special": true
804
+ },
805
+ "156991": {
806
+ "content": "<|reserved_token_96|>",
807
+ "lstrip": false,
808
+ "normalized": false,
809
+ "rstrip": false,
810
+ "single_word": false,
811
+ "special": true
812
+ },
813
+ "156992": {
814
+ "content": "<|reserved_token_97|>",
815
+ "lstrip": false,
816
+ "normalized": false,
817
+ "rstrip": false,
818
+ "single_word": false,
819
+ "special": true
820
+ },
821
+ "156993": {
822
+ "content": "<|reserved_token_98|>",
823
+ "lstrip": false,
824
+ "normalized": false,
825
+ "rstrip": false,
826
+ "single_word": false,
827
+ "special": true
828
+ },
829
+ "156994": {
830
+ "content": "<|reserved_token_99|>",
831
+ "lstrip": false,
832
+ "normalized": false,
833
+ "rstrip": false,
834
+ "single_word": false,
835
+ "special": true
836
+ },
837
+ "156995": {
838
+ "content": "<|reserved_token_100|>",
839
+ "lstrip": false,
840
+ "normalized": false,
841
+ "rstrip": false,
842
+ "single_word": false,
843
+ "special": true
844
+ },
845
+ "156996": {
846
+ "content": "<|reserved_token_101|>",
847
+ "lstrip": false,
848
+ "normalized": false,
849
+ "rstrip": false,
850
+ "single_word": false,
851
+ "special": true
852
+ },
853
+ "156997": {
854
+ "content": "<|reserved_token_102|>",
855
+ "lstrip": false,
856
+ "normalized": false,
857
+ "rstrip": false,
858
+ "single_word": false,
859
+ "special": true
860
+ },
861
+ "156998": {
862
+ "content": "<|reserved_token_103|>",
863
+ "lstrip": false,
864
+ "normalized": false,
865
+ "rstrip": false,
866
+ "single_word": false,
867
+ "special": true
868
+ },
869
+ "156999": {
870
+ "content": "<|reserved_token_104|>",
871
+ "lstrip": false,
872
+ "normalized": false,
873
+ "rstrip": false,
874
+ "single_word": false,
875
+ "special": true
876
+ },
877
+ "157000": {
878
+ "content": "<|reserved_token_105|>",
879
+ "lstrip": false,
880
+ "normalized": false,
881
+ "rstrip": false,
882
+ "single_word": false,
883
+ "special": true
884
+ },
885
+ "157001": {
886
+ "content": "<|reserved_token_106|>",
887
+ "lstrip": false,
888
+ "normalized": false,
889
+ "rstrip": false,
890
+ "single_word": false,
891
+ "special": true
892
+ },
893
+ "157002": {
894
+ "content": "<|reserved_token_107|>",
895
+ "lstrip": false,
896
+ "normalized": false,
897
+ "rstrip": false,
898
+ "single_word": false,
899
+ "special": true
900
+ },
901
+ "157003": {
902
+ "content": "<|reserved_token_108|>",
903
+ "lstrip": false,
904
+ "normalized": false,
905
+ "rstrip": false,
906
+ "single_word": false,
907
+ "special": true
908
+ },
909
+ "157004": {
910
+ "content": "<|reserved_token_109|>",
911
+ "lstrip": false,
912
+ "normalized": false,
913
+ "rstrip": false,
914
+ "single_word": false,
915
+ "special": true
916
+ },
917
+ "157005": {
918
+ "content": "<|reserved_token_110|>",
919
+ "lstrip": false,
920
+ "normalized": false,
921
+ "rstrip": false,
922
+ "single_word": false,
923
+ "special": true
924
+ },
925
+ "157006": {
926
+ "content": "<|reserved_token_111|>",
927
+ "lstrip": false,
928
+ "normalized": false,
929
+ "rstrip": false,
930
+ "single_word": false,
931
+ "special": true
932
+ },
933
+ "157007": {
934
+ "content": "<|reserved_token_112|>",
935
+ "lstrip": false,
936
+ "normalized": false,
937
+ "rstrip": false,
938
+ "single_word": false,
939
+ "special": true
940
+ },
941
+ "157008": {
942
+ "content": "<|reserved_token_113|>",
943
+ "lstrip": false,
944
+ "normalized": false,
945
+ "rstrip": false,
946
+ "single_word": false,
947
+ "special": true
948
+ },
949
+ "157009": {
950
+ "content": "<|reserved_token_114|>",
951
+ "lstrip": false,
952
+ "normalized": false,
953
+ "rstrip": false,
954
+ "single_word": false,
955
+ "special": true
956
+ },
957
+ "157010": {
958
+ "content": "<|reserved_token_115|>",
959
+ "lstrip": false,
960
+ "normalized": false,
961
+ "rstrip": false,
962
+ "single_word": false,
963
+ "special": true
964
+ },
965
+ "157011": {
966
+ "content": "<|reserved_token_116|>",
967
+ "lstrip": false,
968
+ "normalized": false,
969
+ "rstrip": false,
970
+ "single_word": false,
971
+ "special": true
972
+ },
973
+ "157012": {
974
+ "content": "<|reserved_token_117|>",
975
+ "lstrip": false,
976
+ "normalized": false,
977
+ "rstrip": false,
978
+ "single_word": false,
979
+ "special": true
980
+ },
981
+ "157013": {
982
+ "content": "<|reserved_token_118|>",
983
+ "lstrip": false,
984
+ "normalized": false,
985
+ "rstrip": false,
986
+ "single_word": false,
987
+ "special": true
988
+ },
989
+ "157014": {
990
+ "content": "<|reserved_token_119|>",
991
+ "lstrip": false,
992
+ "normalized": false,
993
+ "rstrip": false,
994
+ "single_word": false,
995
+ "special": true
996
+ },
997
+ "157015": {
998
+ "content": "<|reserved_token_120|>",
999
+ "lstrip": false,
1000
+ "normalized": false,
1001
+ "rstrip": false,
1002
+ "single_word": false,
1003
+ "special": true
1004
+ },
1005
+ "157016": {
1006
+ "content": "<|reserved_token_121|>",
1007
+ "lstrip": false,
1008
+ "normalized": false,
1009
+ "rstrip": false,
1010
+ "single_word": false,
1011
+ "special": true
1012
+ },
1013
+ "157017": {
1014
+ "content": "<|reserved_token_122|>",
1015
+ "lstrip": false,
1016
+ "normalized": false,
1017
+ "rstrip": false,
1018
+ "single_word": false,
1019
+ "special": true
1020
+ },
1021
+ "157018": {
1022
+ "content": "<|reserved_token_123|>",
1023
+ "lstrip": false,
1024
+ "normalized": false,
1025
+ "rstrip": false,
1026
+ "single_word": false,
1027
+ "special": true
1028
+ },
1029
+ "157019": {
1030
+ "content": "<|reserved_token_124|>",
1031
+ "lstrip": false,
1032
+ "normalized": false,
1033
+ "rstrip": false,
1034
+ "single_word": false,
1035
+ "special": true
1036
+ },
1037
+ "157020": {
1038
+ "content": "<|reserved_token_125|>",
1039
+ "lstrip": false,
1040
+ "normalized": false,
1041
+ "rstrip": false,
1042
+ "single_word": false,
1043
+ "special": true
1044
+ },
1045
+ "157021": {
1046
+ "content": "<|reserved_token_126|>",
1047
+ "lstrip": false,
1048
+ "normalized": false,
1049
+ "rstrip": false,
1050
+ "single_word": false,
1051
+ "special": true
1052
+ },
1053
+ "157022": {
1054
+ "content": "<|reserved_token_127|>",
1055
+ "lstrip": false,
1056
+ "normalized": false,
1057
+ "rstrip": false,
1058
+ "single_word": false,
1059
+ "special": true
1060
+ },
1061
+ "157023": {
1062
+ "content": "<|reserved_token_128|>",
1063
+ "lstrip": false,
1064
+ "normalized": false,
1065
+ "rstrip": false,
1066
+ "single_word": false,
1067
+ "special": true
1068
+ },
1069
+ "157024": {
1070
+ "content": "<|reserved_token_129|>",
1071
+ "lstrip": false,
1072
+ "normalized": false,
1073
+ "rstrip": false,
1074
+ "single_word": false,
1075
+ "special": true
1076
+ },
1077
+ "157025": {
1078
+ "content": "<|reserved_token_130|>",
1079
+ "lstrip": false,
1080
+ "normalized": false,
1081
+ "rstrip": false,
1082
+ "single_word": false,
1083
+ "special": true
1084
+ },
1085
+ "157026": {
1086
+ "content": "<|reserved_token_131|>",
1087
+ "lstrip": false,
1088
+ "normalized": false,
1089
+ "rstrip": false,
1090
+ "single_word": false,
1091
+ "special": true
1092
+ },
1093
+ "157027": {
1094
+ "content": "<|reserved_token_132|>",
1095
+ "lstrip": false,
1096
+ "normalized": false,
1097
+ "rstrip": false,
1098
+ "single_word": false,
1099
+ "special": true
1100
+ },
1101
+ "157028": {
1102
+ "content": "<|reserved_token_133|>",
1103
+ "lstrip": false,
1104
+ "normalized": false,
1105
+ "rstrip": false,
1106
+ "single_word": false,
1107
+ "special": true
1108
+ },
1109
+ "157029": {
1110
+ "content": "<|reserved_token_134|>",
1111
+ "lstrip": false,
1112
+ "normalized": false,
1113
+ "rstrip": false,
1114
+ "single_word": false,
1115
+ "special": true
1116
+ },
1117
+ "157030": {
1118
+ "content": "<|reserved_token_135|>",
1119
+ "lstrip": false,
1120
+ "normalized": false,
1121
+ "rstrip": false,
1122
+ "single_word": false,
1123
+ "special": true
1124
+ },
1125
+ "157031": {
1126
+ "content": "<|reserved_token_136|>",
1127
+ "lstrip": false,
1128
+ "normalized": false,
1129
+ "rstrip": false,
1130
+ "single_word": false,
1131
+ "special": true
1132
+ },
1133
+ "157032": {
1134
+ "content": "<|reserved_token_137|>",
1135
+ "lstrip": false,
1136
+ "normalized": false,
1137
+ "rstrip": false,
1138
+ "single_word": false,
1139
+ "special": true
1140
+ },
1141
+ "157033": {
1142
+ "content": "<|reserved_token_138|>",
1143
+ "lstrip": false,
1144
+ "normalized": false,
1145
+ "rstrip": false,
1146
+ "single_word": false,
1147
+ "special": true
1148
+ },
1149
+ "157034": {
1150
+ "content": "<|reserved_token_139|>",
1151
+ "lstrip": false,
1152
+ "normalized": false,
1153
+ "rstrip": false,
1154
+ "single_word": false,
1155
+ "special": true
1156
+ },
1157
+ "157035": {
1158
+ "content": "<|reserved_token_140|>",
1159
+ "lstrip": false,
1160
+ "normalized": false,
1161
+ "rstrip": false,
1162
+ "single_word": false,
1163
+ "special": true
1164
+ },
1165
+ "157036": {
1166
+ "content": "<|reserved_token_141|>",
1167
+ "lstrip": false,
1168
+ "normalized": false,
1169
+ "rstrip": false,
1170
+ "single_word": false,
1171
+ "special": true
1172
+ },
1173
+ "157037": {
1174
+ "content": "<|reserved_token_142|>",
1175
+ "lstrip": false,
1176
+ "normalized": false,
1177
+ "rstrip": false,
1178
+ "single_word": false,
1179
+ "special": true
1180
+ },
1181
+ "157038": {
1182
+ "content": "<|reserved_token_143|>",
1183
+ "lstrip": false,
1184
+ "normalized": false,
1185
+ "rstrip": false,
1186
+ "single_word": false,
1187
+ "special": true
1188
+ },
1189
+ "157039": {
1190
+ "content": "<|reserved_token_144|>",
1191
+ "lstrip": false,
1192
+ "normalized": false,
1193
+ "rstrip": false,
1194
+ "single_word": false,
1195
+ "special": true
1196
+ },
1197
+ "157040": {
1198
+ "content": "<|reserved_token_145|>",
1199
+ "lstrip": false,
1200
+ "normalized": false,
1201
+ "rstrip": false,
1202
+ "single_word": false,
1203
+ "special": true
1204
+ },
1205
+ "157041": {
1206
+ "content": "<|reserved_token_146|>",
1207
+ "lstrip": false,
1208
+ "normalized": false,
1209
+ "rstrip": false,
1210
+ "single_word": false,
1211
+ "special": true
1212
+ },
1213
+ "157042": {
1214
+ "content": "<|reserved_token_147|>",
1215
+ "lstrip": false,
1216
+ "normalized": false,
1217
+ "rstrip": false,
1218
+ "single_word": false,
1219
+ "special": true
1220
+ },
1221
+ "157043": {
1222
+ "content": "<|reserved_token_148|>",
1223
+ "lstrip": false,
1224
+ "normalized": false,
1225
+ "rstrip": false,
1226
+ "single_word": false,
1227
+ "special": true
1228
+ },
1229
+ "157044": {
1230
+ "content": "<|reserved_token_149|>",
1231
+ "lstrip": false,
1232
+ "normalized": false,
1233
+ "rstrip": false,
1234
+ "single_word": false,
1235
+ "special": true
1236
+ },
1237
+ "157045": {
1238
+ "content": "<|reserved_token_150|>",
1239
+ "lstrip": false,
1240
+ "normalized": false,
1241
+ "rstrip": false,
1242
+ "single_word": false,
1243
+ "special": true
1244
+ },
1245
+ "157046": {
1246
+ "content": "<|reserved_token_151|>",
1247
+ "lstrip": false,
1248
+ "normalized": false,
1249
+ "rstrip": false,
1250
+ "single_word": false,
1251
+ "special": true
1252
+ },
1253
+ "157047": {
1254
+ "content": "<|reserved_token_152|>",
1255
+ "lstrip": false,
1256
+ "normalized": false,
1257
+ "rstrip": false,
1258
+ "single_word": false,
1259
+ "special": true
1260
+ },
1261
+ "157048": {
1262
+ "content": "<|reserved_token_153|>",
1263
+ "lstrip": false,
1264
+ "normalized": false,
1265
+ "rstrip": false,
1266
+ "single_word": false,
1267
+ "special": true
1268
+ },
1269
+ "157049": {
1270
+ "content": "<|reserved_token_154|>",
1271
+ "lstrip": false,
1272
+ "normalized": false,
1273
+ "rstrip": false,
1274
+ "single_word": false,
1275
+ "special": true
1276
+ },
1277
+ "157050": {
1278
+ "content": "<|reserved_token_155|>",
1279
+ "lstrip": false,
1280
+ "normalized": false,
1281
+ "rstrip": false,
1282
+ "single_word": false,
1283
+ "special": true
1284
+ },
1285
+ "157051": {
1286
+ "content": "<|reserved_token_156|>",
1287
+ "lstrip": false,
1288
+ "normalized": false,
1289
+ "rstrip": false,
1290
+ "single_word": false,
1291
+ "special": true
1292
+ },
1293
+ "157052": {
1294
+ "content": "<|reserved_token_157|>",
1295
+ "lstrip": false,
1296
+ "normalized": false,
1297
+ "rstrip": false,
1298
+ "single_word": false,
1299
+ "special": true
1300
+ },
1301
+ "157053": {
1302
+ "content": "<|reserved_token_158|>",
1303
+ "lstrip": false,
1304
+ "normalized": false,
1305
+ "rstrip": false,
1306
+ "single_word": false,
1307
+ "special": true
1308
+ },
1309
+ "157054": {
1310
+ "content": "<|reserved_token_159|>",
1311
+ "lstrip": false,
1312
+ "normalized": false,
1313
+ "rstrip": false,
1314
+ "single_word": false,
1315
+ "special": true
1316
+ },
1317
+ "157055": {
1318
+ "content": "<|reserved_token_160|>",
1319
+ "lstrip": false,
1320
+ "normalized": false,
1321
+ "rstrip": false,
1322
+ "single_word": false,
1323
+ "special": true
1324
+ },
1325
+ "157056": {
1326
+ "content": "<|reserved_token_161|>",
1327
+ "lstrip": false,
1328
+ "normalized": false,
1329
+ "rstrip": false,
1330
+ "single_word": false,
1331
+ "special": true
1332
+ },
1333
+ "157057": {
1334
+ "content": "<|reserved_token_162|>",
1335
+ "lstrip": false,
1336
+ "normalized": false,
1337
+ "rstrip": false,
1338
+ "single_word": false,
1339
+ "special": true
1340
+ },
1341
+ "157058": {
1342
+ "content": "<|reserved_token_163|>",
1343
+ "lstrip": false,
1344
+ "normalized": false,
1345
+ "rstrip": false,
1346
+ "single_word": false,
1347
+ "special": true
1348
+ },
1349
+ "157059": {
