jklyuiozcx commited on
Commit
74194bb
·
verified ·
1 Parent(s): 9f81cc8

Upload folder using huggingface_hub

Browse files
.gitattributes CHANGED
@@ -42,3 +42,4 @@ examples/ink-journey/ink-journey-demo.mp4 filter=lfs diff=lfs merge=lfs -text
42
  examples/ink-journey/ink-journey-preview.gif filter=lfs diff=lfs merge=lfs -text
43
  examples/moon-simulation/moon-simulation.gif filter=lfs diff=lfs merge=lfs -text
44
  examples/teacher-day/teacher-day-horizontal.jpg filter=lfs diff=lfs merge=lfs -text
 
 
42
  examples/ink-journey/ink-journey-preview.gif filter=lfs diff=lfs merge=lfs -text
43
  examples/moon-simulation/moon-simulation.gif filter=lfs diff=lfs merge=lfs -text
44
  examples/teacher-day/teacher-day-horizontal.jpg filter=lfs diff=lfs merge=lfs -text
45
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -1,201 +1,67 @@
1
  ---
2
  license: apache-2.0
3
- language:
4
- - en
5
- - zh
6
  library_name: transformers
7
  pipeline_tag: image-text-to-text
8
  tags:
9
- - multimodal
10
- - text-generation
11
- - image-text-to-text
12
- - looped-transformer
13
- - mhc
14
  - custom_code
 
15
  ---
16
 
17
- # SenseNova-6.8-Loop-9B-Preview
18
-
19
- <p align="center">
20
- <strong>English</strong> | <a href="https://huggingface.co/sensenova/SenseNova-6.8-Loop-9B-Preview/blob/main/README_CN.md">简体中文</a>
21
- </p>
22
-
23
- ![Looped Transformer architecture with a shared looped block and mHC residual streams.](assets/loop-mhc-overview.jpg)
24
-
25
- ## 📣 News
26
-
27
- - `[2026.10.30]` We release **SenseNova-6.8-Loop-9B-Preview**, a looped multimodal model for long-context and agentic workflows.
28
-
29
- ## Overview
30
-
31
- **SenseNova-6.8-Loop-9B-Preview** is a multimodal language model developed for long-context and agentic workflows. It accepts text and images and generates text for applications that combine document understanding, coding, and interaction with external tools.
32
-
33
- The model combines a **looped decoder with two-stream manifold-constrained hyper-connections (mHC)**. A contiguous block of decoder layers is executed twice with shared weights, while learned mappings manage information flow between two residual streams in the looped section. Together, these components provide a deeper execution path without duplicating the repeated block's parameters.
34
-
35
- Documented training recipes combine long sequences, multimodal inputs, and supervised task trajectories, with a maximum training sequence length of 262,144 tokens. Model behavior and deployment limits depend on the released checkpoint and inference implementation.
36
-
37
- ### Model highlights
38
-
39
- - **Shared looped block:** 24 decoder layers are applied twice in sequence, producing 56 decoder-layer executions from 32 distinct layers.
40
- - **Two-stream mHC:** learned mappings route and combine information across two residual streams in the looped section.
41
- - **Hybrid attention:** the decoder interleaves linear and full attention in a repeating four-layer pattern.
42
- - **Long-context multimodal training:** documented recipes include text, images, code, and tool trajectories at sequence lengths up to 256K tokens.
43
-
44
- ## Showcase (Created by Raccoon Work)
45
-
46
- ### Motion Design
47
-
48
- [![Motion Design](examples/ink-journey/ink-journey-preview.gif)](examples/ink-journey/ink-journey-demo.mp4)
49
-
50
- ### PPT Design
51
-
52
- ![Teacher’s Day presentation: slides 1, 3, and 6.](examples/teacher-day/teacher-day-horizontal.jpg)
53
-
54
- ![Iceland presentation: slides 1, 8, and 14.](examples/iceland/iceland-horizontal-v2.jpg)
55
-
56
- ![Chocolate presentation: slides 1, 4, and 5.](examples/chocolate/chocolate-horizontal.jpg)
57
-
58
- ### Data Analysis
59
-
60
- ![Lunar mission simulation with orbital trajectories and telemetry.](examples/moon-simulation/moon-simulation.gif)
61
-
62
- ![Energy analysis: temperature, electricity consumption, and peak-hour purchasing.](examples/energy-analysis/energy-analysis.png)
63
-
64
- ## 📊 Key Benchmarks
65
-
66
- ![SenseNova 6.8 Loop-9B-Preview benchmark results compared with other models on BIRD-Dev, WorkBuddy Bench, AIDABench, TeamBench, Terminal Bench 2.1, T³-Banking, AutomationBench, and RaccoonPPTBench.](assets/loop9b-benchmark-results.png)
67
-
68
- ## Model Specifications
69
-
70
- | Property | SenseNova-6.8-Loop-9B-Preview |
71
- | --- | --- |
72
- | Architecture | Autoregressive multimodal decoder; looped block with two-stream mHC |
73
- | Backbone scale | ~9B total parameters; ~14.6B inference parameters |
74
- | Input / output | Text and images / text |
75
- | Distinct decoder layers | 32 |
76
- | Looped layers / passes | [4, 28) / 2, with shared weights |
77
- | Decoder layer executions | 56 per forward pass |
78
- | mHC streams | 2 |
79
- | Training context | Up to 262,144 tokens |
80
- | Existing export precision | BF16 backbone; FP32 mHC routing parameters |
81
-
82
- Training sequence length is a configuration, not a guarantee of inference support or performance at every length.
83
-
84
- ## Model Architecture
85
 
86
- SenseNova-6.8-Loop-9B-Preview combines a 32-layer multimodal decoder with a repeated middle block. The figure illustrates the order of computation; the detailed operations inside each layer and the vision encoder are omitted.
 
 
87
 
88
- ![Input passes through early layers, then layers 4 through 27 twice with shared weights and two mHC streams, followed by final layers and text output.](assets/loop9b-architecture.svg)
89
 
90
- *Figure 1. Decoder execution order. Layer indices are zero-based; the two passes share block weights.*
 
91
 
92
- ### Looped decoder
93
-
94
- The decoder first processes the input through layers 0–3. The same block of layers 4–27 is then applied twice in sequence, after which layers 28��31 and the output head produce the model's output. If *B* represents the repeated block and *h* its input, the looped computation can be written as **B(B(h))**.
95
-
96
- The second pass takes the first pass's output as its input. Because both passes use the same layer weights, the model has **32 distinct decoder layers but 56 decoder-layer executions** per forward pass. Parameter sharing avoids duplicating the repeated block, while the additional pass still adds computation.
97
-
98
- ### Hybrid attention
99
-
100
- The decoder interleaves linear-attention and full-attention layers. Each four-layer group contains three linear-attention layers followed by one full-attention layer. The loop boundaries align with complete four-layer groups, so the attention pattern is preserved across both passes.
101
-
102
- ### Two-stream mHC
103
-
104
- Inside the looped section, mHC carries two residual streams around the decoder-layer computations. Learned mappings govern how a layer reads from the streams, mixes stream information, and incorporates the layer output into the residual state. Sinkhorn normalization constrains the stream-mixing map; the documented configuration uses 20 iterations.
105
-
106
- The loop range, pass count, and mHC routing parameters form part of the checkpoint's execution configuration. A compatible inference implementation must load these parameters and reproduce the same execution order.
107
-
108
- | Configuration | Value |
109
- | --- | --- |
110
- | Distinct decoder layers / executions | 32 / 56 |
111
- | Repeated block / passes | [4, 28) / 2 |
112
- | Hidden dimension | 4,096 |
113
- | Feed-forward dimension | 12,288 |
114
- | Vocabulary size | 248,320 |
115
- | Full-attention heads / KV heads | 16 / 4 |
116
- | mHC streams / Sinkhorn iterations | 2 / 20 |
117
-
118
- ## Inference Acceleration
119
-
120
- SenseNova-6.8-Loop-9B-Preview supports **MTP** for accelerated inference. **Dspark** and **Dflash2** are under development, with training currently in progress.
121
-
122
- | Method | Status |
123
- | --- | --- |
124
- | MTP | Available |
125
- | Dspark | Training in progress |
126
- | Dflash2 | Training in progress |
127
-
128
- Practical speedups depend on the workload, hardware, and serving configuration.
129
-
130
- See [OpenSenseNova/SenseSpec](https://github.com/OpenSenseNova/SenseSpec) for our speculative decoding training code.
131
-
132
- ## Models and Resources
133
-
134
- | Resource | Status |
135
- | --- | --- |
136
- | Model weights | [Hugging Face model page](https://huggingface.co/sensenova/SenseNova-6.8-Loop-9B-Preview) |
137
- | Transformers inference implementation | Publication details to be announced |
138
- | Technical report | To be announced |
139
-
140
- ## Quick Start
141
-
142
- Use Transformers to load the model and its custom implementation from a prepared model directory. The example below runs text generation on a single GPU.
143
-
144
- ### Requirements
145
-
146
- The checked inference environment uses **Transformers 5.18.0** and **PyTorch 2.11.0 with CUDA 13.0**. The model directory must include the configuration, tokenizer, chat template, model code, and accessible weight shards.
147
-
148
- ### Text generation with Transformers
149
-
150
- Set `model_id` to the prepared model directory, then run:
151
 
152
  ```python
153
  import torch
154
  from transformers import AutoModelForCausalLM, AutoTokenizer
155
 
156
- model_id = "/path/to/model-directory"
157
- device = "cuda:0"
158
-
159
- tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
160
  model = AutoModelForCausalLM.from_pretrained(
161
- model_id,
162
  trust_remote_code=True,
163
  dtype=torch.bfloat16,
164
- device_map={"": device},
165
- ).eval()
 
166
 
167
- messages = [{"role": "user", "content": "What is 13 times 7? Reply with the number only."}]
168
  prompt = tokenizer.apply_chat_template(
169
- messages,
170
- tokenize=False,
171
- add_generation_prompt=True,
172
- enable_thinking=False,
173
  )
174
- inputs = tokenizer(prompt, return_tensors="pt").to(device)
175
-
176
  with torch.inference_mode():
177
- outputs = model.generate(
178
- **inputs,
179
- max_new_tokens=256,
180
- do_sample=False,
181
- use_cache=True,
182
- )
183
-
184
- response = tokenizer.decode(
185
- outputs[0, inputs.input_ids.shape[1]:],
186
- skip_special_tokens=True,
187
- )
188
- print(response)
189
  ```
190
 
191
- The example uses greedy decoding and generates up to 256 new tokens. Set `enable_thinking=True` to enable thinking mode, or change `device` to select another GPU.
 