1350
+ "content": "<|reserved_token_164|>",
1351
+ "lstrip": false,
1352
+ "normalized": false,
1353
+ "rstrip": false,
1354
+ "single_word": false,
1355
+ "special": true
1356
+ },
1357
+ "157060": {
1358
+ "content": "<|reserved_token_165|>",
1359
+ "lstrip": false,
1360
+ "normalized": false,
1361
+ "rstrip": false,
1362
+ "single_word": false,
1363
+ "special": true
1364
+ },
1365
+ "157061": {
1366
+ "content": "<|reserved_token_166|>",
1367
+ "lstrip": false,
1368
+ "normalized": false,
1369
+ "rstrip": false,
1370
+ "single_word": false,
1371
+ "special": true
1372
+ },
1373
+ "157062": {
1374
+ "content": "<|reserved_token_167|>",
1375
+ "lstrip": false,
1376
+ "normalized": false,
1377
+ "rstrip": false,
1378
+ "single_word": false,
1379
+ "special": true
1380
+ },
1381
+ "157063": {
1382
+ "content": "<|reserved_token_168|>",
1383
+ "lstrip": false,
1384
+ "normalized": false,
1385
+ "rstrip": false,
1386
+ "single_word": false,
1387
+ "special": true
1388
+ },
1389
+ "157064": {
1390
+ "content": "<|reserved_token_169|>",
1391
+ "lstrip": false,
1392
+ "normalized": false,
1393
+ "rstrip": false,
1394
+ "single_word": false,
1395
+ "special": true
1396
+ },
1397
+ "157065": {
1398
+ "content": "<|reserved_token_170|>",
1399
+ "lstrip": false,
1400
+ "normalized": false,
1401
+ "rstrip": false,
1402
+ "single_word": false,
1403
+ "special": true
1404
+ },
1405
+ "157066": {
1406
+ "content": "<|reserved_token_171|>",
1407
+ "lstrip": false,
1408
+ "normalized": false,
1409
+ "rstrip": false,
1410
+ "single_word": false,
1411
+ "special": true
1412
+ },
1413
+ "157067": {
1414
+ "content": "<|reserved_token_172|>",
1415
+ "lstrip": false,
1416
+ "normalized": false,
1417
+ "rstrip": false,
1418
+ "single_word": false,
1419
+ "special": true
1420
+ },
1421
+ "157068": {
1422
+ "content": "<|reserved_token_173|>",
1423
+ "lstrip": false,
1424
+ "normalized": false,
1425
+ "rstrip": false,
1426
+ "single_word": false,
1427
+ "special": true
1428
+ },
1429
+ "157069": {
1430
+ "content": "<|reserved_token_174|>",
1431
+ "lstrip": false,
1432
+ "normalized": false,
1433
+ "rstrip": false,
1434
+ "single_word": false,
1435
+ "special": true
1436
+ },
1437
+ "157070": {
1438
+ "content": "<|reserved_token_175|>",
1439
+ "lstrip": false,
1440
+ "normalized": false,
1441
+ "rstrip": false,
1442
+ "single_word": false,
1443
+ "special": true
1444
+ },
1445
+ "157071": {
1446
+ "content": "<|reserved_token_176|>",
1447
+ "lstrip": false,
1448
+ "normalized": false,
1449
+ "rstrip": false,
1450
+ "single_word": false,
1451
+ "special": true
1452
+ },
1453
+ "157072": {
1454
+ "content": "<|reserved_token_177|>",
1455
+ "lstrip": false,
1456
+ "normalized": false,
1457
+ "rstrip": false,
1458
+ "single_word": false,
1459
+ "special": true
1460
+ },
1461
+ "157073": {
1462
+ "content": "<|reserved_token_178|>",
1463
+ "lstrip": false,
1464
+ "normalized": false,
1465
+ "rstrip": false,
1466
+ "single_word": false,
1467
+ "special": true
1468
+ },
1469
+ "157074": {
1470
+ "content": "<|reserved_token_179|>",
1471
+ "lstrip": false,
1472
+ "normalized": false,
1473
+ "rstrip": false,
1474
+ "single_word": false,
1475
+ "special": true
1476
+ },
1477
+ "157075": {
1478
+ "content": "<|reserved_token_180|>",
1479
+ "lstrip": false,
1480
+ "normalized": false,
1481
+ "rstrip": false,
1482
+ "single_word": false,
1483
+ "special": true
1484
+ },
1485
+ "157076": {
1486
+ "content": "<|reserved_token_181|>",
1487
+ "lstrip": false,
1488
+ "normalized": false,
1489
+ "rstrip": false,
1490
+ "single_word": false,
1491
+ "special": true
1492
+ },
1493
+ "157077": {
1494
+ "content": "<|reserved_token_182|>",
1495
+ "lstrip": false,
1496
+ "normalized": false,
1497
+ "rstrip": false,
1498
+ "single_word": false,
1499
+ "special": true
1500
+ },
1501
+ "157078": {
1502
+ "content": "<|reserved_token_183|>",
1503
+ "lstrip": false,
1504
+ "normalized": false,
1505
+ "rstrip": false,
1506
+ "single_word": false,
1507
+ "special": true
1508
+ },
1509
+ "157079": {
1510
+ "content": "<|reserved_token_184|>",
1511
+ "lstrip": false,
1512
+ "normalized": false,
1513
+ "rstrip": false,
1514
+ "single_word": false,
1515
+ "special": true
1516
+ },
1517
+ "157080": {
1518
+ "content": "<|reserved_token_185|>",
1519
+ "lstrip": false,
1520
+ "normalized": false,
1521
+ "rstrip": false,
1522
+ "single_word": false,
1523
+ "special": true
1524
+ },
1525
+ "157081": {
1526
+ "content": "<|reserved_token_186|>",
1527
+ "lstrip": false,
1528
+ "normalized": false,
1529
+ "rstrip": false,
1530
+ "single_word": false,
1531
+ "special": true
1532
+ },
1533
+ "157082": {
1534
+ "content": "<|reserved_token_187|>",
1535
+ "lstrip": false,
1536
+ "normalized": false,
1537
+ "rstrip": false,
1538
+ "single_word": false,
1539
+ "special": true
1540
+ },
1541
+ "157083": {
1542
+ "content": "<|reserved_token_188|>",
1543
+ "lstrip": false,
1544
+ "normalized": false,
1545
+ "rstrip": false,
1546
+ "single_word": false,
1547
+ "special": true
1548
+ },
1549
+ "157084": {
1550
+ "content": "<|reserved_token_189|>",
1551
+ "lstrip": false,
1552
+ "normalized": false,
1553
+ "rstrip": false,
1554
+ "single_word": false,
1555
+ "special": true
1556
+ },
1557
+ "157085": {
1558
+ "content": "<|reserved_token_190|>",
1559
+ "lstrip": false,
1560
+ "normalized": false,
1561
+ "rstrip": false,
1562
+ "single_word": false,
1563
+ "special": true
1564
+ },
1565
+ "157086": {
1566
+ "content": "<|reserved_token_191|>",
1567
+ "lstrip": false,
1568
+ "normalized": false,
1569
+ "rstrip": false,
1570
+ "single_word": false,
1571
+ "special": true
1572
+ },
1573
+ "157087": {
1574
+ "content": "<|reserved_token_192|>",
1575
+ "lstrip": false,
1576
+ "normalized": false,
1577
+ "rstrip": false,
1578
+ "single_word": false,
1579
+ "special": true
1580
+ },
1581
+ "157088": {
1582
+ "content": "<|reserved_token_193|>",
1583
+ "lstrip": false,
1584
+ "normalized": false,
1585
+ "rstrip": false,
1586
+ "single_word": false,
1587
+ "special": true
1588
+ },
1589
+ "157089": {
1590
+ "content": "<|reserved_token_194|>",
1591
+ "lstrip": false,
1592
+ "normalized": false,
1593
+ "rstrip": false,
1594
+ "single_word": false,
1595
+ "special": true
1596
+ },
1597
+ "157090": {
1598
+ "content": "<|reserved_token_195|>",
1599
+ "lstrip": false,
1600
+ "normalized": false,
1601
+ "rstrip": false,
1602
+ "single_word": false,
1603
+ "special": true
1604
+ },
1605
+ "157091": {
1606
+ "content": "<|reserved_token_196|>",
1607
+ "lstrip": false,
1608
+ "normalized": false,
1609
+ "rstrip": false,
1610
+ "single_word": false,
1611
+ "special": true
1612
+ },
1613
+ "157092": {
1614
+ "content": "<|reserved_token_197|>",
1615
+ "lstrip": false,
1616
+ "normalized": false,
1617
+ "rstrip": false,
1618
+ "single_word": false,
1619
+ "special": true
1620
+ },
1621
+ "157093": {
1622
+ "content": "<|reserved_token_198|>",
1623
+ "lstrip": false,
1624
+ "normalized": false,
1625
+ "rstrip": false,
1626
+ "single_word": false,
1627
+ "special": true
1628
+ },
1629
+ "157094": {
1630
+ "content": "<|reserved_token_199|>",
1631
+ "lstrip": false,
1632
+ "normalized": false,
1633
+ "rstrip": false,
1634
+ "single_word": false,
1635
+ "special": true
1636
+ },
1637
+ "157095": {
1638
+ "content": "<|reserved_token_200|>",
1639
+ "lstrip": false,
1640
+ "normalized": false,
1641
+ "rstrip": false,
1642
+ "single_word": false,
1643
+ "special": true
1644
+ },
1645
+ "157096": {
1646
+ "content": "<|reserved_token_201|>",
1647
+ "lstrip": false,
1648
+ "normalized": false,
1649
+ "rstrip": false,
1650
+ "single_word": false,
1651
+ "special": true
1652
+ },
1653
+ "157097": {
1654
+ "content": "<|reserved_token_202|>",
1655
+ "lstrip": false,
1656
+ "normalized": false,
1657
+ "rstrip": false,
1658
+ "single_word": false,
1659
+ "special": true
1660
+ },
1661
+ "157098": {
1662
+ "content": "<|reserved_token_203|>",
1663
+ "lstrip": false,
1664
+ "normalized": false,
1665
+ "rstrip": false,
1666
+ "single_word": false,
1667
+ "special": true
1668
+ },
1669
+ "157099": {
1670
+ "content": "<|reserved_token_204|>",
1671
+ "lstrip": false,
1672
+ "normalized": false,
1673
+ "rstrip": false,
1674
+ "single_word": false,
1675
+ "special": true
1676
+ },
1677
+ "157100": {
1678
+ "content": "<|reserved_token_205|>",
1679
+ "lstrip": false,
1680
+ "normalized": false,
1681
+ "rstrip": false,
1682
+ "single_word": false,
1683
+ "special": true
1684
+ },
1685
+ "157101": {
1686
+ "content": "<|reserved_token_206|>",
1687
+ "lstrip": false,
1688
+ "normalized": false,
1689
+ "rstrip": false,
1690
+ "single_word": false,
1691
+ "special": true
1692
+ },
1693
+ "157102": {
1694
+ "content": "<|reserved_token_207|>",
1695
+ "lstrip": false,
1696
+ "normalized": false,
1697
+ "rstrip": false,
1698
+ "single_word": false,
1699
+ "special": true
1700
+ },
1701
+ "157103": {
1702
+ "content": "<|reserved_token_208|>",
1703
+ "lstrip": false,
1704
+ "normalized": false,
1705
+ "rstrip": false,
1706
+ "single_word": false,
1707
+ "special": true
1708
+ },
1709
+ "157104": {
1710
+ "content": "<|reserved_token_209|>",
1711
+ "lstrip": false,
1712
+ "normalized": false,
1713
+ "rstrip": false,
1714
+ "single_word": false,
1715
+ "special": true
1716
+ },
1717
+ "157105": {
1718
+ "content": "<|reserved_token_210|>",
1719
+ "lstrip": false,
1720
+ "normalized": false,
1721
+ "rstrip": false,
1722
+ "single_word": false,
1723
+ "special": true
1724
+ },
1725
+ "157106": {
1726
+ "content": "<|reserved_token_211|>",
1727
+ "lstrip": false,
1728
+ "normalized": false,
1729
+ "rstrip": false,
1730
+ "single_word": false,
1731
+ "special": true
1732
+ },
1733
+ "157107": {
1734
+ "content": "<|reserved_token_212|>",
1735
+ "lstrip": false,
1736
+ "normalized": false,
1737
+ "rstrip": false,
1738
+ "single_word": false,
1739
+ "special": true
1740
+ },
1741
+ "157108": {
1742
+ "content": "<|reserved_token_213|>",
1743
+ "lstrip": false,
1744
+ "normalized": false,
1745
+ "rstrip": false,
1746
+ "single_word": false,
1747
+ "special": true
1748
+ },
1749
+ "157109": {
1750
+ "content": "<|reserved_token_214|>",
1751
+ "lstrip": false,
1752
+ "normalized": false,
1753
+ "rstrip": false,
1754
+ "single_word": false,
1755
+ "special": true
1756
+ },
1757
+ "157110": {
1758
+ "content": "<|reserved_token_215|>",
1759
+ "lstrip": false,
1760
+ "normalized": false,
1761
+ "rstrip": false,
1762
+ "single_word": false,
1763
+ "special": true
1764
+ },
1765
+ "157111": {
1766
+ "content": "<|reserved_token_216|>",
1767
+ "lstrip": false,
1768
+ "normalized": false,
1769
+ "rstrip": false,
1770
+ "single_word": false,
1771
+ "special": true
1772
+ },
1773
+ "157112": {
1774
+ "content": "<|reserved_token_217|>",
1775
+ "lstrip": false,
1776
+ "normalized": false,
1777
+ "rstrip": false,
1778
+ "single_word": false,
1779
+ "special": true
1780
+ },
1781
+ "157113": {
1782
+ "content": "<|reserved_token_218|>",
1783
+ "lstrip": false,
1784
+ "normalized": false,
1785
+ "rstrip": false,
1786
+ "single_word": false,
1787
+ "special": true
1788
+ },
1789
+ "157114": {
1790
+ "content": "<|reserved_token_219|>",
1791
+ "lstrip": false,
1792
+ "normalized": false,
1793
+ "rstrip": false,
1794
+ "single_word": false,
1795
+ "special": true
1796
+ },
1797
+ "157115": {
1798
+ "content": "<|reserved_token_220|>",
1799
+ "lstrip": false,
1800
+ "normalized": false,
1801
+ "rstrip": false,
1802
+ "single_word": false,
1803
+ "special": true
1804
+ },
1805
+ "157116": {
1806
+ "content": "<|reserved_token_221|>",
1807
+ "lstrip": false,
1808
+ "normalized": false,
1809
+ "rstrip": false,
1810
+ "single_word": false,
1811
+ "special": true
1812
+ },
1813
+ "157117": {
1814
+ "content": "<|reserved_token_222|>",
1815
+ "lstrip": false,
1816
+ "normalized": false,
1817
+ "rstrip": false,
1818
+ "single_word": false,
1819
+ "special": true
1820
+ },
1821
+ "157118": {
1822
+ "content": "<|reserved_token_223|>",
1823
+ "lstrip": false,
1824
+ "normalized": false,
1825
+ "rstrip": false,
1826
+ "single_word": false,
1827
+ "special": true
1828
+ },
1829
+ "157119": {
1830
+ "content": "<|reserved_token_224|>",
1831
+ "lstrip": false,
1832
+ "normalized": false,
1833
+ "rstrip": false,
1834
+ "single_word": false,
1835
+ "special": true
1836
+ },
1837
+ "157120": {
1838
+ "content": "<|reserved_token_225|>",
1839
+ "lstrip": false,
1840
+ "normalized": false,
1841
+ "rstrip": false,
1842
+ "single_word": false,
1843
+ "special": true
1844
+ },
1845
+ "157121": {
1846
+ "content": "<|reserved_token_226|>",
1847
+ "lstrip": false,
1848
+ "normalized": false,
1849
+ "rstrip": false,
1850
+ "single_word": false,
1851
+ "special": true
1852
+ },
1853
+ "157122": {
1854
+ "content": "<|reserved_token_227|>",
1855
+ "lstrip": false,
1856
+ "normalized": false,
1857
+ "rstrip": false,
1858
+ "single_word": false,
1859
+ "special": true
1860
+ },
1861
+ "157123": {
1862
+ "content": "<|reserved_token_228|>",
1863
+ "lstrip": false,
1864
+ "normalized": false,
1865
+ "rstrip": false,
1866
+ "single_word": false,
1867
+ "special": true
1868
+ },
1869
+ "157124": {
1870
+ "content": "<|reserved_token_229|>",
1871
+ "lstrip": false,
1872
+ "normalized": false,
1873
+ "rstrip": false,
1874
+ "single_word": false,
1875
+ "special": true
1876
+ },
1877
+ "157125": {
1878
+ "content": "<|reserved_token_230|>",
1879
+ "lstrip": false,
1880
+ "normalized": false,
1881
+ "rstrip": false,
1882
+ "single_word": false,
1883
+ "special": true
1884
+ },
1885
+ "157126": {
1886
+ "content": "<|reserved_token_231|>",