 
192
 
193
- `trust_remote_code=True` loads the custom model implementation from the model directory. Review model code before loading it from an untrusted source. The example follows the previously validated single-GPU inference script; image input and tool calling are not covered by this example.
194
 
195
- ### Training
 
 
196
 
197
- The model is trained using [OpenSenseNova/torchtitan](https://github.com/OpenSenseNova/torchtitan); see the `example/train` directory in the repository for training examples.
198
 
199
- ## License and citation
 
 
 
200
 
201
- **License:** To be determined for the released weights and code.<br>**Citation:** To be added with the release.
 
 
1
  ---
2
  license: apache-2.0
 
 
 
3
  library_name: transformers
4
  pipeline_tag: image-text-to-text
5
  tags:
 
 
 
 
 
6
  - custom_code
7
+ - multimodal
8
  ---
9
 
10
+ # SenseNova 6.8 Loop 9B Preview
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
11
 
12
+ SenseNova 6.8 Loop 9B Preview is a multimodal model with shared-weight Loop
13
+ execution. This repository contains the model configuration, tokenizer,
14
+ processor, Transformers modeling code, model weights, and MTP draft weights.
15
 
16
+ ## Text inference with Transformers
17
 
18
+ The custom implementation is included in this repository. The example below
19
+ uses the tested Transformers version and runs on a CUDA device with BF16 support.
20
 
21
+ ```bash
22
+ pip install torch transformers==5.18.0 accelerate safetensors
23
+ ```
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
24
 
25
  ```python
26
  import torch
27
  from transformers import AutoModelForCausalLM, AutoTokenizer
28
 
29
+ model_path = "sensenova/SenseNova-6.8-Loop-9B-Preview" # Replace with your Hub repository ID.
30
+ tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
 
 
31
  model = AutoModelForCausalLM.from_pretrained(
32
+ model_path,
33
  trust_remote_code=True,
34
  dtype=torch.bfloat16,
35
+ device_map="auto",
36
+ )
37
+ model.eval()
38
 
39
+ messages = [{"role": "user", "content": "What is 13 multiplied by 7?"}]
40
  prompt = tokenizer.apply_chat_template(
41
+ messages, tokenize=False, add_generation_prompt=True, enable_thinking=False
 
 
 
42
  )
43
+ inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
 
44
  with torch.inference_mode():
45
+ output_ids = model.generate(**inputs, max_new_tokens=128, do_sample=False)
46
+ print(tokenizer.decode(output_ids[0, inputs.input_ids.shape[1]:], skip_special_tokens=True))
 
 
 
 
 
 
 
 
 
 
47
  ```
48
 
49
+ The model defaults to PyTorch SDPA attention. An appropriate PyTorch, CUDA and
50
+ GPU setup is required for GPU inference. Automatic device placement requires
51
+ Accelerate. Other library versions have not all been validated.
52
 
53
+ ## Multimodal inputs
54
 
55
+ The repository contains image and video processor configuration. Multimodal
56
+ examples and supported inference runtimes will be documented after end-to-end
57
+ validation.
58
 
59
+ ## MTP
60
 
61
+ The checkpoint includes one MTP draft layer. Standard Transformers generation
62
+ loads the main model and does not use MTP for speculative decoding. MTP
63
+ acceleration requires a compatible inference engine and a supported draft
64
+ loading configuration.
65
 
66
+ The main model weights are stored in FP32 and the MTP weights in BF16. The
67
+ example requests BF16 inference; the implementation keeps mHC routing in FP32.
chat_template.jinja ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set image_count = namespace(value=0) %}
2
+ {%- set video_count = namespace(value=0) %}
3
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
4
+ {%- if content is string %}
5
+ {{- content }}
6
+ {%- elif content is iterable and content is not mapping %}
7
+ {%- for item in content %}
8
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
9
+ {%- if is_system_content %}
10
+ {{- raise_exception('System message cannot contain images.') }}
11
+ {%- endif %}
12
+ {%- if do_vision_count %}
13
+ {%- set image_count.value = image_count.value + 1 %}
14
+ {%- endif %}
15
+ {%- if add_vision_id %}
16
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
17
+ {%- endif %}
18
+ {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
19
+ {%- elif 'video' in item or item.type == 'video' %}
20
+ {%- if is_system_content %}
21
+ {{- raise_exception('System message cannot contain videos.') }}
22
+ {%- endif %}
23
+ {%- if do_vision_count %}
24
+ {%- set video_count.value = video_count.value + 1 %}
25
+ {%- endif %}
26
+ {%- if add_vision_id %}
27
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
28
+ {%- endif %}
29
+ {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
30
+ {%- elif 'text' in item %}
31
+ {{- item.text }}
32
+ {%- else %}
33
+ {{- raise_exception('Unexpected item type in content.') }}
34
+ {%- endif %}
35
+ {%- endfor %}
36
+ {%- elif content is none or content is undefined %}
37
+ {{- '' }}
38
+ {%- else %}
39
+ {{- raise_exception('Unexpected content type.') }}
40
+ {%- endif %}
41
+ {%- endmacro %}
42
+ {%- if not messages %}
43
+ {{- raise_exception('No messages provided.') }}
44
+ {%- endif %}
45
+ {%- if tools and tools is iterable and tools is not mapping %}
46
+ {{- '<|im_start|>system\n' }}
47
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
48
+ {%- for tool in tools %}
49
+ {{- "\n" }}
50
+ {{- tool | tojson }}
51
+ {%- endfor %}
52
+ {{- "\n</tools>" }}
53
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
54
+ {%- if messages[0].role == 'system' %}
55
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
56
+ {%- if content %}
57
+ {{- '\n\n' + content }}
58
+ {%- endif %}
59
+ {%- endif %}
60
+ {{- '<|im_end|>\n' }}
61
+ {%- else %}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
65
+ {%- endif %}
66
+ {%- endif %}
67
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
68
+ {%- for message in messages[::-1] %}
69
+ {%- set index = (messages|length - 1) - loop.index0 %}
70
+ {%- if ns.multi_step_tool and message.role == "user" %}
71
+ {%- set content = render_content(message.content, false)|trim %}
72
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
73
+ {%- set ns.multi_step_tool = false %}
74
+ {%- set ns.last_query_index = index %}
75
+ {%- endif %}
76
+ {%- endif %}
77
+ {%- endfor %}
78
+ {%- if ns.multi_step_tool %}
79
+ {{- raise_exception('No user query found in messages.') }}
80
+ {%- endif %}
81
+ {%- for message in messages %}
82
+ {%- set content = render_content(message.content, true)|trim %}
83
+ {%- if message.role == "system" %}
84
+ {%- if not loop.first %}
85
+ {{- raise_exception('System message must be at the beginning.') }}
86
+ {%- endif %}
87
+ {%- elif message.role == "user" %}
88
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
89
+ {%- elif message.role == "assistant" %}
90
+ {%- set reasoning_content = '' %}
91
+ {%- if message.reasoning_content is string %}
92
+ {%- set reasoning_content = message.reasoning_content %}
93
+ {%- else %}
94
+ {%- if '</think>' in content %}
95
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
96
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
97
+ {%- endif %}
98
+ {%- endif %}
99
+ {%- set reasoning_content = reasoning_content|trim %}
100
+ {%- if loop.index0 > ns.last_query_index %}
101
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
102
+ {%- else %}
103
+ {{- '<|im_start|>' + message.role + '\n' + content }}
104
+ {%- endif %}
105
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
106
+ {%- for tool_call in message.tool_calls %}
107
+ {%- if tool_call.function is defined %}
108
+ {%- set tool_call = tool_call.function %}
109
+ {%- endif %}
110
+ {%- if loop.first %}
111
+ {%- if content|trim %}
112
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
113
+ {%- else %}
114
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
115
+ {%- endif %}
116
+ {%- else %}
117
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
118
+ {%- endif %}
119
+ {%- if tool_call.arguments is defined %}
120
+ {%- for args_name, args_value in tool_call.arguments|items %}
121
+ {{- '<parameter=' + args_name + '>\n' }}
122
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
123
+ {{- args_value }}
124
+ {{- '\n</parameter>\n' }}
125
+ {%- endfor %}
126
+ {%- endif %}
127
+ {{- '</function>\n</tool_call>' }}
128
+ {%- endfor %}
129
+ {%- endif %}
130
+ {{- '<|im_end|>\n' }}
131
+ {%- elif message.role == "tool" %}
132
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
133
+ {{- '<|im_start|>user' }}
134
+ {%- endif %}
135
+ {{- '\n<tool_response>\n' }}
136
+ {{- content }}
137
+ {{- '\n</tool_response>' }}
138
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
139
+ {{- '<|im_end|>\n' }}
140
+ {%- elif loop.last %}
141
+ {{- '<|im_end|>\n' }}
142
+ {%- endif %}
143
+ {%- else %}
144
+ {{- raise_exception('Unexpected message role.') }}
145
+ {%- endif %}
146
+ {%- endfor %}
147
+ {%- if add_generation_prompt %}
148
+ {{- '<|im_start|>assistant\n' }}
149
+ {%- if enable_thinking is defined and enable_thinking is false %}
150
+ {{- '<think>\n\n</think>\n\n' }}
151
+ {%- else %}
152
+ {{- '<think>\n' }}
153
+ {%- endif %}
154
+ {%- endif %}
config.json ADDED
@@ -0,0 +1,145 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "SenseNovaLoopForConditionalGeneration"
4
+ ],
5
+ "attn_implementation": "sdpa",
6
+ "bos_token_id": null,
7
+ "dtype": "bfloat16",
8
+ "eos_token_id": 248046,
9
+ "hidden_size": 4096,
10
+ "image_token_id": 248056,
11
+ "loop_settings": {
12
+ "counts": [
13
+ 2
14
+ ],
15
+ "end": 28,
16
+ "eval_count": 2,
17
+ "gate_init": 0.02,
18
+ "probabilities": [
19
+ 1.0
20
+ ],
21
+ "seed": 42,
22
+ "start": 4,
23
+ "use_gate": false
24
+ },
25
+ "model_type": "sensenova_loop",
26
+ "pad_token_id": 248044,
27
+ "patch_conv3d": true,
28
+ "text_config": {
29
+ "attention_bias": false,
30
+ "attention_dropout": 0.0,
31
+ "attn_output_gate": true,
32
+ "bos_token_id": null,
33
+ "dtype": "bfloat16",
34
+ "eos_token_id": 248044,
35
+ "full_attention_interval": 4,
36
+ "head_dim": 256,
37
+ "hidden_act": "silu",
38
+ "hidden_size": 4096,
39
+ "initializer_range": 0.02,
40
+ "intermediate_size": 12288,
41
+ "layer_types": [
42
+ "linear_attention",
43
+ "linear_attention",
44
+ "linear_attention",
45
+ "full_attention",
46
+ "linear_attention",
47
+ "linear_attention",
48
+ "linear_attention",
49
+ "full_attention",
50
+ "linear_attention",
51
+ "linear_attention",
52
+ "linear_attention",
53
+ "full_attention",
54
+ "linear_attention",
55
+ "linear_attention",
56
+ "linear_attention",
57
+ "full_attention",
58
+ "linear_attention",
59
+ "linear_attention",
60
+ "linear_attention",
61
+ "full_attention",
62
+ "linear_attention",
63
+ "linear_attention",
64
+ "linear_attention",
65
+ "full_attention",
66
+ "linear_attention",
67
+ "linear_attention",
68
+ "linear_attention",
69
+ "full_attention",
70
+ "linear_attention",
71
+ "linear_attention",
72
+ "linear_attention",
73
+ "full_attention"
74
+ ],
75
+ "linear_conv_kernel_dim": 4,
76
+ "linear_key_head_dim": 128,
77
+ "linear_num_key_heads": 16,
78
+ "linear_num_value_heads": 32,
79
+ "linear_value_head_dim": 128,
80
+ "mamba_ssm_dtype": "float32",
81
+ "max_position_embeddings": 262144,
82
+ "mlp_only_layers": [],
83
+ "model_type": "qwen3_5_text",
84
+ "mtp_num_hidden_layers": 1,
85
+ "mtp_use_dedicated_embeddings": false,
86
+ "num_attention_heads": 16,
87
+ "num_hidden_layers": 32,
88
+ "num_key_value_heads": 4,
89
+ "pad_token_id": 248044,
90
+ "partial_rotary_factor": 0.25,
91
+ "rms_norm_eps": 1e-06,
92
+ "rope_parameters": {
93
+ "mrope_interleaved": true,
94
+ "mrope_section": [
95
+ 11,
96
+ 11,
97
+ 10
98
+ ],
99
+ "partial_rotary_factor": 0.25,
100
+ "rope_theta": 10000000,
101
+ "rope_type": "default"
102
+ },
103
+ "tie_word_embeddings": false,
104
+ "use_cache": true,
105
+ "vocab_size": 248320
106
+ },
107
+ "tie_word_embeddings": false,
108
+ "transformers_version": "5.8.1",
109
+ "use_cache": true,
110
+ "video_token_id": 248057,
111
+ "vision_config": {
112
+ "deepstack_visual_indexes": [],
113
+ "depth": 27,
114
+ "dtype": "bfloat16",
115
+ "hidden_act": "gelu_pytorch_tanh",
116
+ "hidden_size": 1152,
117
+ "in_channels": 3,
118
+ "initializer_range": 0.02,
119
+ "intermediate_size": 4304,
120
+ "model_type": "qwen3_5_vision",
121
+ "num_heads": 16,
122
+ "num_position_embeddings": 2304,
123
+ "out_hidden_size": 4096,
124
+ "patch_size": 16,
125
+ "spatial_merge_size": 2,
126
+ "temporal_patch_size": 2
127
+ },
128
+ "vision_end_token_id": 248054,
129
+ "vision_start_token_id": 248053,
130
+ "mhc": {
131
+ "end": 28,
132
+ "fixed_identity": false,
133
+ "iterations": 20,
134
+ "pre_scale": 1.0,
135
+ "start": 4,
136
+ "streams": 2
137
+ },
138
+ "model_name": "SenseNova 6.8 Loop 9B preview",
139
+ "auto_map": {
140
+ "AutoConfig": "modeling_sensenova_loop.SenseNovaLoopConfig",
141
+ "AutoModel": "modeling_sensenova_loop.SenseNovaLoopModel",
142