1887
+ "lstrip": false,
1888
+ "normalized": false,
1889
+ "rstrip": false,
1890
+ "single_word": false,
1891
+ "special": true
1892
+ },
1893
+ "157127": {
1894
+ "content": "<|reserved_token_232|>",
1895
+ "lstrip": false,
1896
+ "normalized": false,
1897
+ "rstrip": false,
1898
+ "single_word": false,
1899
+ "special": true
1900
+ },
1901
+ "157128": {
1902
+ "content": "<|reserved_token_233|>",
1903
+ "lstrip": false,
1904
+ "normalized": false,
1905
+ "rstrip": false,
1906
+ "single_word": false,
1907
+ "special": true
1908
+ },
1909
+ "157129": {
1910
+ "content": "<|reserved_token_234|>",
1911
+ "lstrip": false,
1912
+ "normalized": false,
1913
+ "rstrip": false,
1914
+ "single_word": false,
1915
+ "special": true
1916
+ },
1917
+ "157130": {
1918
+ "content": "<|reserved_token_235|>",
1919
+ "lstrip": false,
1920
+ "normalized": false,
1921
+ "rstrip": false,
1922
+ "single_word": false,
1923
+ "special": true
1924
+ },
1925
+ "157131": {
1926
+ "content": "<|reserved_token_236|>",
1927
+ "lstrip": false,
1928
+ "normalized": false,
1929
+ "rstrip": false,
1930
+ "single_word": false,
1931
+ "special": true
1932
+ },
1933
+ "157132": {
1934
+ "content": "<|reserved_token_237|>",
1935
+ "lstrip": false,
1936
+ "normalized": false,
1937
+ "rstrip": false,
1938
+ "single_word": false,
1939
+ "special": true
1940
+ },
1941
+ "157133": {
1942
+ "content": "<|reserved_token_238|>",
1943
+ "lstrip": false,
1944
+ "normalized": false,
1945
+ "rstrip": false,
1946
+ "single_word": false,
1947
+ "special": true
1948
+ },
1949
+ "157134": {
1950
+ "content": "<|reserved_token_239|>",
1951
+ "lstrip": false,
1952
+ "normalized": false,
1953
+ "rstrip": false,
1954
+ "single_word": false,
1955
+ "special": true
1956
+ },
1957
+ "157135": {
1958
+ "content": "<|reserved_token_240|>",
1959
+ "lstrip": false,
1960
+ "normalized": false,
1961
+ "rstrip": false,
1962
+ "single_word": false,
1963
+ "special": true
1964
+ },
1965
+ "157136": {
1966
+ "content": "<|reserved_token_241|>",
1967
+ "lstrip": false,
1968
+ "normalized": false,
1969
+ "rstrip": false,
1970
+ "single_word": false,
1971
+ "special": true
1972
+ },
1973
+ "157137": {
1974
+ "content": "<|reserved_token_242|>",
1975
+ "lstrip": false,
1976
+ "normalized": false,
1977
+ "rstrip": false,
1978
+ "single_word": false,
1979
+ "special": true
1980
+ },
1981
+ "157138": {
1982
+ "content": "<|reserved_token_243|>",
1983
+ "lstrip": false,
1984
+ "normalized": false,
1985
+ "rstrip": false,
1986
+ "single_word": false,
1987
+ "special": true
1988
+ },
1989
+ "157139": {
1990
+ "content": "<|reserved_token_244|>",
1991
+ "lstrip": false,
1992
+ "normalized": false,
1993
+ "rstrip": false,
1994
+ "single_word": false,
1995
+ "special": true
1996
+ },
1997
+ "157140": {
1998
+ "content": "<|reserved_token_245|>",
1999
+ "lstrip": false,
2000
+ "normalized": false,
2001
+ "rstrip": false,
2002
+ "single_word": false,
2003
+ "special": true
2004
+ },
2005
+ "157141": {
2006
+ "content": "<|reserved_token_246|>",
2007
+ "lstrip": false,
2008
+ "normalized": false,
2009
+ "rstrip": false,
2010
+ "single_word": false,
2011
+ "special": true
2012
+ },
2013
+ "157142": {
2014
+ "content": "<|reserved_token_247|>",
2015
+ "lstrip": false,
2016
+ "normalized": false,
2017
+ "rstrip": false,
2018
+ "single_word": false,
2019
+ "special": true
2020
+ },
2021
+ "157143": {
2022
+ "content": "<|reserved_token_248|>",
2023
+ "lstrip": false,
2024
+ "normalized": false,
2025
+ "rstrip": false,
2026
+ "single_word": false,
2027
+ "special": true
2028
+ },
2029
+ "157144": {
2030
+ "content": "<|reserved_token_249|>",
2031
+ "lstrip": false,
2032
+ "normalized": false,
2033
+ "rstrip": false,
2034
+ "single_word": false,
2035
+ "special": true
2036
+ },
2037
+ "157145": {
2038
+ "content": "<|reserved_token_250|>",
2039
+ "lstrip": false,
2040
+ "normalized": false,
2041
+ "rstrip": false,
2042
+ "single_word": false,
2043
+ "special": true
2044
+ },
2045
+ "157146": {
2046
+ "content": "<|reserved_token_251|>",
2047
+ "lstrip": false,
2048
+ "normalized": false,
2049
+ "rstrip": false,
2050
+ "single_word": false,
2051
+ "special": true
2052
+ },
2053
+ "157147": {
2054
+ "content": "<|reserved_token_252|>",
2055
+ "lstrip": false,
2056
+ "normalized": false,
2057
+ "rstrip": false,
2058
+ "single_word": false,
2059
+ "special": true
2060
+ },
2061
+ "157148": {
2062
+ "content": "<|reserved_token_253|>",
2063
+ "lstrip": false,
2064
+ "normalized": false,
2065
+ "rstrip": false,
2066
+ "single_word": false,
2067
+ "special": true
2068
+ },
2069
+ "157149": {
2070
+ "content": "<|reserved_token_254|>",
2071
+ "lstrip": false,
2072
+ "normalized": false,
2073
+ "rstrip": false,
2074
+ "single_word": false,
2075
+ "special": true
2076
+ },
2077
+ "157150": {
2078
+ "content": "<|reserved_token_255|>",
2079
+ "lstrip": false,
2080
+ "normalized": false,
2081
+ "rstrip": false,
2082
+ "single_word": false,
2083
+ "special": true
2084
+ },
2085
+ "157151": {
2086
+ "content": "<role>",
2087
+ "lstrip": false,
2088
+ "normalized": false,
2089
+ "rstrip": false,
2090
+ "single_word": false,
2091
+ "special": true
2092
+ },
2093
+ "157152": {
2094
+ "content": "</role>",
2095
+ "lstrip": false,
2096
+ "normalized": false,
2097
+ "rstrip": false,
2098
+ "single_word": false,
2099
+ "special": true
2100
+ },
2101
+ "157153": {
2102
+ "content": "<|arithmetic_end|>",
2103
+ "lstrip": false,
2104
+ "normalized": false,
2105
+ "rstrip": false,
2106
+ "single_word": false,
2107
+ "special": true
2108
+ },
2109
+ "157154": {
2110
+ "content": "<|number_end|>",
2111
+ "lstrip": false,
2112
+ "normalized": false,
2113
+ "rstrip": false,
2114
+ "single_word": false,
2115
+ "special": true
2116
+ },
2117
+ "157155": {
2118
+ "content": "<|arithmetic_start|>",
2119
+ "lstrip": false,
2120
+ "normalized": false,
2121
+ "rstrip": false,
2122
+ "single_word": false,
2123
+ "special": true
2124
+ },
2125
+ "157156": {
2126
+ "content": "<|number_start|>",
2127
+ "lstrip": false,
2128
+ "normalized": false,
2129
+ "rstrip": false,
2130
+ "single_word": false,
2131
+ "special": true
2132
+ },
2133
+ "157157": {
2134
+ "content": "<imagePatch>",
2135
+ "lstrip": false,
2136
+ "normalized": false,
2137
+ "rstrip": false,
2138
+ "single_word": false,
2139
+ "special": true
2140
+ },
2141
+ "157158": {
2142
+ "content": "<image>",
2143
+ "lstrip": false,
2144
+ "normalized": false,
2145
+ "rstrip": false,
2146
+ "single_word": false,
2147
+ "special": true
2148
+ },
2149
+ "157159": {
2150
+ "content": "</image>",
2151
+ "lstrip": false,
2152
+ "normalized": false,
2153
+ "rstrip": false,
2154
+ "single_word": false,
2155
+ "special": true
2156
+ },
2157
+ "157160": {
2158
+ "content": "<video>",
2159
+ "lstrip": false,
2160
+ "normalized": false,
2161
+ "rstrip": false,
2162
+ "single_word": false,
2163
+ "special": true
2164
+ },
2165
+ "157161": {
2166
+ "content": "</video>",
2167
+ "lstrip": false,
2168
+ "normalized": false,
2169
+ "rstrip": false,
2170
+ "single_word": false,
2171
+ "special": true
2172
+ },
2173
+ "157162": {
2174
+ "content": "<gen_imagePatch>",
2175
+ "lstrip": false,
2176
+ "normalized": false,
2177
+ "rstrip": false,
2178
+ "single_word": false,
2179
+ "special": true
2180
+ },
2181
+ "157163": {
2182
+ "content": "<gen_image>",
2183
+ "lstrip": false,
2184
+ "normalized": false,
2185
+ "rstrip": false,
2186
+ "single_word": false,
2187
+ "special": true
2188
+ },
2189
+ "157164": {
2190
+ "content": "</gen_image>",
2191
+ "lstrip": false,
2192
+ "normalized": false,
2193
+ "rstrip": false,
2194
+ "single_word": false,
2195
+ "special": true
2196
+ },
2197
+ "157165": {
2198
+ "content": "<imageHere>",
2199
+ "lstrip": false,
2200
+ "normalized": false,
2201
+ "rstrip": false,
2202
+ "single_word": false,
2203
+ "special": true
2204
+ },
2205
+ "157166": {
2206
+ "content": "<end_of_chunk>",
2207
+ "lstrip": false,
2208
+ "normalized": false,
2209
+ "rstrip": false,
2210
+ "single_word": false,
2211
+ "special": true
2212
+ },
2213
+ "157167": {
2214
+ "content": "<end_of_audio>",
2215
+ "lstrip": false,
2216
+ "normalized": false,
2217
+ "rstrip": false,
2218
+ "single_word": false,
2219
+ "special": true
2220
+ },
2221
+ "157168": {
2222
+ "content": "<audioPatch>",
2223
+ "lstrip": false,
2224
+ "normalized": false,
2225
+ "rstrip": false,
2226
+ "single_word": false,
2227
+ "special": true
2228
+ },
2229
+ "157169": {
2230
+ "content": "<audio>",
2231
+ "lstrip": false,
2232
+ "normalized": false,
2233
+ "rstrip": false,
2234
+ "single_word": false,
2235
+ "special": true
2236
+ },
2237
+ "157170": {
2238
+ "content": "</audio>",
2239
+ "lstrip": false,
2240
+ "normalized": false,
2241
+ "rstrip": false,
2242
+ "single_word": false,
2243
+ "special": true
2244
+ },
2245
+ "157171": {
2246
+ "content": "<gen_audioPatch>",
2247
+ "lstrip": false,
2248
+ "normalized": false,
2249
+ "rstrip": false,
2250
+ "single_word": false,
2251
+ "special": true
2252
+ },
2253
+ "157172": {
2254
+ "content": "<gen_audio>",
2255
+ "lstrip": false,
2256
+ "normalized": false,
2257
+ "rstrip": false,
2258
+ "single_word": false,
2259
+ "special": true
2260
+ },
2261
+ "157173": {
2262
+ "content": "</gen_audio>",
2263
+ "lstrip": false,
2264
+ "normalized": false,
2265
+ "rstrip": false,
2266
+ "single_word": false,
2267
+ "special": true
2268
+ },
2269
+ "157174": {
2270
+ "content": "<audioHere>",
2271
+ "lstrip": false,
2272
+ "normalized": false,
2273
+ "rstrip": false,
2274
+ "single_word": false,
2275
+ "special": true
2276
+ },
2277
+ "157175": {
2278
+ "content": "<framePatch>",
2279
+ "lstrip": false,
2280
+ "normalized": false,
2281
+ "rstrip": false,
2282
+ "single_word": false,
2283
+ "special": true
2284
+ },
2285
+ "157176": {
2286
+ "content": "<text>",
2287
+ "lstrip": false,
2288
+ "normalized": false,
2289
+ "rstrip": false,
2290
+ "single_word": false,
2291
+ "special": true
2292
+ },
2293
+ "157177": {
2294
+ "content": "<asr>",
2295
+ "lstrip": false,
2296
+ "normalized": false,
2297
+ "rstrip": false,
2298
+ "single_word": false,
2299
+ "special": true
2300
+ },
2301
+ "157178": {
2302
+ "content": "<tts>",
2303
+ "lstrip": false,
2304
+ "normalized": false,
2305
+ "rstrip": false,
2306
+ "single_word": false,
2307
+ "special": true
2308
+ }
2309
+ },
2310
+ "additional_special_tokens": [
2311
+ "<|arithmetic_end|>",
2312
+ "<role>",
2313
+ "<|number_end|>",
2314
+ "<|arithmetic_start|>",
2315
+ "</role>",
2316
+ "<|number_start|>",
2317
+ "<|role_end|>",
2318
+ "<tool_call>",
2319
+ "</tool_call>",
2320
+ "<tool_response>",
2321
+ "</tool_response>"
2322
+ ],
2323
+ "bos_token": "<|startoftext|>",
2324
+ "chat_template": "{% set thinking_option = 'off' %}\n{{- '<role>SYSTEM</role>' }}\n{%- if messages[0].role == 'system' %}\n {{- messages[0].content + '\\n\\n' }}\n{%- endif %}\n{%- if tools %}\n {{- \"# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n {%- for tool in tools %}\n {{- \"\\n\" }}\n {{- tool | tojson }}\n {%- endfor %}\n {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call>\\n\" }}\n{%- endif %}\n{{- 'detailed thinking ' + thinking_option + '<|role_end|>' }}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n {%- set index = (messages|length - 1) - loop.index0 %}\n {%- if ns.multi_step_tool and message.role == \"user\" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}\n {%- set ns.multi_step_tool = false %}\n {%- set ns.last_query_index = index %}\n {%- endif %}\n{%- endfor %}\n{%- for message in messages %}\n {%- if message.content is string %}\n {%- set content = message.content %}\n {%- else %}\n {%- set content = '' %}\n {%- endif %}\n {%- if message.role == \"user\" %}\n {{- '<role>HUMAN</role>' + message.content + '<|role_end|>' }}\n {%- elif message.role == \"system\" and not loop.first %}\n {{- '<role>SYSTEM</role>' + message.content + '<|role_end|>' }}\n {%- elif message.role == \"assistant\" %}\n {%- set reasoning_content = '' %}\n {%- if message.reasoning_content is string %}\n {%- set reasoning_content = message.reasoning_content %}\n {%- else %}\n {%- if '</think>' in content %}\n {%- set reasoning_content = content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n {%- set content = content.split('</think>')[-1].lstrip('\\n') %}\n {%- endif %}\n {%- endif %}\n {%- if loop.index0 > ns.last_query_index %}\n {%- if reasoning_content %}\n {{- '<role>ASSISTANT</role>' + '\\n<think>\\n' + reasoning_content.strip('\\n') + '\\n</think>\\n\\n' + content.lstrip('\\n') }}\n {%- else %}\n {{- '<role>ASSISTANT</role>' + content }}\n {%- endif %}\n {%- else %}\n {{- '<role>ASSISTANT</role>' + content }}\n {%- endif %}\n {%- if message.tool_calls %}\n {%- for tool_call in message.tool_calls %}\n {%- if (loop.first and content) or (not loop.first) %}\n {{- '\\n' }}\n {%- endif %}\n {%- if tool_call.function %}\n {%- set tool_call = tool_call.function %}\n {%- endif %}\n {{- '<tool_call>\\n{\"name\": \"' }}\n {{- tool_call.name }}\n {{- '\", \"arguments\": ' }}\n {%- if tool_call.arguments is string %}\n {{- tool_call.arguments }}\n {%- else %}\n {{- tool_call.arguments | tojson }}\n {%- endif %}\n {{- '}\\n</tool_call>' }}\n {%- endfor %}\n {%- endif %}\n {{- '<|role_end|>' }}\n {%- elif message.role == \"tool\" %}\n {%- if loop.first or (messages[loop.index0 - 1].role != \"tool\") %}\n {{- '<role>OBSERVATION</role>' }}\n {%- endif %}\n {{- '\\n<tool_response>\\n' }}\n {{- content }}\n {{- '\\n</tool_response>' }}\n {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n {{- '<|role_end|>' }}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<role>ASSISTANT</role>' }}\n{%- endif %}",