+ "AutoModelForCausalLM": "modeling_sensenova_loop.SenseNovaLoopForConditionalGeneration",
143
+ "AutoModelForImageTextToText": "modeling_sensenova_loop.SenseNovaLoopForConditionalGeneration"
144
+ }
145
+ }
generation_config.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "eos_token_id": [
4
+ 248044,
5
+ 248046
6
+ ],
7
+ "transformers_version": "5.8.1",
8
+ "use_cache": true,
9
+ "temperature": 1.0,
10
+ "top_k": 20,
11
+ "top_p": 0.95
12
+ }
model-00001-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2dce742fbfa66210f82c5233ca29c4d60a9316ca7e4013309dfba55c3b935f9d
3
+ size 4068475008
model-00002-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:283728eccc8398a3b6452ffda881704564000f17089a95ec056f1484b8750b05
3
+ size 4943302320
model-00003-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c341e1836a728731cf59867b933fadf21535a824a08343b6fae71de9d525300f
3
+ size 4805805352
model-00004-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:72e7c9a54107c5c0871408cb59c9237d6688ec2f9c834d4a75fdd4e4070add5e
3
+ size 4973612456
model-00005-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7f8e974e4f4ae09c62e028e50522105c5fd1c945fb7a68ddeeba283b58d0e9ed
3
+ size 4941235864
model-00006-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:97ddbb10982d6684d89b94480d6f46a379036801c0118f436b26e13112c950e2
3
+ size 4954587112
model-00007-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:11a3c4502352be133dde8892879f22a0aa33c6d7b94cd11e2c1b0886481dd6a7
3
+ size 4956292552
model-00008-of-00008.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fbbad135d34f72cdb04fb568c8c6dc493be2ee347c77f3f190bf0c1d5186341d
3
+ size 4008633000
model.safetensors.index.json ADDED
@@ -0,0 +1,878 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 38138420672
4
+ },
5
+ "weight_map": {
6
+ "lm_head.weight": "model-00001-of-00008.safetensors",
7
+ "model.language_model.embed_tokens.weight": "model-00002-of-00008.safetensors",
8
+ "model.language_model.layers.0.input_layernorm.weight": "model-00002-of-00008.safetensors",
9
+ "model.language_model.layers.0.linear_attn.A_log": "model-00002-of-00008.safetensors",
10
+ "model.language_model.layers.0.linear_attn.conv1d.weight": "model-00002-of-00008.safetensors",
11
+ "model.language_model.layers.0.linear_attn.dt_bias": "model-00002-of-00008.safetensors",
12
+ "model.language_model.layers.0.linear_attn.in_proj_a.weight": "model-00002-of-00008.safetensors",
13
+ "model.language_model.layers.0.linear_attn.in_proj_b.weight": "model-00002-of-00008.safetensors",
14
+ "model.language_model.layers.0.linear_attn.in_proj_qkv.weight": "model-00002-of-00008.safetensors",
15
+ "model.language_model.layers.0.linear_attn.in_proj_z.weight": "model-00002-of-00008.safetensors",
16
+ "model.language_model.layers.0.linear_attn.norm.weight": "model-00002-of-00008.safetensors",
17
+ "model.language_model.layers.0.linear_attn.out_proj.weight": "model-00002-of-00008.safetensors",
18
+ "model.language_model.layers.0.mlp.down_proj.weight": "model-00002-of-00008.safetensors",
19
+ "model.language_model.layers.0.mlp.gate_proj.weight": "model-00002-of-00008.safetensors",
20
+ "model.language_model.layers.0.mlp.up_proj.weight": "model-00002-of-00008.safetensors",
21
+ "model.language_model.layers.0.post_attention_layernorm.weight": "model-00002-of-00008.safetensors",
22
+ "model.language_model.layers.1.input_layernorm.weight": "model-00002-of-00008.safetensors",
23
+ "model.language_model.layers.1.linear_attn.A_log": "model-00002-of-00008.safetensors",
24
+ "model.language_model.layers.1.linear_attn.conv1d.weight": "model-00002-of-00008.safetensors",
25
+ "model.language_model.layers.1.linear_attn.dt_bias": "model-00002-of-00008.safetensors",
26
+ "model.language_model.layers.1.linear_attn.in_proj_a.weight": "model-00002-of-00008.safetensors",
27
+ "model.language_model.layers.1.linear_attn.in_proj_b.weight": "model-00002-of-00008.safetensors",
28
+ "model.language_model.layers.1.linear_attn.in_proj_qkv.weight": "model-00003-of-00008.safetensors",
29
+ "model.language_model.layers.1.linear_attn.in_proj_z.weight": "model-00003-of-00008.safetensors",
30
+ "model.language_model.layers.1.linear_attn.norm.weight": "model-00003-of-00008.safetensors",
31
+ "model.language_model.layers.1.linear_attn.out_proj.weight": "model-00003-of-00008.safetensors",
32
+ "model.language_model.layers.1.mlp.down_proj.weight": "model-00003-of-00008.safetensors",
33
+ "model.language_model.layers.1.mlp.gate_proj.weight": "model-00003-of-00008.safetensors",
34
+ "model.language_model.layers.1.mlp.up_proj.weight": "model-00003-of-00008.safetensors",
35
+ "model.language_model.layers.1.post_attention_layernorm.weight": "model-00003-of-00008.safetensors",
36
+ "model.language_model.layers.10.input_layernorm.weight": "model-00003-of-00008.safetensors",
37
+ "model.language_model.layers.10.linear_attn.A_log": "model-00003-of-00008.safetensors",
38
+ "model.language_model.layers.10.linear_attn.conv1d.weight": "model-00003-of-00008.safetensors",
39
+ "model.language_model.layers.10.linear_attn.dt_bias": "model-00003-of-00008.safetensors",
40
+ "model.language_model.layers.10.linear_attn.in_proj_a.weight": "model-00003-of-00008.safetensors",
41
+ "model.language_model.layers.10.linear_attn.in_proj_b.weight": "model-00003-of-00008.safetensors",
42
+ "model.language_model.layers.10.linear_attn.in_proj_qkv.weight": "model-00003-of-00008.safetensors",
43
+ "model.language_model.layers.10.linear_attn.in_proj_z.weight": "model-00003-of-00008.safetensors",
44
+ "model.language_model.layers.10.linear_attn.norm.weight": "model-00003-of-00008.safetensors",
45
+ "model.language_model.layers.10.linear_attn.out_proj.weight": "model-00003-of-00008.safetensors",
46
+ "model.language_model.layers.10.mhc_attn.bias": "model-00003-of-00008.safetensors",
47
+ "model.language_model.layers.10.mhc_attn.weight": "model-00003-of-00008.safetensors",
48
+ "model.language_model.layers.10.mhc_ffn.bias": "model-00003-of-00008.safetensors",
49
+ "model.language_model.layers.10.mhc_ffn.weight": "model-00003-of-00008.safetensors",
50
+ "model.language_model.layers.10.mlp.down_proj.weight": "model-00003-of-00008.safetensors",
51
+ "model.language_model.layers.10.mlp.gate_proj.weight": "model-00003-of-00008.safetensors",
52
+ "model.language_model.layers.10.mlp.up_proj.weight": "model-00003-of-00008.safetensors",
53
+ "model.language_model.layers.10.post_attention_layernorm.weight": "model-00003-of-00008.safetensors",
54
+ "model.language_model.layers.11.input_layernorm.weight": "model-00003-of-00008.safetensors",
55
+ "model.language_model.layers.11.mhc_attn.bias": "model-00003-of-00008.safetensors",
56
+ "model.language_model.layers.11.mhc_attn.weight": "model-00003-of-00008.safetensors",
57
+ "model.language_model.layers.11.mhc_ffn.bias": "model-00003-of-00008.safetensors",
58
+ "model.language_model.layers.11.mhc_ffn.weight": "model-00003-of-00008.safetensors",
59
+ "model.language_model.layers.11.mlp.down_proj.weight": "model-00003-of-00008.safetensors",
60
+ "model.language_model.layers.11.mlp.gate_proj.weight": "model-00003-of-00008.safetensors",
61
+ "model.language_model.layers.11.mlp.up_proj.weight": "model-00003-of-00008.safetensors",
62
+ "model.language_model.layers.11.post_attention_layernorm.weight": "model-00003-of-00008.safetensors",
63
+ "model.language_model.layers.11.self_attn.k_norm.weight": "model-00003-of-00008.safetensors",
64
+ "model.language_model.layers.11.self_attn.k_proj.weight": "model-00003-of-00008.safetensors",
65
+ "model.language_model.layers.11.self_attn.o_proj.weight": "model-00003-of-00008.safetensors",
66
+ "model.language_model.layers.11.self_attn.q_norm.weight": "model-00003-of-00008.safetensors",
67
+ "model.language_model.layers.11.self_attn.q_proj.weight": "model-00003-of-00008.safetensors",
68
+ "model.language_model.layers.11.self_attn.v_proj.weight": "model-00003-of-00008.safetensors",
69
+ "model.language_model.layers.12.input_layernorm.weight": "model-00003-of-00008.safetensors",
70
+ "model.language_model.layers.12.linear_attn.A_log": "model-00003-of-00008.safetensors",
71
+ "model.language_model.layers.12.linear_attn.conv1d.weight": "model-00003-of-00008.safetensors",
72
+ "model.language_model.layers.12.linear_attn.dt_bias": "model-00003-of-00008.safetensors",
73
+ "model.language_model.layers.12.linear_attn.in_proj_a.weight": "model-00003-of-00008.safetensors",
74
+ "model.language_model.layers.12.linear_attn.in_proj_b.weight": "model-00003-of-00008.safetensors",
75
+ "model.language_model.layers.12.linear_attn.in_proj_qkv.weight": "model-00003-of-00008.safetensors",
76
+ "model.language_model.layers.12.linear_attn.in_proj_z.weight": "model-00003-of-00008.safetensors",
77
+ "model.language_model.layers.12.linear_attn.norm.weight": "model-00003-of-00008.safetensors",
78
+ "model.language_model.layers.12.linear_attn.out_proj.weight": "model-00003-of-00008.safetensors",
79
+ "model.language_model.layers.12.mhc_attn.bias": "model-00003-of-00008.safetensors",
80
+ "model.language_model.layers.12.mhc_attn.weight": "model-00003-of-00008.safetensors",
81
+ "model.language_model.layers.12.mhc_ffn.bias": "model-00003-of-00008.safetensors",
82
+ "model.language_model.layers.12.mhc_ffn.weight": "model-00003-of-00008.safetensors",
83
+ "model.language_model.layers.12.mlp.down_proj.weight": "model-00003-of-00008.safetensors",
84
+ "model.language_model.layers.12.mlp.gate_proj.weight": "model-00003-of-00008.safetensors",
85
+ "model.language_model.layers.12.mlp.up_proj.weight": "model-00003-of-00008.safetensors",
86
+ "model.language_model.layers.12.post_attention_layernorm.weight": "model-00003-of-00008.safetensors",
87
+ "model.language_model.layers.13.input_layernorm.weight": "model-00003-of-00008.safetensors",
88
+ "model.language_model.layers.13.linear_attn.A_log": "model-00003-of-00008.safetensors",
89
+ "model.language_model.layers.13.linear_attn.conv1d.weight": "model-00003-of-00008.safetensors",
90
+ "model.language_model.layers.13.linear_attn.dt_bias": "model-00003-of-00008.safetensors",
91
+ "model.language_model.layers.13.linear_attn.in_proj_a.weight": "model-00003-of-00008.safetensors",
92
+ "model.language_model.layers.13.linear_attn.in_proj_b.weight": "model-00003-of-00008.safetensors",
93
+ "model.language_model.layers.13.linear_attn.in_proj_qkv.weight": "model-00003-of-00008.safetensors",
94
+ "model.language_model.layers.13.linear_attn.in_proj_z.weight": "model-00003-of-00008.safetensors",
95
+ "model.language_model.layers.13.linear_attn.norm.weight": "model-00003-of-00008.safetensors",
96
+ "model.language_model.layers.13.linear_attn.out_proj.weight": "model-00003-of-00008.safetensors",
97
+ "model.language_model.layers.13.mhc_attn.bias": "model-00003-of-00008.safetensors",
98
+ "model.language_model.layers.13.mhc_attn.weight": "model-00003-of-00008.safetensors",
99
+ "model.language_model.layers.13.mhc_ffn.bias": "model-00003-of-00008.safetensors",
100
+ "model.language_model.layers.13.mhc_ffn.weight": "model-00003-of-00008.safetensors",
101
+ "model.language_model.layers.13.mlp.down_proj.weight": "model-00003-of-00008.safetensors",
102
+ "model.language_model.layers.13.mlp.gate_proj.weight": "model-00003-of-00008.safetensors",
103
+ "model.language_model.layers.13.mlp.up_proj.weight": "model-00003-of-00008.safetensors",
104
+ "model.language_model.layers.13.post_attention_layernorm.weight": "model-00003-of-00008.safetensors",
105
+ "model.language_model.layers.14.input_layernorm.weight": "model-00003-of-00008.safetensors",
106
+ "model.language_model.layers.14.linear_attn.A_log": "model-00003-of-00008.safetensors",
107
+ "model.language_model.layers.14.linear_attn.conv1d.weight": "model-00003-of-00008.safetensors",
108
+ "model.language_model.layers.14.linear_attn.dt_bias": "model-00003-of-00008.safetensors",
109
+ "model.language_model.layers.14.linear_attn.in_proj_a.weight": "model-00003-of-00008.safetensors",
110
+ "model.language_model.layers.14.linear_attn.in_proj_b.weight": "model-00003-of-00008.safetensors",
111
+ "model.language_model.layers.14.linear_attn.in_proj_qkv.weight": "model-00003-of-00008.safetensors",
112
+ "model.language_model.layers.14.linear_attn.in_proj_z.weight": "model-00003-of-00008.safetensors",
113
+ "model.language_model.layers.14.linear_attn.norm.weight": "model-00003-of-00008.safetensors",
114
+ "model.language_model.layers.14.linear_attn.out_proj.weight": "model-00003-of-00008.safetensors",
115
+ "model.language_model.layers.14.mhc_attn.bias": "model-00003-of-00008.safetensors",
116
+ "model.language_model.layers.14.mhc_attn.weight": "model-00003-of-00008.safetensors",
117
+ "model.language_model.layers.14.mhc_ffn.bias": "model-00003-of-00008.safetensors",
118
+ "model.language_model.layers.14.mhc_ffn.weight": "model-00003-of-00008.safetensors",
119
+ "model.language_model.layers.14.mlp.down_proj.weight": "model-00003-of-00008.safetensors",
120
+ "model.language_model.layers.14.mlp.gate_proj.weight": "model-00004-of-00008.safetensors",
121
+ "model.language_model.layers.14.mlp.up_proj.weight": "model-00004-of-00008.safetensors",
122
+ "model.language_model.layers.14.post_attention_layernorm.weight": "model-00004-of-00008.safetensors",
123
+ "model.language_model.layers.15.input_layernorm.weight": "model-00004-of-00008.safetensors",
124
+ "model.language_model.layers.15.mhc_attn.bias": "model-00004-of-00008.safetensors",
125
+ "model.language_model.layers.15.mhc_attn.weight": "model-00004-of-00008.safetensors",
126
+ "model.language_model.layers.15.mhc_ffn.bias": "model-00004-of-00008.safetensors",
127
+ "model.language_model.layers.15.mhc_ffn.weight": "model-00004-of-00008.safetensors",
128
+ "model.language_model.layers.15.mlp.down_proj.weight": "model-00004-of-00008.safetensors",
129
+ "model.language_model.layers.15.mlp.gate_proj.weight": "model-00004-of-00008.safetensors",
130
+ "model.language_model.layers.15.mlp.up_proj.weight": "model-00004-of-00008.safetensors",
131
+ "model.language_model.layers.15.post_attention_layernorm.weight": "model-00004-of-00008.safetensors",
132
+ "model.language_model.layers.15.self_attn.k_norm.weight": "model-00004-of-00008.safetensors",
133
+ "model.language_model.layers.15.self_attn.k_proj.weight": "model-00004-of-00008.safetensors",
134
+ "model.language_model.layers.15.self_attn.o_proj.weight": "model-00004-of-00008.safetensors",
135
+ "model.language_model.layers.15.self_attn.q_norm.weight": "model-00004-of-00008.safetensors",
136
+ "model.language_model.layers.15.self_attn.q_proj.weight": "model-00004-of-00008.safetensors",
137
+ "model.language_model.layers.15.self_attn.v_proj.weight": "model-00004-of-00008.safetensors",
138
+ "model.language_model.layers.16.input_layernorm.weight": "model-00004-of-00008.safetensors",
139
+ "model.language_model.layers.16.linear_attn.A_log": "model-00004-of-00008.safetensors",
140
+ "model.language_model.layers.16.linear_attn.conv1d.weight": "model-00004-of-00008.safetensors",
141
+ "model.language_model.layers.16.linear_attn.dt_bias": "model-00004-of-00008.safetensors",
142
+ "model.language_model.layers.16.linear_attn.in_proj_a.weight": "model-00004-of-00008.safetensors",
143
+ "model.language_model.layers.16.linear_attn.in_proj_b.weight": "model-00004-of-00008.safetensors",
144
+ "model.language_model.layers.16.linear_attn.in_proj_qkv.weight": "model-00004-of-00008.safetensors",
145
+ "model.language_model.layers.16.linear_attn.in_proj_z.weight": "model-00004-of-00008.safetensors",
146
+ "model.language_model.layers.16.linear_attn.norm.weight": "model-00004-of-00008.safetensors",
147
+ "model.language_model.layers.16.linear_attn.out_proj.weight": "model-00004-of-00008.safetensors",
148
+ "model.language_model.layers.16.mhc_attn.bias": "model-00004-of-00008.safetensors",
149
+ "model.language_model.layers.16.mhc_attn.weight": "model-00004-of-00008.safetensors",
150
+ "model.language_model.layers.16.mhc_ffn.bias": "model-00004-of-00008.safetensors",
151
+ "model.language_model.layers.16.mhc_ffn.weight": "model-00004-of-00008.safetensors",
152
+ "model.language_model.layers.16.mlp.down_proj.weight": "model-00004-of-00008.safetensors",
153
+ "model.language_model.layers.16.mlp.gate_proj.weight": "model-00004-of-00008.safetensors",
154
+ "model.language_model.layers.16.mlp.up_proj.weight": "model-00004-of-00008.safetensors",
155
+ "model.language_model.layers.16.post_attention_layernorm.weight": "model-00004-of-00008.safetensors",
156
+ "model.language_model.layers.17.input_layernorm.weight": "model-00004-of-00008.safetensors",
157
+ "model.language_model.layers.17.linear_attn.A_log": "model-00004-of-00008.safetensors",
158
+ "model.language_model.layers.17.linear_attn.conv1d.weight": "model-00004-of-00008.safetensors",
159
+ "model.language_model.layers.17.linear_attn.dt_bias": "model-00004-of-00008.safetensors",
160
+ "model.language_model.layers.17.linear_attn.in_proj_a.weight": "model-00004-of-00008.safetensors",
161
+ "model.language_model.layers.17.linear_attn.in_proj_b.weight": "model-00004-of-00008.safetensors",
162
+ "model.language_model.layers.17.linear_attn.in_proj_qkv.weight": "model-00004-of-00008.safetensors",
163
+ "model.language_model.layers.17.linear_attn.in_proj_z.weight": "model-00004-of-00008.safetensors",
164
+ "model.language_model.layers.17.linear_attn.norm.weight": "model-00004-of-00008.safetensors",
165
+ "model.language_model.layers.17.linear_attn.out_proj.weight": "model-00004-of-00008.safetensors",
166
+ "model.language_model.layers.17.mhc_attn.bias": "model-00004-of-00008.safetensors",
167
+ "model.language_model.layers.17.mhc_attn.weight": "model-00004-of-00008.safetensors",
168
+ "model.language_model.layers.17.mhc_ffn.bias": "model-00004-of-00008.safetensors",
169
+ "model.language_model.layers.17.mhc_ffn.weight": "model-00004-of-00008.safetensors",
170
+ "model.language_model.layers.17.mlp.down_proj.weight": "model-00004-of-00008.safetensors",
171
+ "model.language_model.layers.17.mlp.gate_proj.weight": "model-00004-of-00008.safetensors",
172
+ "model.language_model.layers.17.mlp.up_proj.weight": "model-00004-of-00008.safetensors",
173
+ "model.language_model.layers.17.post_attention_layernorm.weight": "model-00004-of-00008.safetensors",
174
+ "model.language_model.layers.18.input_layernorm.weight": "model-00004-of-00008.safetensors",
175
+ "model.language_model.layers.18.linear_attn.A_log": "model-00004-of-00008.safetensors",
176
+ "model.language_model.layers.18.linear_attn.conv1d.weight": "model-00004-of-00008.safetensors",
177
+ "model.language_model.layers.18.linear_attn.dt_bias": "model-00004-of-00008.safetensors",
178
+ "model.language_model.layers.18.linear_attn.in_proj_a.weight": "model-00004-of-00008.safetensors",
179
+ "model.language_model.layers.18.linear_attn.in_proj_b.weight": "model-00004-of-00008.safetensors",
180
+ "model.language_model.layers.18.linear_attn.in_proj_qkv.weight": "model-00004-of-00008.safetensors",
181
+ "model.language_model.layers.18.linear_attn.in_proj_z.weight": "model-00004-of-00008.safetensors",
182
+ "model.language_model.layers.18.linear_attn.norm.weight": "model-00004-of-00008.safetensors",
183
+ "model.language_model.layers.18.linear_attn.out_proj.weight": "model-00004-of-00008.safetensors",
184
+ "model.language_model.layers.18.mhc_attn.bias": "model-00004-of-00008.safetensors",
185