2325
+ "clean_up_tokenization_spaces": false,
2326
+ "cls_token": "[CLS]",
2327
+ "eos_token": "<|role_end|>",
2328
+ "extra_special_tokens": {},
2329
+ "fast_tokenizer": true,
2330
+ "gmask_token": "[gMASK]",
2331
+ "merges_file": null,
2332
+ "model_max_length": 1000000000000000019884624838656,
2333
+ "pad_token": "<|role_end|>"
2334
+ }
transformer/config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "MingImageTransformer2DModel",
3
+ "_diffusers_version": "0.41.0.dev0",
4
+ "alignment_padding_mode": "zero_masked",
5
+ "auto_map": {
6
+ "AutoModel": "transformer_ming_image.MingImageTransformer2DModel"
7
+ },
8
+ "axes_dims": [
9
+ 32,
10
+ 48,
11
+ 48
12
+ ],
13
+ "axes_lens": [
14
+ 20480,
15
+ 512,
16
+ 512
17
+ ],
18
+ "cap_feat_dim": 2560,
19
+ "dim": 3840,
20
+ "f_patch_size": 1,
21
+ "in_channels": 16,
22
+ "multi_frame_output": false,
23
+ "n_heads": 30,
24
+ "n_layers": 30,
25
+ "n_refiner_layers": 2,
26
+ "norm_eps": 1e-05,
27
+ "patch_size": 2,
28
+ "rope_theta": 256.0,
29
+ "t_scale": 1000.0
30
+ }
transformer/diffusion_pytorch_model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dfb8adf57ad2a9d59544062dcb1ad7546b69d4de855e6a009328e86636a858b3
3
+ size 9973711608
transformer/diffusion_pytorch_model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:400c9d40787dfe0c435128332267f60a8d25808c421d017f7f7fac75c9ee72b7
3
+ size 2336146728
transformer/diffusion_pytorch_model.safetensors.index.json ADDED
@@ -0,0 +1,526 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 12309802112
4
+ },
5
+ "weight_map": {
6
+ "all_final_layer.2-1.adaLN_modulation.1.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
7
+ "all_final_layer.2-1.adaLN_modulation.1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
8
+ "all_final_layer.2-1.linear.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
9
+ "all_final_layer.2-1.linear.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
10
+ "all_x_embedder.2-1.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
11
+ "all_x_embedder.2-1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
12
+ "cap_embedder.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
13
+ "cap_embedder.1.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
14
+ "cap_embedder.1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
15
+ "context_refiner.0.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
16
+ "context_refiner.0.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
17
+ "context_refiner.0.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
18
+ "context_refiner.0.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
19
+ "context_refiner.0.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
20
+ "context_refiner.0.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
21
+ "context_refiner.0.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
22
+ "context_refiner.0.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
23
+ "context_refiner.0.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
24
+ "context_refiner.0.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
25
+ "context_refiner.0.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
26
+ "context_refiner.0.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
27
+ "context_refiner.0.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
28
+ "context_refiner.1.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
29
+ "context_refiner.1.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
30
+ "context_refiner.1.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
31
+ "context_refiner.1.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
32
+ "context_refiner.1.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
33
+ "context_refiner.1.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
34
+ "context_refiner.1.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
35
+ "context_refiner.1.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
36
+ "context_refiner.1.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
37
+ "context_refiner.1.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
38
+ "context_refiner.1.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
39
+ "context_refiner.1.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
40
+ "context_refiner.1.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
41
+ "layers.0.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
42
+ "layers.0.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
43
+ "layers.0.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
44
+ "layers.0.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
45
+ "layers.0.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
46
+ "layers.0.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
47
+ "layers.0.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
48
+ "layers.0.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
49
+ "layers.0.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
50
+ "layers.0.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
51
+ "layers.0.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
52
+ "layers.0.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
53
+ "layers.0.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
54
+ "layers.0.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
55
+ "layers.0.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
56
+ "layers.1.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
57
+ "layers.1.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
58
+ "layers.1.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
59
+ "layers.1.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
60
+ "layers.1.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
61
+ "layers.1.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
62
+ "layers.1.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
63
+ "layers.1.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
64
+ "layers.1.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
65
+ "layers.1.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
66
+ "layers.1.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
67
+ "layers.1.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
68
+ "layers.1.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
69
+ "layers.1.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
70
+ "layers.1.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
71
+ "layers.10.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
72
+ "layers.10.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
73
+ "layers.10.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
74
+ "layers.10.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
75
+ "layers.10.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
76
+ "layers.10.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
77
+ "layers.10.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
78
+ "layers.10.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
79
+ "layers.10.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
80
+ "layers.10.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
81
+ "layers.10.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
82
+ "layers.10.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
83
+ "layers.10.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
84
+ "layers.10.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
85
+ "layers.10.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
86
+ "layers.11.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
87
+ "layers.11.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
88
+ "layers.11.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
89
+ "layers.11.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
90
+ "layers.11.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
91
+ "layers.11.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
92
+ "layers.11.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
93
+ "layers.11.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
94
+ "layers.11.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
95
+ "layers.11.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
96
+ "layers.11.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
97
+ "layers.11.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
98
+ "layers.11.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
99
+ "layers.11.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
100
+ "layers.11.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
101
+ "layers.12.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
102
+ "layers.12.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
103
+ "layers.12.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
104
+ "layers.12.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
105
+ "layers.12.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
106
+ "layers.12.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
107
+ "layers.12.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
108
+ "layers.12.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
109
+ "layers.12.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
110
+ "layers.12.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
111
+ "layers.12.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
112
+ "layers.12.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
113
+ "layers.12.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
114
+ "layers.12.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
115
+ "layers.12.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
116
+ "layers.13.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
117
+ "layers.13.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
118
+ "layers.13.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
119
+ "layers.13.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
120
+ "layers.13.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
121
+ "layers.13.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
122
+ "layers.13.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
123
+ "layers.13.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
124
+ "layers.13.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
125
+ "layers.13.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
126
+ "layers.13.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
127
+ "layers.13.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
128
+ "layers.13.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
129
+ "layers.13.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
130
+ "layers.13.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
131
+ "layers.14.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
132
+ "layers.14.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
133
+ "layers.14.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
134
+ "layers.14.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
135
+ "layers.14.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
136
+ "layers.14.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
137
+ "layers.14.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
138
+ "layers.14.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
139
+ "layers.14.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
140
+ "layers.14.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
141
+ "layers.14.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
142
+ "layers.14.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
143
+ "layers.14.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
144
+ "layers.14.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
145
+ "layers.14.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
146
+ "layers.15.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
147
+ "layers.15.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
148
+ "layers.15.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
149
+ "layers.15.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
150
+ "layers.15.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
151
+ "layers.15.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
152
+ "layers.15.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
153
+ "layers.15.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
154
+ "layers.15.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
155
+ "layers.15.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
156
+ "layers.15.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
157
+ "layers.15.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
158
+ "layers.15.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
159
+ "layers.15.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
160
+ "layers.15.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
161
+ "layers.16.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
162
+ "layers.16.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
163
+ "layers.16.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
164
+ "layers.16.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
165
+ "layers.16.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
166
+ "layers.16.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
167
+ "layers.16.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
168
+ "layers.16.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
169
+ "layers.16.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
170