+ "model.language_model.layers.18.mhc_attn.weight": "model-00004-of-00008.safetensors",
186
+ "model.language_model.layers.18.mhc_ffn.bias": "model-00004-of-00008.safetensors",
187
+ "model.language_model.layers.18.mhc_ffn.weight": "model-00004-of-00008.safetensors",
188
+ "model.language_model.layers.18.mlp.down_proj.weight": "model-00004-of-00008.safetensors",
189
+ "model.language_model.layers.18.mlp.gate_proj.weight": "model-00004-of-00008.safetensors",
190
+ "model.language_model.layers.18.mlp.up_proj.weight": "model-00004-of-00008.safetensors",
191
+ "model.language_model.layers.18.post_attention_layernorm.weight": "model-00004-of-00008.safetensors",
192
+ "model.language_model.layers.19.input_layernorm.weight": "model-00004-of-00008.safetensors",
193
+ "model.language_model.layers.19.mhc_attn.bias": "model-00004-of-00008.safetensors",
194
+ "model.language_model.layers.19.mhc_attn.weight": "model-00004-of-00008.safetensors",
195
+ "model.language_model.layers.19.mhc_ffn.bias": "model-00004-of-00008.safetensors",
196
+ "model.language_model.layers.19.mhc_ffn.weight": "model-00004-of-00008.safetensors",
197
+ "model.language_model.layers.19.mlp.down_proj.weight": "model-00004-of-00008.safetensors",
198
+ "model.language_model.layers.19.mlp.gate_proj.weight": "model-00004-of-00008.safetensors",
199
+ "model.language_model.layers.19.mlp.up_proj.weight": "model-00004-of-00008.safetensors",
200
+ "model.language_model.layers.19.post_attention_layernorm.weight": "model-00004-of-00008.safetensors",
201
+ "model.language_model.layers.19.self_attn.k_norm.weight": "model-00004-of-00008.safetensors",
202
+ "model.language_model.layers.19.self_attn.k_proj.weight": "model-00004-of-00008.safetensors",
203
+ "model.language_model.layers.19.self_attn.o_proj.weight": "model-00004-of-00008.safetensors",
204
+ "model.language_model.layers.19.self_attn.q_norm.weight": "model-00004-of-00008.safetensors",
205
+ "model.language_model.layers.19.self_attn.q_proj.weight": "model-00004-of-00008.safetensors",
206
+ "model.language_model.layers.19.self_attn.v_proj.weight": "model-00004-of-00008.safetensors",
207
+ "model.language_model.layers.2.input_layernorm.weight": "model-00004-of-00008.safetensors",
208
+ "model.language_model.layers.2.linear_attn.A_log": "model-00004-of-00008.safetensors",
209
+ "model.language_model.layers.2.linear_attn.conv1d.weight": "model-00004-of-00008.safetensors",
210
+ "model.language_model.layers.2.linear_attn.dt_bias": "model-00004-of-00008.safetensors",
211
+ "model.language_model.layers.2.linear_attn.in_proj_a.weight": "model-00004-of-00008.safetensors",
212
+ "model.language_model.layers.2.linear_attn.in_proj_b.weight": "model-00004-of-00008.safetensors",
213
+ "model.language_model.layers.2.linear_attn.in_proj_qkv.weight": "model-00004-of-00008.safetensors",
214
+ "model.language_model.layers.2.linear_attn.in_proj_z.weight": "model-00004-of-00008.safetensors",
215
+ "model.language_model.layers.2.linear_attn.norm.weight": "model-00004-of-00008.safetensors",
216
+ "model.language_model.layers.2.linear_attn.out_proj.weight": "model-00004-of-00008.safetensors",
217
+ "model.language_model.layers.2.mlp.down_proj.weight": "model-00005-of-00008.safetensors",
218
+ "model.language_model.layers.2.mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
219
+ "model.language_model.layers.2.mlp.up_proj.weight": "model-00005-of-00008.safetensors",
220
+ "model.language_model.layers.2.post_attention_layernorm.weight": "model-00005-of-00008.safetensors",
221
+ "model.language_model.layers.20.input_layernorm.weight": "model-00005-of-00008.safetensors",
222
+ "model.language_model.layers.20.linear_attn.A_log": "model-00005-of-00008.safetensors",
223
+ "model.language_model.layers.20.linear_attn.conv1d.weight": "model-00005-of-00008.safetensors",
224
+ "model.language_model.layers.20.linear_attn.dt_bias": "model-00005-of-00008.safetensors",
225
+ "model.language_model.layers.20.linear_attn.in_proj_a.weight": "model-00005-of-00008.safetensors",
226
+ "model.language_model.layers.20.linear_attn.in_proj_b.weight": "model-00005-of-00008.safetensors",
227
+ "model.language_model.layers.20.linear_attn.in_proj_qkv.weight": "model-00005-of-00008.safetensors",
228
+ "model.language_model.layers.20.linear_attn.in_proj_z.weight": "model-00005-of-00008.safetensors",
229
+ "model.language_model.layers.20.linear_attn.norm.weight": "model-00005-of-00008.safetensors",
230
+ "model.language_model.layers.20.linear_attn.out_proj.weight": "model-00005-of-00008.safetensors",
231
+ "model.language_model.layers.20.mhc_attn.bias": "model-00005-of-00008.safetensors",
232
+ "model.language_model.layers.20.mhc_attn.weight": "model-00005-of-00008.safetensors",
233
+ "model.language_model.layers.20.mhc_ffn.bias": "model-00005-of-00008.safetensors",
234
+ "model.language_model.layers.20.mhc_ffn.weight": "model-00005-of-00008.safetensors",
235
+ "model.language_model.layers.20.mlp.down_proj.weight": "model-00005-of-00008.safetensors",
236
+ "model.language_model.layers.20.mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
237
+ "model.language_model.layers.20.mlp.up_proj.weight": "model-00005-of-00008.safetensors",
238
+ "model.language_model.layers.20.post_attention_layernorm.weight": "model-00005-of-00008.safetensors",
239
+ "model.language_model.layers.21.input_layernorm.weight": "model-00005-of-00008.safetensors",
240
+ "model.language_model.layers.21.linear_attn.A_log": "model-00005-of-00008.safetensors",
241
+ "model.language_model.layers.21.linear_attn.conv1d.weight": "model-00005-of-00008.safetensors",
242
+ "model.language_model.layers.21.linear_attn.dt_bias": "model-00005-of-00008.safetensors",
243
+ "model.language_model.layers.21.linear_attn.in_proj_a.weight": "model-00005-of-00008.safetensors",
244
+ "model.language_model.layers.21.linear_attn.in_proj_b.weight": "model-00005-of-00008.safetensors",
245
+ "model.language_model.layers.21.linear_attn.in_proj_qkv.weight": "model-00005-of-00008.safetensors",
246
+ "model.language_model.layers.21.linear_attn.in_proj_z.weight": "model-00005-of-00008.safetensors",
247
+ "model.language_model.layers.21.linear_attn.norm.weight": "model-00005-of-00008.safetensors",
248
+ "model.language_model.layers.21.linear_attn.out_proj.weight": "model-00005-of-00008.safetensors",
249
+ "model.language_model.layers.21.mhc_attn.bias": "model-00005-of-00008.safetensors",
250
+ "model.language_model.layers.21.mhc_attn.weight": "model-00005-of-00008.safetensors",
251
+ "model.language_model.layers.21.mhc_ffn.bias": "model-00005-of-00008.safetensors",
252
+ "model.language_model.layers.21.mhc_ffn.weight": "model-00005-of-00008.safetensors",
253
+ "model.language_model.layers.21.mlp.down_proj.weight": "model-00005-of-00008.safetensors",
254
+ "model.language_model.layers.21.mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
255
+ "model.language_model.layers.21.mlp.up_proj.weight": "model-00005-of-00008.safetensors",
256
+ "model.language_model.layers.21.post_attention_layernorm.weight": "model-00005-of-00008.safetensors",
257
+ "model.language_model.layers.22.input_layernorm.weight": "model-00005-of-00008.safetensors",
258
+ "model.language_model.layers.22.linear_attn.A_log": "model-00005-of-00008.safetensors",
259
+ "model.language_model.layers.22.linear_attn.conv1d.weight": "model-00005-of-00008.safetensors",
260
+ "model.language_model.layers.22.linear_attn.dt_bias": "model-00005-of-00008.safetensors",
261
+ "model.language_model.layers.22.linear_attn.in_proj_a.weight": "model-00005-of-00008.safetensors",
262
+ "model.language_model.layers.22.linear_attn.in_proj_b.weight": "model-00005-of-00008.safetensors",
263
+ "model.language_model.layers.22.linear_attn.in_proj_qkv.weight": "model-00005-of-00008.safetensors",
264
+ "model.language_model.layers.22.linear_attn.in_proj_z.weight": "model-00005-of-00008.safetensors",
265
+ "model.language_model.layers.22.linear_attn.norm.weight": "model-00005-of-00008.safetensors",
266
+ "model.language_model.layers.22.linear_attn.out_proj.weight": "model-00005-of-00008.safetensors",
267
+ "model.language_model.layers.22.mhc_attn.bias": "model-00005-of-00008.safetensors",
268
+ "model.language_model.layers.22.mhc_attn.weight": "model-00005-of-00008.safetensors",
269
+ "model.language_model.layers.22.mhc_ffn.bias": "model-00005-of-00008.safetensors",
270
+ "model.language_model.layers.22.mhc_ffn.weight": "model-00005-of-00008.safetensors",
271
+ "model.language_model.layers.22.mlp.down_proj.weight": "model-00005-of-00008.safetensors",
272
+ "model.language_model.layers.22.mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
273
+ "model.language_model.layers.22.mlp.up_proj.weight": "model-00005-of-00008.safetensors",
274
+ "model.language_model.layers.22.post_attention_layernorm.weight": "model-00005-of-00008.safetensors",
275
+ "model.language_model.layers.23.input_layernorm.weight": "model-00005-of-00008.safetensors",
276
+ "model.language_model.layers.23.mhc_attn.bias": "model-00005-of-00008.safetensors",
277
+ "model.language_model.layers.23.mhc_attn.weight": "model-00005-of-00008.safetensors",
278
+ "model.language_model.layers.23.mhc_ffn.bias": "model-00005-of-00008.safetensors",
279
+ "model.language_model.layers.23.mhc_ffn.weight": "model-00005-of-00008.safetensors",
280
+ "model.language_model.layers.23.mlp.down_proj.weight": "model-00005-of-00008.safetensors",
281
+ "model.language_model.layers.23.mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
282
+ "model.language_model.layers.23.mlp.up_proj.weight": "model-00005-of-00008.safetensors",
283
+ "model.language_model.layers.23.post_attention_layernorm.weight": "model-00005-of-00008.safetensors",
284