+ "layers.16.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
171
+ "layers.16.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
172
+ "layers.16.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
173
+ "layers.16.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
174
+ "layers.16.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
175
+ "layers.16.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
176
+ "layers.17.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
177
+ "layers.17.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
178
+ "layers.17.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
179
+ "layers.17.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
180
+ "layers.17.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
181
+ "layers.17.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
182
+ "layers.17.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
183
+ "layers.17.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
184
+ "layers.17.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
185
+ "layers.17.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
186
+ "layers.17.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
187
+ "layers.17.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
188
+ "layers.17.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
189
+ "layers.17.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
190
+ "layers.17.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
191
+ "layers.18.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
192
+ "layers.18.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
193
+ "layers.18.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
194
+ "layers.18.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
195
+ "layers.18.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
196
+ "layers.18.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
197
+ "layers.18.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
198
+ "layers.18.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
199
+ "layers.18.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
200
+ "layers.18.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
201
+ "layers.18.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
202
+ "layers.18.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
203
+ "layers.18.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
204
+ "layers.18.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
205
+ "layers.18.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
206
+ "layers.19.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
207
+ "layers.19.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
208
+ "layers.19.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
209
+ "layers.19.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
210
+ "layers.19.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
211
+ "layers.19.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
212
+ "layers.19.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
213
+ "layers.19.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
214
+ "layers.19.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
215
+ "layers.19.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
216
+ "layers.19.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
217
+ "layers.19.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
218
+ "layers.19.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
219
+ "layers.19.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
220
+ "layers.19.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
221
+ "layers.2.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
222
+ "layers.2.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
223
+ "layers.2.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
224
+ "layers.2.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
225
+ "layers.2.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
226
+ "layers.2.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
227
+ "layers.2.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
228
+ "layers.2.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
229
+ "layers.2.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
230
+ "layers.2.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
231
+ "layers.2.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
232
+ "layers.2.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
233
+ "layers.2.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
234
+ "layers.2.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
235
+ "layers.2.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
236
+ "layers.20.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
237
+ "layers.20.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
238
+ "layers.20.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
239
+ "layers.20.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
240
+ "layers.20.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
241
+ "layers.20.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
242
+ "layers.20.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
243
+ "layers.20.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
244
+ "layers.20.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
245
+ "layers.20.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
246
+ "layers.20.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
247
+ "layers.20.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
248
+ "layers.20.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
249
+ "layers.20.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
250
+ "layers.20.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
251
+ "layers.21.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
252
+ "layers.21.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
253
+ "layers.21.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
254
+ "layers.21.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
255
+ "layers.21.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
256
+ "layers.21.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
257
+ "layers.21.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
258
+ "layers.21.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
259
+ "layers.21.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
260
+ "layers.21.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
261
+ "layers.21.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
262
+ "layers.21.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
263
+ "layers.21.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
264
+ "layers.21.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
265
+ "layers.21.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
266
+ "layers.22.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
267
+ "layers.22.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
268
+ "layers.22.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
269
+ "layers.22.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
270
+ "layers.22.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
271
+ "layers.22.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
272
+ "layers.22.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
273
+ "layers.22.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
274
+ "layers.22.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
275
+ "layers.22.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
276
+ "layers.22.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
277
+ "layers.22.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
278
+ "layers.22.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
279
+ "layers.22.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
280
+ "layers.22.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
281
+ "layers.23.adaLN_modulation.0.bias": "diffusion_pytorch_model-00002-of-00002.safetensors",
282
+ "layers.23.adaLN_modulation.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
283
+ "layers.23.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
284
+ "layers.23.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
285
+ "layers.23.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
286
+ "layers.23.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
287
+ "layers.23.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
288
+ "layers.23.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
289
+ "layers.23.attention_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
290
+ "layers.23.attention_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
291
+ "layers.23.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
292
+ "layers.23.feed_forward.w2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
293
+ "layers.23.feed_forward.w3.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
294
+ "layers.23.ffn_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
295
+ "layers.23.ffn_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
296
+ "layers.24.adaLN_modulation.0.bias": "diffusion_pytorch_model-00002-of-00002.safetensors",
297
+ "layers.24.adaLN_modulation.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
298
+ "layers.24.attention.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
299
+ "layers.24.attention.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
300
+ "layers.24.attention.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
301
+ "layers.24.attention.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
302
+ "layers.24.attention.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
303
+ "layers.24.attention.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
304
+ "layers.24.attention_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
305
+ "layers.24.attention_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
306
+ "layers.24.feed_forward.w1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
307
+ "layers.24.feed_forward.w2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
308
+ "layers.24.feed_forward.w3.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
309
+ "layers.24.ffn_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
310
+ "layers.24.ffn_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
311
+ "layers.25.adaLN_modulation.0.bias": "diffusion_pytorch_model-00002-of-00002.safetensors",
312
+ "layers.25.adaLN_modulation.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
313
+ "layers.25.attention.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
314
+ "layers.25.attention.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
315
+ "layers.25.attention.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
316
+ "layers.25.attention.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
317
+ "layers.25.attention.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
318
+ "layers.25.attention.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
319
+ "layers.25.attention_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
320
+ "layers.25.attention_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
321
+ "layers.25.feed_forward.w1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
322
+ "layers.25.feed_forward.w2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
323
+ "layers.25.feed_forward.w3.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
324
+ "layers.25.ffn_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
325
+ "layers.25.ffn_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
326
+ "layers.26.adaLN_modulation.0.bias": "diffusion_pytorch_model-00002-of-00002.safetensors",
327
+ "layers.26.adaLN_modulation.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
328
+ "layers.26.attention.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
329
+ "layers.26.attention.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
330
+ "layers.26.attention.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
331
+ "layers.26.attention.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
332
+ "layers.26.attention.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
333
+ "layers.26.attention.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
334
+ "layers.26.attention_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
335
+ "layers.26.attention_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
336
+ "layers.26.feed_forward.w1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
337
+ "layers.26.feed_forward.w2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
338
+ "layers.26.feed_forward.w3.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
339
+ "layers.26.ffn_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
340
+ "layers.26.ffn_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
341
+ "layers.27.adaLN_modulation.0.bias": "diffusion_pytorch_model-00002-of-00002.safetensors",
342
+ "layers.27.adaLN_modulation.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
343
+ "layers.27.attention.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
344
+ "layers.27.attention.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
345
+ "layers.27.attention.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
346
+ "layers.27.attention.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
347
+ "layers.27.attention.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
348
+ "layers.27.attention.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
349
+ "layers.27.attention_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
350
+ "layers.27.attention_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
351
+ "layers.27.feed_forward.w1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
352
+ "layers.27.feed_forward.w2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
353
+ "layers.27.feed_forward.w3.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
354
+ "layers.27.ffn_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
355
+ "layers.27.ffn_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
356
+ "layers.28.adaLN_modulation.0.bias": "diffusion_pytorch_model-00002-of-00002.safetensors",
357
+ "layers.28.adaLN_modulation.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
358
+ "layers.28.attention.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
359
+ "layers.28.attention.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
360
+ "layers.28.attention.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
361
+ "layers.28.attention.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
362
+ "layers.28.attention.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
363
+ "layers.28.attention.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
364
+ "layers.28.attention_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
365
+ "layers.28.attention_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
366
+ "layers.28.feed_forward.w1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
367
+ "layers.28.feed_forward.w2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
368
+ "layers.28.feed_forward.w3.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
369
+ "layers.28.ffn_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
370
+ "layers.28.ffn_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
371
+ "layers.29.adaLN_modulation.0.bias": "diffusion_pytorch_model-00002-of-00002.safetensors",
372
+ "layers.29.adaLN_modulation.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
373
+ "layers.29.attention.norm_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
374
+ "layers.29.attention.norm_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
375
+ "layers.29.attention.to_k.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
376
+ "layers.29.attention.to_out.0.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
377
+ "layers.29.attention.to_q.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
378
+ "layers.29.attention.to_v.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
379
+ "layers.29.attention_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
380
+ "layers.29.attention_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
381
+ "layers.29.feed_forward.w1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
382
+ "layers.29.feed_forward.w2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
383
+ "layers.29.feed_forward.w3.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
384
+ "layers.29.ffn_norm1.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
385
+ "layers.29.ffn_norm2.weight": "diffusion_pytorch_model-00002-of-00002.safetensors",
386
+ "layers.3.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
387
+ "layers.3.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
388
+ "layers.3.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
389
+ "layers.3.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
390
+ "layers.3.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
391
+ "layers.3.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
392
+ "layers.3.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
393
+ "layers.3.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
394
+ "layers.3.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
395
+ "layers.3.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
396
+ "layers.3.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
397
+ "layers.3.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
398
+ "layers.3.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
399
+ "layers.3.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
400
+ "layers.3.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
401
+ "layers.4.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
402
+ "layers.4.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
403
+ "layers.4.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
404
+ "layers.4.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
405
+ "layers.4.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
406
+ "layers.4.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
407
+ "layers.4.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
408
+ "layers.4.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
409
+ "layers.4.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
410
+ "layers.4.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
411
+ "layers.4.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
412
+ "layers.4.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
413
+ "layers.4.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
414
+ "layers.4.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
415
+ "layers.4.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
416
+ "layers.5.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
417
+ "layers.5.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
418
+ "layers.5.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
419
+ "layers.5.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
420
+ "layers.5.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
421
+ "layers.5.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
422
+ "layers.5.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
423
+ "layers.5.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
424
+ "layers.5.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
425
+ "layers.5.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
426
+ "layers.5.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
427
+ "layers.5.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
428
+ "layers.5.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
429
+ "layers.5.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
430
+ "layers.5.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
431
+ "layers.6.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
432
+ "layers.6.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
433
+ "layers.6.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
434
+ "layers.6.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
435
+ "layers.6.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
436
+ "layers.6.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
437
+ "layers.6.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
438
+ "layers.6.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
439
+ "layers.6.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
440
+ "layers.6.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
441
+ "layers.6.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
442
+ "layers.6.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
443
+ "layers.6.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
444
+ "layers.6.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
445
+ "layers.6.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
446
+ "layers.7.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
447
+ "layers.7.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
448
+ "layers.7.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
449
+ "layers.7.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
450
+ "layers.7.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
451
+ "layers.7.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
452
+ "layers.7.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
453
+ "layers.7.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
454
+ "layers.7.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
455
+ "layers.7.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
456
+ "layers.7.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
457
+ "layers.7.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
458
+ "layers.7.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
459
+ "layers.7.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
460
+ "layers.7.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
461
+ "layers.8.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
462
+ "layers.8.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
463
+ "layers.8.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
464
+ "layers.8.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
465
+ "layers.8.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
466
+ "layers.8.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
467
+ "layers.8.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
468
+ "layers.8.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
469
+ "layers.8.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
470
+ "layers.8.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
471
+ "layers.8.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
472
+ "layers.8.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
473
+ "layers.8.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
474
+ "layers.8.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
475
+ "layers.8.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
476
+ "layers.9.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
477
+ "layers.9.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
478
+ "layers.9.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
479
+ "layers.9.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
480
+ "layers.9.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
481
+ "layers.9.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
482
+ "layers.9.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
483
+ "layers.9.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
484
+ "layers.9.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
485
+ "layers.9.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
486
+ "layers.9.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
487
+ "layers.9.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
488
+ "layers.9.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
489
+ "layers.9.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
490
+ "layers.9.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
491
+ "noise_refiner.0.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
492
+ "noise_refiner.0.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
493
+ "noise_refiner.0.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
494
+ "noise_refiner.0.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
495
+ "noise_refiner.0.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
496
+ "noise_refiner.0.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
497
+ "noise_refiner.0.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
498
+ "noise_refiner.0.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
499
+ "noise_refiner.0.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
500
+ "noise_refiner.0.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
501
+ "noise_refiner.0.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
502
+ "noise_refiner.0.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
503
+ "noise_refiner.0.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
504
+ "noise_refiner.0.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
505
+ "noise_refiner.0.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
506
+ "noise_refiner.1.adaLN_modulation.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
507
+ "noise_refiner.1.adaLN_modulation.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
508
+ "noise_refiner.1.attention.norm_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
509
+ "noise_refiner.1.attention.norm_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
510
+ "noise_refiner.1.attention.to_k.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
511
+ "noise_refiner.1.attention.to_out.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
512
+ "noise_refiner.1.attention.to_q.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
513
+ "noise_refiner.1.attention.to_v.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
514
+ "noise_refiner.1.attention_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
515
+ "noise_refiner.1.attention_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
516
+ "noise_refiner.1.feed_forward.w1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
517
+ "noise_refiner.1.feed_forward.w2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
518
+ "noise_refiner.1.feed_forward.w3.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
519
+ "noise_refiner.1.ffn_norm1.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
520
+ "noise_refiner.1.ffn_norm2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
521
+ "t_embedder.mlp.0.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
522
+ "t_embedder.mlp.0.weight": "diffusion_pytorch_model-00001-of-00002.safetensors",
523
+ "t_embedder.mlp.2.bias": "diffusion_pytorch_model-00001-of-00002.safetensors",
524
+ "t_embedder.mlp.2.weight": "diffusion_pytorch_model-00001-of-00002.safetensors"
525
+ }
526
+ }
transformer/transformer_ming_image.py ADDED
@@ -0,0 +1,503 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright 2026 inclusionAI and The HuggingFace Team. All rights reserved.
2
+ #
3
+ # Licensed under the MIT License. See the LICENSE file in this repository.