+ "model.language_model.layers.23.self_attn.k_norm.weight": "model-00005-of-00008.safetensors",
285
+ "model.language_model.layers.23.self_attn.k_proj.weight": "model-00005-of-00008.safetensors",
286
+ "model.language_model.layers.23.self_attn.o_proj.weight": "model-00005-of-00008.safetensors",
287
+ "model.language_model.layers.23.self_attn.q_norm.weight": "model-00005-of-00008.safetensors",
288
+ "model.language_model.layers.23.self_attn.q_proj.weight": "model-00005-of-00008.safetensors",
289
+ "model.language_model.layers.23.self_attn.v_proj.weight": "model-00005-of-00008.safetensors",
290
+ "model.language_model.layers.24.input_layernorm.weight": "model-00005-of-00008.safetensors",
291
+ "model.language_model.layers.24.linear_attn.A_log": "model-00005-of-00008.safetensors",
292
+ "model.language_model.layers.24.linear_attn.conv1d.weight": "model-00005-of-00008.safetensors",
293
+ "model.language_model.layers.24.linear_attn.dt_bias": "model-00005-of-00008.safetensors",
294
+ "model.language_model.layers.24.linear_attn.in_proj_a.weight": "model-00005-of-00008.safetensors",
295
+ "model.language_model.layers.24.linear_attn.in_proj_b.weight": "model-00005-of-00008.safetensors",
296
+ "model.language_model.layers.24.linear_attn.in_proj_qkv.weight": "model-00005-of-00008.safetensors",
297
+ "model.language_model.layers.24.linear_attn.in_proj_z.weight": "model-00005-of-00008.safetensors",
298
+ "model.language_model.layers.24.linear_attn.norm.weight": "model-00005-of-00008.safetensors",
299
+ "model.language_model.layers.24.linear_attn.out_proj.weight": "model-00005-of-00008.safetensors",
300
+ "model.language_model.layers.24.mhc_attn.bias": "model-00005-of-00008.safetensors",
301
+ "model.language_model.layers.24.mhc_attn.weight": "model-00005-of-00008.safetensors",
302
+ "model.language_model.layers.24.mhc_ffn.bias": "model-00005-of-00008.safetensors",
303
+ "model.language_model.layers.24.mhc_ffn.weight": "model-00005-of-00008.safetensors",
304
+ "model.language_model.layers.24.mlp.down_proj.weight": "model-00005-of-00008.safetensors",
305
+ "model.language_model.layers.24.mlp.gate_proj.weight": "model-00005-of-00008.safetensors",
306
+ "model.language_model.layers.24.mlp.up_proj.weight": "model-00005-of-00008.safetensors",
307
+ "model.language_model.layers.24.post_attention_layernorm.weight": "model-00005-of-00008.safetensors",
308
+ "model.language_model.layers.25.input_layernorm.weight": "model-00005-of-00008.safetensors",
309
+ "model.language_model.layers.25.linear_attn.A_log": "model-00005-of-00008.safetensors",
310
+ "model.language_model.layers.25.linear_attn.conv1d.weight": "model-00005-of-00008.safetensors",
311
+ "model.language_model.layers.25.linear_attn.dt_bias": "model-00005-of-00008.safetensors",
312
+ "model.language_model.layers.25.linear_attn.in_proj_a.weight": "model-00005-of-00008.safetensors",
313
+ "model.language_model.layers.25.linear_attn.in_proj_b.weight": "model-00005-of-00008.safetensors",
314
+ "model.language_model.layers.25.linear_attn.in_proj_qkv.weight": "model-00006-of-00008.safetensors",
315
+ "model.language_model.layers.25.linear_attn.in_proj_z.weight": "model-00006-of-00008.safetensors",
316
+ "model.language_model.layers.25.linear_attn.norm.weight": "model-00006-of-00008.safetensors",
317
+ "model.language_model.layers.25.linear_attn.out_proj.weight": "model-00006-of-00008.safetensors",
318
+ "model.language_model.layers.25.mhc_attn.bias": "model-00006-of-00008.safetensors",
319
+ "model.language_model.layers.25.mhc_attn.weight": "model-00006-of-00008.safetensors",
320
+ "model.language_model.layers.25.mhc_ffn.bias": "model-00006-of-00008.safetensors",
321
+ "model.language_model.layers.25.mhc_ffn.weight": "model-00006-of-00008.safetensors",
322
+ "model.language_model.layers.25.mlp.down_proj.weight": "model-00006-of-00008.safetensors",
323
+ "model.language_model.layers.25.mlp.gate_proj.weight": "model-00006-of-00008.safetensors",
324
+ "model.language_model.layers.25.mlp.up_proj.weight": "model-00006-of-00008.safetensors",
325
+ "model.language_model.layers.25.post_attention_layernorm.weight": "model-00006-of-00008.safetensors",
326
+ "model.language_model.layers.26.input_layernorm.weight": "model-00006-of-00008.safetensors",
327
+ "model.language_model.layers.26.linear_attn.A_log": "model-00006-of-00008.safetensors",
328
+ "model.language_model.layers.26.linear_attn.conv1d.weight": "model-00006-of-00008.safetensors",
329
+ "model.language_model.layers.26.linear_attn.dt_bias": "model-00006-of-00008.safetensors",
330
+ "model.language_model.layers.26.linear_attn.in_proj_a.weight": "model-00006-of-00008.safetensors",
331
+ "model.language_model.layers.26.linear_attn.in_proj_b.weight": "model-00006-of-00008.safetensors",
332
+ "model.language_model.layers.26.linear_attn.in_proj_qkv.weight": "model-00006-of-00008.safetensors",
333
+ "model.language_model.layers.26.linear_attn.in_proj_z.weight": "model-00006-of-00008.safetensors",
334
+ "model.language_model.layers.26.linear_attn.norm.weight": "model-00006-of-00008.safetensors",
335
+ "model.language_model.layers.26.linear_attn.out_proj.weight": "model-00006-of-00008.safetensors",
336
+ "model.language_model.layers.26.mhc_attn.bias": "model-00006-of-00008.safetensors",
337
+ "model.language_model.layers.26.mhc_attn.weight": "model-00006-of-00008.safetensors",
338
+ "model.language_model.layers.26.mhc_ffn.bias": "model-00006-of-00008.safetensors",
339
+ "model.language_model.layers.26.mhc_ffn.weight": "model-00006-of-00008.safetensors",
340
+ "model.language_model.layers.26.mlp.down_proj.weight": "model-00006-of-00008.safetensors",
341
+ "model.language_model.layers.26.mlp.gate_proj.weight": "model-00006-of-00008.safetensors",
342
+ "model.language_model.layers.26.mlp.up_proj.weight": "model-00006-of-00008.safetensors",
343
+ "model.language_model.layers.26.post_attention_layernorm.weight": "model-00006-of-00008.safetensors",
344
+ "model.language_model.layers.27.input_layernorm.weight": "model-00006-of-00008.safetensors",
345
+ "model.language_model.layers.27.mhc_attn.bias": "model-00006-of-00008.safetensors",
346
+ "model.language_model.layers.27.mhc_attn.weight": "model-00006-of-00008.safetensors",
347
+ "model.language_model.layers.27.mhc_ffn.bias": "model-00006-of-00008.safetensors",
348
+ "model.language_model.layers.27.mhc_ffn.weight": "model-00006-of-00008.safetensors",
349
+ "model.language_model.layers.27.mlp.down_proj.weight": "model-00006-of-00008.safetensors",
350
+ "model.language_model.layers.27.mlp.gate_proj.weight": "model-00006-of-00008.safetensors",
351
+ "model.language_model.layers.27.mlp.up_proj.weight": "model-00006-of-00008.safetensors",
352
+ "model.language_model.layers.27.post_attention_layernorm.weight": "model-00006-of-00008.safetensors",
353
+ "model.language_model.layers.27.self_attn.k_norm.weight": "model-00006-of-00008.safetensors",
354
+ "model.language_model.layers.27.self_attn.k_proj.weight": "model-00006-of-00008.safetensors",
355
+ "model.language_model.layers.27.self_attn.o_proj.weight": "model-00006-of-00008.safetensors",
356
+ "model.language_model.layers.27.self_attn.q_norm.weight": "model-00006-of-00008.safetensors",
357
+ "model.language_model.layers.27.self_attn.q_proj.weight": "model-00006-of-00008.safetensors",
358
+ "model.language_model.layers.27.self_attn.v_proj.weight": "model-00006-of-00008.safetensors",
359
+ "model.language_model.layers.28.input_layernorm.weight": "model-00006-of-00008.safetensors",
360
+ "model.language_model.layers.28.linear_attn.A_log": "model-00006-of-00008.safetensors",
361
+ "model.language_model.layers.28.linear_attn.conv1d.weight": "model-00006-of-00008.safetensors",
362
+ "model.language_model.layers.28.linear_attn.dt_bias": "model-00006-of-00008.safetensors",
363
+ "model.language_model.layers.28.linear_attn.in_proj_a.weight": "model-00006-of-00008.safetensors",
364
+ "model.language_model.layers.28.linear_attn.in_proj_b.weight": "model-00006-of-00008.safetensors",
365
+ "model.language_model.layers.28.linear_attn.in_proj_qkv.weight": "model-00006-of-00008.safetensors",
366
+ "model.language_model.layers.28.linear_attn.in_proj_z.weight": "model-00006-of-00008.safetensors",
367
+ "model.language_model.layers.28.linear_attn.norm.weight": "model-00006-of-00008.safetensors",
368
+ "model.language_model.layers.28.linear_attn.out_proj.weight": "model-00006-of-00008.safetensors",
369
+ "model.language_model.layers.28.mlp.down_proj.weight": "model-00006-of-00008.safetensors",
370
+ "model.language_model.layers.28.mlp.gate_proj.weight": "model-00006-of-00008.safetensors",
371
+ "model.language_model.layers.28.mlp.up_proj.weight": "model-00006-of-00008.safetensors",
372
+ "model.language_model.layers.28.post_attention_layernorm.weight": "model-00006-of-00008.safetensors",
373
+ "model.language_model.layers.29.input_layernorm.weight": "model-00006-of-00008.safetensors",
374
+ "model.language_model.layers.29.linear_attn.A_log": "model-00006-of-00008.safetensors",
375
+ "model.language_model.layers.29.linear_attn.conv1d.weight": "model-00006-of-00008.safetensors",
376
+ "model.language_model.layers.29.linear_attn.dt_bias": "model-00006-of-00008.safetensors",
377
+ "model.language_model.layers.29.linear_attn.in_proj_a.weight": "model-00006-of-00008.safetensors",
378
+ "model.language_model.layers.29.linear_attn.in_proj_b.weight": "model-00006-of-00008.safetensors",
379
+ "model.language_model.layers.29.linear_attn.in_proj_qkv.weight": "model-00006-of-00008.safetensors",
380
+ "model.language_model.layers.29.linear_attn.in_proj_z.weight": "model-00006-of-00008.safetensors",
381
+ "model.language_model.layers.29.linear_attn.norm.weight": "model-00006-of-00008.safetensors",