4
+
5
+ import math
6
+
7
+ import torch
8
+ import torch.nn as nn
9
+ import torch.nn.functional as F
10
+ from torch.nn.utils.rnn import pad_sequence
11
+
12
+ from diffusers.configuration_utils import ConfigMixin, register_to_config
13
+ from diffusers.models.attention import AttentionModuleMixin
14
+ from diffusers.models.attention_dispatch import dispatch_attention_fn
15
+ from diffusers.models.modeling_outputs import Transformer2DModelOutput
16
+ from diffusers.models.modeling_utils import ModelMixin
17
+ from diffusers.models.normalization import RMSNorm
18
+ from diffusers.utils.torch_utils import maybe_allow_in_graph
19
+
20
+
21
+ ADALN_EMBED_DIM = 256
22
+ SEQ_MULTI_OF = 32
23
+ PADDING_MODES = ("zero_masked", "learned")
24
+
25
+
26
+ # Copied from diffusers.models.transformers.transformer_z_image.TimestepEmbedder
27
+ class TimestepEmbedder(nn.Module):
28
+ def __init__(self, out_size, mid_size=None, frequency_embedding_size=256):
29
+ super().__init__()
30
+ if mid_size is None:
31
+ mid_size = out_size
32
+ self.mlp = nn.Sequential(
33
+ nn.Linear(frequency_embedding_size, mid_size, bias=True),
34
+ nn.SiLU(),
35
+ nn.Linear(mid_size, out_size, bias=True),
36
+ )
37
+
38
+ self.frequency_embedding_size = frequency_embedding_size
39
+
40
+ @staticmethod
41
+ def timestep_embedding(t, dim, max_period=10000):
42
+ with torch.amp.autocast("cuda", enabled=False):
43
+ half = dim // 2
44
+ freqs = torch.exp(
45
+ -math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32, device=t.device) / half
46
+ )
47
+ args = t[:, None].float() * freqs[None]
48
+ embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
49
+ if dim % 2:
50
+ embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
51
+ return embedding
52
+
53
+ def forward(self, t) -> torch.Tensor:
54
+ t_freq = self.timestep_embedding(t, self.frequency_embedding_size)
55
+ weight_dtype = self.mlp[0].weight.dtype
56
+ compute_dtype = getattr(self.mlp[0], "compute_dtype", None)
57
+ if weight_dtype.is_floating_point:
58
+ t_freq = t_freq.to(weight_dtype)
59
+ elif compute_dtype is not None:
60
+ t_freq = t_freq.to(compute_dtype)
61
+ t_emb = self.mlp(t_freq)
62
+ return t_emb
63
+
64
+
65
+ class MingImageAttnProcessor:
66
+ _attention_backend = None
67
+ _parallel_config = None
68
+
69
+ def __call__(
70
+ self,
71
+ attn: "MingImageAttention",
72
+ hidden_states: torch.Tensor,
73
+ attention_mask: torch.Tensor | None = None,
74
+ freqs_cis: torch.Tensor | None = None,
75
+ ) -> torch.Tensor:
76
+ query = attn.to_q(hidden_states).unflatten(-1, (attn.heads, -1))
77
+ key = attn.to_k(hidden_states).unflatten(-1, (attn.heads, -1))
78
+ value = attn.to_v(hidden_states).unflatten(-1, (attn.heads, -1))
79
+
80
+ query = attn.norm_q(query)
81
+ key = attn.norm_k(key)
82
+
83
+ def apply_rotary_emb(x_in: torch.Tensor, freqs_cis: torch.Tensor) -> torch.Tensor:
84
+ with torch.amp.autocast("cuda", enabled=False):
85
+ x = torch.view_as_complex(x_in.float().reshape(*x_in.shape[:-1], -1, 2))
86
+ x_out = torch.view_as_real(x * freqs_cis.unsqueeze(2)).flatten(3)
87
+ return x_out.type_as(x_in)
88
+
89
+ if freqs_cis is not None:
90
+ query = apply_rotary_emb(query, freqs_cis)
91
+ key = apply_rotary_emb(key, freqs_cis)
92
+
93
+ # (batch, seq_len) -> (batch, 1, 1, seq_len), broadcast over heads and queries
94
+ if attention_mask is not None:
95
+ attention_mask = attention_mask[:, None, None, :]
96
+
97
+ dtype = query.dtype
98
+ hidden_states = dispatch_attention_fn(
99
+ query,
100
+ key,
101
+ value,
102
+ attn_mask=attention_mask,
103
+ dropout_p=0.0,
104
+ is_causal=False,
105
+ backend=self._attention_backend,
106
+ parallel_config=self._parallel_config,
107
+ )
108
+ hidden_states = hidden_states.flatten(2, 3).to(dtype)
109
+
110
+ hidden_states = attn.to_out[0](hidden_states)
111
+ hidden_states = attn.to_out[1](hidden_states)
112
+ return hidden_states
113
+
114
+
115
+ class MingImageAttention(nn.Module, AttentionModuleMixin):
116
+ _default_processor_cls = MingImageAttnProcessor
117
+ _available_processors = [MingImageAttnProcessor]
118
+
119
+ def __init__(self, dim: int, heads: int, eps: float = 1e-5, processor=None):
120
+ super().__init__()
121
+ head_dim = dim // heads
122
+ self.heads = heads
123
+
124
+ self.to_q = nn.Linear(dim, dim, bias=False)
125
+ self.to_k = nn.Linear(dim, dim, bias=False)
126
+ self.to_v = nn.Linear(dim, dim, bias=False)
127
+ self.norm_q = RMSNorm(head_dim, eps=eps)
128
+ self.norm_k = RMSNorm(head_dim, eps=eps)
129
+ self.to_out = nn.ModuleList([nn.Linear(dim, dim, bias=False), nn.Dropout(0.0)])
130
+
131
+ if processor is None:
132
+ processor = self._default_processor_cls()
133
+ self.set_processor(processor)
134
+
135
+ def forward(
136
+ self,
137
+ hidden_states: torch.Tensor,
138
+ attention_mask: torch.Tensor | None = None,
139
+ freqs_cis: torch.Tensor | None = None,
140
+ ) -> torch.Tensor:
141
+ return self.processor(self, hidden_states, attention_mask, freqs_cis)
142
+
143
+
144
+ # Copied from diffusers.models.transformers.transformer_z_image.FeedForward
145
+ class FeedForward(nn.Module):
146
+ def __init__(self, dim: int, hidden_dim: int):
147
+ super().__init__()
148
+ self.w1 = nn.Linear(dim, hidden_dim, bias=False)
149
+ self.w2 = nn.Linear(hidden_dim, dim, bias=False)
150
+ self.w3 = nn.Linear(dim, hidden_dim, bias=False)
151
+
152
+ def _forward_silu_gating(self, x1, x3):
153
+ return F.silu(x1) * x3
154
+
155
+ def forward(self, x) -> torch.Tensor:
156
+ return self.w2(self._forward_silu_gating(self.w1(x), self.w3(x)))
157
+
158
+
159
+ @maybe_allow_in_graph
160
+ class MingImageTransformerBlock(nn.Module):
161
+ def __init__(self, dim: int, n_heads: int, norm_eps: float, modulation: bool = True):
162
+ super().__init__()
163
+ self.attention = MingImageAttention(dim, n_heads, eps=1e-5)
164
+ self.feed_forward = FeedForward(dim=dim, hidden_dim=int(dim / 3 * 8))
165
+
166
+ self.attention_norm1 = RMSNorm(dim, eps=norm_eps)
167
+ self.ffn_norm1 = RMSNorm(dim, eps=norm_eps)
168
+ self.attention_norm2 = RMSNorm(dim, eps=norm_eps)
169
+ self.ffn_norm2 = RMSNorm(dim, eps=norm_eps)
170
+
171
+ self.modulation = modulation
172
+ if modulation:
173
+ self.adaLN_modulation = nn.Sequential(nn.Linear(min(dim, ADALN_EMBED_DIM), 4 * dim, bias=True))
174
+
175
+ def forward(
176
+ self,
177
+ x: torch.Tensor,
178
+ attn_mask: torch.Tensor | None,
179
+ freqs_cis: torch.Tensor,
180
+ adaln_input: torch.Tensor | None = None,
181
+ ) -> torch.Tensor:
182
+ if self.modulation:
183
+ scale_msa, gate_msa, scale_mlp, gate_mlp = self.adaLN_modulation(adaln_input).unsqueeze(1).chunk(4, dim=2)
184
+ gate_msa, gate_mlp = gate_msa.tanh(), gate_mlp.tanh()
185
+ scale_msa, scale_mlp = 1.0 + scale_msa, 1.0 + scale_mlp
186
+
187
+ attn_out = self.attention(
188
+ self.attention_norm1(x) * scale_msa, attention_mask=attn_mask, freqs_cis=freqs_cis
189
+ )
190
+ x = x + gate_msa * self.attention_norm2(attn_out)
191
+ x = x + gate_mlp * self.ffn_norm2(self.feed_forward(self.ffn_norm1(x) * scale_mlp))
192
+ else:
193
+ attn_out = self.attention(self.attention_norm1(x), attention_mask=attn_mask, freqs_cis=freqs_cis)
194
+ x = x + self.attention_norm2(attn_out)
195
+ x = x + self.ffn_norm2(self.feed_forward(self.ffn_norm1(x)))
196
+
197
+ return x
198
+
199
+
200
+ class MingImageFinalLayer(nn.Module):
201
+ def __init__(self, hidden_size, out_channels):
202
+ super().__init__()
203
+ self.norm_final = nn.LayerNorm(hidden_size, elementwise_affine=False, eps=1e-6)
204
+ self.linear = nn.Linear(hidden_size, out_channels, bias=True)
205
+ self.adaLN_modulation = nn.Sequential(
206
+ nn.SiLU(),
207
+ nn.Linear(min(hidden_size, ADALN_EMBED_DIM), hidden_size, bias=True),
208
+ )
209
+
210
+ def forward(self, x: torch.Tensor, c: torch.Tensor) -> torch.Tensor:
211
+ scale = 1.0 + self.adaLN_modulation(c)
212
+ x = self.norm_final(x) * scale.unsqueeze(1)
213
+ return self.linear(x)
214
+
215
+
216
+ class MingImageRopeEmbedder:
217
+ def __init__(self, theta: float, axes_dims: list[int], axes_lens: list[int]):
218
+ self.theta = theta
219
+ self.axes_dims = axes_dims
220
+ self.axes_lens = axes_lens
221
+ self.freqs_cis = None
222
+
223
+ @staticmethod
224
+ def precompute_freqs_cis(dim: list[int], end: list[int], theta: float):
225
+ # Precomputed once on CPU in float64: the frame axis reaches position 20480, where float32 angles drift
226
+ # by ~1e-3 rad. The complex64 tables are then moved to the model's device.
227
+ freqs_cis = []
228
+ for d, e in zip(dim, end):
229
+ freqs = 1.0 / (theta ** (torch.arange(0, d, 2, dtype=torch.float64, device="cpu") / d))
230
+ timestep = torch.arange(e, dtype=torch.float64, device="cpu")
231
+ freqs = torch.outer(timestep, freqs).float()
232
+ freqs_cis.append(torch.polar(torch.ones_like(freqs), freqs).to(torch.complex64))
233
+ return freqs_cis
234
+
235
+ def __call__(self, ids: torch.Tensor) -> torch.Tensor:
236
+ if self.freqs_cis is None:
237
+ self.freqs_cis = self.precompute_freqs_cis(self.axes_dims, self.axes_lens, self.theta)
238
+ if self.freqs_cis[0].device != ids.device:
239
+ self.freqs_cis = [freqs_cis.to(ids.device) for freqs_cis in self.freqs_cis]
240
+ return torch.cat([self.freqs_cis[i][ids[:, i]] for i in range(len(self.axes_dims))], dim=-1)
241
+
242
+
243
+ def create_coordinate_grid(size, start, device):
244
+ axes = [torch.arange(x0, x0 + span, dtype=torch.int32, device=device) for x0, span in zip(start, size)]
245
+ return torch.stack(torch.meshgrid(axes, indexing="ij"), dim=-1).flatten(0, -2)
246
+
247
+
248
+ class MingImageTransformer2DModel(ModelMixin, ConfigMixin):
249
+ r"""
250
+ The single-stream diffusion transformer used by Ming-Image.
251
+
252
+ It is the Z-Image transformer with three additions: a second caption stream (`encoder_hidden_states_2`) that is
253
+ already in model width and skips `cap_embedder`, reference latents concatenated along the frame axis
254
+ (`ref_hidden_states`), and a choice of how sequence-alignment padding is handled (`alignment_padding_mode`).
255
+
256
+ Args:
257
+ patch_size (`int`, defaults to `2`): Spatial patch size.
258
+ f_patch_size (`int`, defaults to `1`): Frame patch size.
259
+ in_channels (`int`, defaults to `16`): Latent channels.
260
+ dim (`int`, defaults to `3840`): Model width.
261
+ n_layers (`int`, defaults to `30`): Number of main transformer blocks.
262
+ n_refiner_layers (`int`, defaults to `2`): Number of noise and context refiner blocks each.
263
+ n_heads (`int`, defaults to `30`): Number of attention heads.
264
+ norm_eps (`float`, defaults to `1e-5`): Epsilon of the RMS norms.
265
+ cap_feat_dim (`int`, defaults to `2560`): Width of `encoder_hidden_states`.
266
+ rope_theta (`float`, defaults to `256.0`): RoPE base.
267
+ t_scale (`float`, defaults to `1000.0`): Multiplier applied to the timestep before embedding.
268
+ axes_dims (`list[int]`, defaults to `[32, 48, 48]`): RoPE dimensions of the frame, height and width axes.
269
+ axes_lens (`list[int]`, defaults to `[20480, 512, 512]`): RoPE table length of each axis.
270
+ alignment_padding_mode (`str`, defaults to `"zero_masked"`):
271
+ How the tokens that pad each sequence to a multiple of 32 are handled. `"zero_masked"` zeroes them and
272
+ masks them out of attention; `"learned"` replaces them with learned pad tokens that stay attended.
273
+ multi_frame_output (`bool`, defaults to `False`):
274
+ Return every frame of the latent sequence instead of only the first one.
275
+ """
276
+
277
+ _supports_gradient_checkpointing = True
278
+ _no_split_modules = ["MingImageTransformerBlock"]
279
+ _repeated_blocks = ["MingImageTransformerBlock"]
280
+ _skip_layerwise_casting_patterns = ["t_embedder", "cap_embedder"]
281
+
282
+ @register_to_config
283
+ def __init__(
284
+ self,
285
+ patch_size: int = 2,
286
+ f_patch_size: int = 1,
287
+ in_channels: int = 16,
288
+ dim: int = 3840,
289
+ n_layers: int = 30,
290
+ n_refiner_layers: int = 2,
291
+ n_heads: int = 30,
292
+ norm_eps: float = 1e-5,
293
+ cap_feat_dim: int = 2560,
294
+ rope_theta: float = 256.0,
295
+ t_scale: float = 1000.0,
296
+ axes_dims: list[int] = [32, 48, 48],
297
+ axes_lens: list[int] = [20480, 512, 512],
298
+ alignment_padding_mode: str = "zero_masked",
299
+ multi_frame_output: bool = False,
300
+ ) -> None:
301
+ super().__init__()
302
+ if alignment_padding_mode not in PADDING_MODES:
303
+ raise ValueError(f"alignment_padding_mode must be one of {PADDING_MODES}, got {alignment_padding_mode!r}")
304
+ if dim // n_heads != sum(axes_dims):
305
+ raise ValueError(f"head dim {dim // n_heads} must equal sum(axes_dims)={sum(axes_dims)}")
306
+
307
+ self.out_channels = in_channels
308
+ self.gradient_checkpointing = False
309
+
310
+ patch_key = f"{patch_size}-{f_patch_size}"
311
+ patch_dim = f_patch_size * patch_size * patch_size * in_channels
312
+ self.all_x_embedder = nn.ModuleDict({patch_key: nn.Linear(patch_dim, dim, bias=True)})
313
+ self.all_final_layer = nn.ModuleDict({patch_key: MingImageFinalLayer(dim, patch_dim)})
314
+
315
+ self.noise_refiner = nn.ModuleList(
316
+ [MingImageTransformerBlock(dim, n_heads, norm_eps, modulation=True) for _ in range(n_refiner_layers)]
317
+ )
318
+ self.context_refiner = nn.ModuleList(
319
+ [MingImageTransformerBlock(dim, n_heads, norm_eps, modulation=False) for _ in range(n_refiner_layers)]
320
+ )
321
+ self.t_embedder = TimestepEmbedder(min(dim, ADALN_EMBED_DIM), mid_size=1024)
322
+ self.cap_embedder = nn.Sequential(RMSNorm(cap_feat_dim, eps=norm_eps), nn.Linear(cap_feat_dim, dim, bias=True))
323
+
324
+ if alignment_padding_mode == "learned":
325
+ self.x_pad_token = nn.Parameter(torch.zeros((1, dim)))
326
+ self.cap_pad_token = nn.Parameter(torch.zeros((1, dim)))
327
+
328
+ self.layers = nn.ModuleList([MingImageTransformerBlock(dim, n_heads, norm_eps) for _ in range(n_layers)])
329
+ self.rope_embedder = MingImageRopeEmbedder(theta=rope_theta, axes_dims=axes_dims, axes_lens=axes_lens)
330
+
331
+ def _attention_mask(self, keep: list[torch.Tensor]) -> torch.Tensor | None:
332
+ """Boolean (batch, max_len) mask from per-item boolean `keep` vectors; `None` when nothing is masked."""
333
+ mask = pad_sequence(keep, batch_first=True, padding_value=False)
334
+ return None if mask.all() else mask
335
+
336
+ def _run_blocks(self, blocks, hidden_states, attn_mask, freqs_cis, adaln_input=None):
337
+ for block in blocks:
338
+ if torch.is_grad_enabled() and self.gradient_checkpointing:
339
+ hidden_states = self._gradient_checkpointing_func(
340
+ block, hidden_states, attn_mask, freqs_cis, adaln_input
341
+ )
342
+ else:
343
+ hidden_states = block(hidden_states, attn_mask, freqs_cis, adaln_input)
344
+ return hidden_states
345
+
346
+ def forward(
347
+ self,
348
+ hidden_states: list[torch.Tensor],
349
+ timestep: torch.Tensor,
350
+ encoder_hidden_states: list[torch.Tensor],
351
+ encoder_hidden_states_2: list[torch.Tensor],
352
+ ref_hidden_states: list[torch.Tensor] | None = None,
353
+ return_dict: bool = True,
354
+ ) -> Transformer2DModelOutput | tuple[list[torch.Tensor]]:
355
+ r"""
356
+ The [`MingImageTransformer2DModel`] forward method.
357
+
358
+ Args:
359
+ hidden_states (`list[torch.Tensor]`):
360
+ Noisy latents, one `(in_channels, num_frames, height, width)` tensor per batch item.
361
+ timestep (`torch.Tensor`):
362
+ Normalized denoising time in `[0, 1]` of shape `(batch_size,)`, where 0 is pure noise.
363
+ encoder_hidden_states (`list[torch.Tensor]`):
364
+ Query-token conditioning, one `(seq_len, cap_feat_dim)` tensor per batch item. Passed through
365
+ `cap_embedder`.
366
+ encoder_hidden_states_2 (`list[torch.Tensor]`):
367
+ Text-token conditioning already in model width, one `(seq_len, dim)` tensor per batch item.
368
+ ref_hidden_states (`list[torch.Tensor]`, *optional*):
369
+ Reference latents, one `(in_channels, num_ref_frames, height, width)` tensor per batch item, appended
370
+ after the noisy frames along the frame axis.
371
+ return_dict (`bool`, defaults to `True`):
372
+ Whether to return a [`~models.modeling_outputs.Transformer2DModelOutput`] instead of a plain tuple.
373
+
374
+ Returns:
375
+ [`~models.modeling_outputs.Transformer2DModelOutput`] or `tuple`:
376
+ The predicted velocity as a list of `(in_channels, num_frames, height, width)` tensors, where
377
+ `num_frames` is 1 unless `multi_frame_output` is set. Returned as a one-element tuple when
378
+ `return_dict` is `False`.
379
+ """
380
+ p = self.config.patch_size
381
+ pf = self.config.f_patch_size
382
+ patch_key = f"{p}-{pf}"
383
+ learned_padding = self.config.alignment_padding_mode == "learned"
384
+ device = hidden_states[0].device
385
+
386
+ if ref_hidden_states is None:
387
+ ref_hidden_states = [None] * len(hidden_states)
388
+
389
+ # Patchify the latents and lay out caption and image positions. Each sequence is padded to a multiple of
390
+ # SEQ_MULTI_OF; caption positions start at 1 on the frame axis, image positions right after the captions.
391
+ x, x_sizes, x_pos_ids, x_pad_masks = [], [], [], []
392
+ cap_feats_2, cap_pos_ids, cap_pad_masks = [], [], []
393
+ for image, ref_image, cap_feat, cap_feat_2 in zip(
394
+ hidden_states, ref_hidden_states, encoder_hidden_states, encoder_hidden_states_2
395
+ ):
396
+ cap_len = len(cap_feat) + len(cap_feat_2)
397
+ cap_padding = (-cap_len) % SEQ_MULTI_OF
398
+ cap_pos_ids.append(create_coordinate_grid((cap_len + cap_padding, 1, 1), (1, 0, 0), device))
399
+ cap_pad_masks.append(
400
+ torch.cat(
401
+ [
402
+ torch.zeros(cap_len, dtype=torch.bool, device=device),
403
+ torch.ones(cap_padding, dtype=torch.bool, device=device),
404
+ ]
405
+ )
406
+ )
407
+ cap_feats_2.append(torch.cat([cap_feat_2, cap_feat_2.new_zeros(cap_padding, cap_feat_2.shape[-1])]))
408
+
409
+ if ref_image is not None:
410
+ image = torch.cat([image, ref_image], dim=1)
411
+ channels, frames, height, width = image.shape
412
+ x_sizes.append((frames, height, width))
413
+ f_tokens, h_tokens, w_tokens = frames // pf, height // p, width // p
414
+ image = image.view(channels, f_tokens, pf, h_tokens, p, w_tokens, p)
415
+ # "c f pf h ph w pw -> (f h w) (pf ph pw c)"
416
+ image = image.permute(1, 3, 5, 2, 4, 6, 0).reshape(f_tokens * h_tokens * w_tokens, -1)
417
+
418
+ image_len = len(image)
419
+ image_padding = (-image_len) % SEQ_MULTI_OF
420
+ x_pos_ids.append(
421
+ torch.cat(
422
+ [
423
+ create_coordinate_grid(
424
+ (f_tokens, h_tokens, w_tokens), (cap_len + cap_padding + 1, 0, 0), device
425
+ ),
426
+ torch.zeros((image_padding, 3), dtype=torch.int32, device=device),
427
+ ]
428
+ )
429
+ )
430
+ x_pad_masks.append(
431
+ torch.cat(
432
+ [
433
+ torch.zeros(image_len, dtype=torch.bool, device=device),
434
+ torch.ones(image_padding, dtype=torch.bool, device=device),
435
+ ]
436
+ )
437
+ )
438
+ x.append(torch.cat([image, image[-1:].repeat(image_padding, 1)]))
439
+
440
+ # Masks: padded batch positions are always excluded; alignment padding only in zero_masked mode.
441
+ x_keep = [~m if not learned_padding else torch.ones_like(m) for m in x_pad_masks]
442
+ cap_keep = [~m if not learned_padding else torch.ones_like(m) for m in cap_pad_masks]
443
+
444
+ adaln_input = self.t_embedder(timestep * self.config.t_scale)
445
+
446
+ # Image embed + noise refiner
447
+ x_lens = [len(item) for item in x]
448
+ x = self.all_x_embedder[patch_key](torch.cat(x))
449
+ adaln_input = adaln_input.type_as(x)
450
+ x[torch.cat(x_pad_masks)] = self.x_pad_token if learned_padding else 0.0
451
+ x_pos_ids_flat = torch.cat(x_pos_ids)
452
+ x_freqs = list(self.rope_embedder(x_pos_ids_flat).split(x_lens))
453
+ x = pad_sequence(list(x.split(x_lens)), batch_first=True, padding_value=0.0)
454
+ x_freqs_padded = pad_sequence(x_freqs, batch_first=True, padding_value=0.0)[:, : x.shape[1]]
455
+ x = self._run_blocks(self.noise_refiner, x, self._attention_mask(x_keep), x_freqs_padded, adaln_input)
456
+
457
+ # Caption embed + context refiner. Only the query-token stream goes through cap_embedder.
458
+ cap_lens_1 = [len(item) for item in encoder_hidden_states]
459
+ cap = list(self.cap_embedder(torch.cat(encoder_hidden_states)).split(cap_lens_1))
460
+ cap = [torch.cat([c1, c2]) for c1, c2 in zip(cap, cap_feats_2)]
461
+ cap_lens = [len(item) for item in cap]
462
+ cap = torch.cat(cap)
463
+ cap[torch.cat(cap_pad_masks)] = self.cap_pad_token if learned_padding else 0.0
464
+ cap_freqs = list(self.rope_embedder(torch.cat(cap_pos_ids)).split(cap_lens))
465
+ cap = pad_sequence(list(cap.split(cap_lens)), batch_first=True, padding_value=0.0)
466
+ cap_freqs_padded = pad_sequence(cap_freqs, batch_first=True, padding_value=0.0)[:, : cap.shape[1]]
467
+ cap = self._run_blocks(self.context_refiner, cap, self._attention_mask(cap_keep), cap_freqs_padded)
468
+
469
+ # Unified [image, caption] sequence through the main blocks
470
+ unified = pad_sequence(
471
+ [torch.cat([x[i][: x_lens[i]], cap[i][: cap_lens[i]]]) for i in range(len(x_lens))],
472
+ batch_first=True,
473
+ padding_value=0.0,
474
+ )
475
+ unified_freqs = pad_sequence(
476
+ [torch.cat([x_freqs[i], cap_freqs[i]]) for i in range(len(x_lens))],
477
+ batch_first=True,
478
+ padding_value=0.0,
479
+ )
480
+ unified_keep = [torch.cat([xk, ck]) for xk, ck in zip(x_keep, cap_keep)]
481
+ unified = self._run_blocks(
482
+ self.layers, unified, self._attention_mask(unified_keep), unified_freqs, adaln_input
483
+ )
484
+ unified = self.all_final_layer[patch_key](unified, adaln_input)
485
+
486
+ # Unpatchify, dropping alignment padding and caption tokens
487
+ output = []
488
+ for item, (frames, height, width) in zip(unified.unbind(dim=0), x_sizes):
489
+ f_tokens, h_tokens, w_tokens = frames // pf, height // p, width // p
490
+ item = (
491
+ item[: f_tokens * h_tokens * w_tokens]
492
+ .view(f_tokens, h_tokens, w_tokens, pf, p, p, self.out_channels)
493
+ # "f h w pf ph pw c -> c (f pf) (h ph) (w pw)"
494
+ .permute(6, 0, 3, 1, 4, 2, 5)
495
+ .reshape(self.out_channels, frames, height, width)
496
+ )
497
+ if not self.config.multi_frame_output:
498
+ item = item[:, :1]
499
+ output.append(item)
500
+
501
+ if not return_dict:
502
+ return (output,)
503
+ return Transformer2DModelOutput(sample=output)
vae/config.json ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_class_name": "AutoencoderKLQwenImage",
3
+ "_diffusers_version": "0.36.0.dev0",
4
+ "attn_scales": [],
5
+ "base_dim": 96,
6
+ "dim_mult": [
7
+ 1,
8
+ 2,
9
+ 4,
10
+ 4
11
+ ],
12
+ "dropout": 0.0,
13
+ "input_channels": 4,
14
+ "latents_mean": null,
15
+ "latents_std": null,
16
+ "scaling_factor": 8.0064,
17
+ "shift_factor": 0.0,
18
+ "num_res_blocks": 2,
19
+ "temperal_downsample": [
20
+ false,
21
+ true,
22
+ true
23
+ ],
24
+ "z_dim": 16
25
+ }
vae/diffusion_pytorch_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:06520463778e64dca1039c7447890065ee220bc408d15412182e5c3e06f304f1
3
+ size 253817336