382
+ "model.language_model.layers.29.linear_attn.out_proj.weight": "model-00006-of-00008.safetensors",
383
+ "model.language_model.layers.29.mlp.down_proj.weight": "model-00006-of-00008.safetensors",
384
+ "model.language_model.layers.29.mlp.gate_proj.weight": "model-00006-of-00008.safetensors",
385
+ "model.language_model.layers.29.mlp.up_proj.weight": "model-00006-of-00008.safetensors",
386
+ "model.language_model.layers.29.post_attention_layernorm.weight": "model-00006-of-00008.safetensors",
387
+ "model.language_model.layers.3.input_layernorm.weight": "model-00006-of-00008.safetensors",
388
+ "model.language_model.layers.3.mlp.down_proj.weight": "model-00006-of-00008.safetensors",
389
+ "model.language_model.layers.3.mlp.gate_proj.weight": "model-00006-of-00008.safetensors",
390
+ "model.language_model.layers.3.mlp.up_proj.weight": "model-00006-of-00008.safetensors",
391
+ "model.language_model.layers.3.post_attention_layernorm.weight": "model-00006-of-00008.safetensors",
392
+ "model.language_model.layers.3.self_attn.k_norm.weight": "model-00006-of-00008.safetensors",
393
+ "model.language_model.layers.3.self_attn.k_proj.weight": "model-00006-of-00008.safetensors",
394
+ "model.language_model.layers.3.self_attn.o_proj.weight": "model-00007-of-00008.safetensors",
395
+ "model.language_model.layers.3.self_attn.q_norm.weight": "model-00007-of-00008.safetensors",
396
+ "model.language_model.layers.3.self_attn.q_proj.weight": "model-00007-of-00008.safetensors",
397
+ "model.language_model.layers.3.self_attn.v_proj.weight": "model-00007-of-00008.safetensors",
398
+ "model.language_model.layers.30.input_layernorm.weight": "model-00007-of-00008.safetensors",
399
+ "model.language_model.layers.30.linear_attn.A_log": "model-00007-of-00008.safetensors",
400
+ "model.language_model.layers.30.linear_attn.conv1d.weight": "model-00007-of-00008.safetensors",
401
+ "model.language_model.layers.30.linear_attn.dt_bias": "model-00007-of-00008.safetensors",
402
+ "model.language_model.layers.30.linear_attn.in_proj_a.weight": "model-00007-of-00008.safetensors",
403
+ "model.language_model.layers.30.linear_attn.in_proj_b.weight": "model-00007-of-00008.safetensors",
404
+ "model.language_model.layers.30.linear_attn.in_proj_qkv.weight": "model-00007-of-00008.safetensors",
405
+ "model.language_model.layers.30.linear_attn.in_proj_z.weight": "model-00007-of-00008.safetensors",
406
+ "model.language_model.layers.30.linear_attn.norm.weight": "model-00007-of-00008.safetensors",
407
+ "model.language_model.layers.30.linear_attn.out_proj.weight": "model-00007-of-00008.safetensors",
408
+ "model.language_model.layers.30.mlp.down_proj.weight": "model-00007-of-00008.safetensors",
409
+ "model.language_model.layers.30.mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
410
+ "model.language_model.layers.30.mlp.up_proj.weight": "model-00007-of-00008.safetensors",
411
+ "model.language_model.layers.30.post_attention_layernorm.weight": "model-00007-of-00008.safetensors",
412
+ "model.language_model.layers.31.input_layernorm.weight": "model-00007-of-00008.safetensors",
413
+ "model.language_model.layers.31.mlp.down_proj.weight": "model-00007-of-00008.safetensors",
414
+ "model.language_model.layers.31.mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
415
+ "model.language_model.layers.31.mlp.up_proj.weight": "model-00007-of-00008.safetensors",
416
+ "model.language_model.layers.31.post_attention_layernorm.weight": "model-00007-of-00008.safetensors",
417
+ "model.language_model.layers.31.self_attn.k_norm.weight": "model-00007-of-00008.safetensors",
418
+ "model.language_model.layers.31.self_attn.k_proj.weight": "model-00007-of-00008.safetensors",
419
+ "model.language_model.layers.31.self_attn.o_proj.weight": "model-00007-of-00008.safetensors",
420
+ "model.language_model.layers.31.self_attn.q_norm.weight": "model-00007-of-00008.safetensors",
421
+ "model.language_model.layers.31.self_attn.q_proj.weight": "model-00007-of-00008.safetensors",
422
+ "model.language_model.layers.31.self_attn.v_proj.weight": "model-00007-of-00008.safetensors",
423
+ "model.language_model.layers.4.input_layernorm.weight": "model-00007-of-00008.safetensors",
424
+ "model.language_model.layers.4.linear_attn.A_log": "model-00007-of-00008.safetensors",
425
+ "model.language_model.layers.4.linear_attn.conv1d.weight": "model-00007-of-00008.safetensors",
426
+ "model.language_model.layers.4.linear_attn.dt_bias": "model-00007-of-00008.safetensors",
427
+ "model.language_model.layers.4.linear_attn.in_proj_a.weight": "model-00007-of-00008.safetensors",
428
+ "model.language_model.layers.4.linear_attn.in_proj_b.weight": "model-00007-of-00008.safetensors",
429
+ "model.language_model.layers.4.linear_attn.in_proj_qkv.weight": "model-00007-of-00008.safetensors",
430
+ "model.language_model.layers.4.linear_attn.in_proj_z.weight": "model-00007-of-00008.safetensors",
431
+ "model.language_model.layers.4.linear_attn.norm.weight": "model-00007-of-00008.safetensors",
432
+ "model.language_model.layers.4.linear_attn.out_proj.weight": "model-00007-of-00008.safetensors",
433
+ "model.language_model.layers.4.mhc_attn.bias": "model-00007-of-00008.safetensors",
434
+ "model.language_model.layers.4.mhc_attn.weight": "model-00007-of-00008.safetensors",
435
+ "model.language_model.layers.4.mhc_ffn.bias": "model-00007-of-00008.safetensors",
436
+ "model.language_model.layers.4.mhc_ffn.weight": "model-00007-of-00008.safetensors",
437
+ "model.language_model.layers.4.mlp.down_proj.weight": "model-00007-of-00008.safetensors",
438
+ "model.language_model.layers.4.mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
439
+ "model.language_model.layers.4.mlp.up_proj.weight": "model-00007-of-00008.safetensors",
440
+ "model.language_model.layers.4.post_attention_layernorm.weight": "model-00007-of-00008.safetensors",
441
+ "model.language_model.layers.5.input_layernorm.weight": "model-00007-of-00008.safetensors",
442
+ "model.language_model.layers.5.linear_attn.A_log": "model-00007-of-00008.safetensors",
443
+ "model.language_model.layers.5.linear_attn.conv1d.weight": "model-00007-of-00008.safetensors",
444
+ "model.language_model.layers.5.linear_attn.dt_bias": "model-00007-of-00008.safetensors",
445
+ "model.language_model.layers.5.linear_attn.in_proj_a.weight": "model-00007-of-00008.safetensors",
446
+ "model.language_model.layers.5.linear_attn.in_proj_b.weight": "model-00007-of-00008.safetensors",
447
+ "model.language_model.layers.5.linear_attn.in_proj_qkv.weight": "model-00007-of-00008.safetensors",
448
+ "model.language_model.layers.5.linear_attn.in_proj_z.weight": "model-00007-of-00008.safetensors",
449
+ "model.language_model.layers.5.linear_attn.norm.weight": "model-00007-of-00008.safetensors",
450
+ "model.language_model.layers.5.linear_attn.out_proj.weight": "model-00007-of-00008.safetensors",
451
+ "model.language_model.layers.5.mhc_attn.bias": "model-00007-of-00008.safetensors",
452
+ "model.language_model.layers.5.mhc_attn.weight": "model-00007-of-00008.safetensors",
453
+ "model.language_model.layers.5.mhc_ffn.bias": "model-00007-of-00008.safetensors",
454
+ "model.language_model.layers.5.mhc_ffn.weight": "model-00007-of-00008.safetensors",
455
+ "model.language_model.layers.5.mlp.down_proj.weight": "model-00007-of-00008.safetensors",
456
+ "model.language_model.layers.5.mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
457
+ "model.language_model.layers.5.mlp.up_proj.weight": "model-00007-of-00008.safetensors",
458
+ "model.language_model.layers.5.post_attention_layernorm.weight": "model-00007-of-00008.safetensors",
459
+ "model.language_model.layers.6.input_layernorm.weight": "model-00007-of-00008.safetensors",
460
+ "model.language_model.layers.6.linear_attn.A_log": "model-00007-of-00008.safetensors",
461
+ "model.language_model.layers.6.linear_attn.conv1d.weight": "model-00007-of-00008.safetensors",
462
+ "model.language_model.layers.6.linear_attn.dt_bias": "model-00007-of-00008.safetensors",
463
+ "model.language_model.layers.6.linear_attn.in_proj_a.weight": "model-00007-of-00008.safetensors",
464
+ "model.language_model.layers.6.linear_attn.in_proj_b.weight": "model-00007-of-00008.safetensors",
465
+ "model.language_model.layers.6.linear_attn.in_proj_qkv.weight": "model-00007-of-00008.safetensors",
466
+ "model.language_model.layers.6.linear_attn.in_proj_z.weight": "model-00007-of-00008.safetensors",
467
+ "model.language_model.layers.6.linear_attn.norm.weight": "model-00007-of-00008.safetensors",
468
+ "model.language_model.layers.6.linear_attn.out_proj.weight": "model-00007-of-00008.safetensors",
469
+ "model.language_model.layers.6.mhc_attn.bias": "model-00007-of-00008.safetensors",
470
+ "model.language_model.layers.6.mhc_attn.weight": "model-00007-of-00008.safetensors",
471
+ "model.language_model.layers.6.mhc_ffn.bias": "model-00007-of-00008.safetensors",
472
+ "model.language_model.layers.6.mhc_ffn.weight": "model-00007-of-00008.safetensors",
473
+ "model.language_model.layers.6.mlp.down_proj.weight": "model-00007-of-00008.safetensors",
474
+ "model.language_model.layers.6.mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
475
+ "model.language_model.layers.6.mlp.up_proj.weight": "model-00007-of-00008.safetensors",
476
+ "model.language_model.layers.6.post_attention_layernorm.weight": "model-00007-of-00008.safetensors",
477
+ "model.language_model.layers.7.input_layernorm.weight": "model-00007-of-00008.safetensors",
478
+ "model.language_model.layers.7.mhc_attn.bias": "model-00007-of-00008.safetensors",
479
+ "model.language_model.layers.7.mhc_attn.weight": "model-00007-of-00008.safetensors",
480
+ "model.language_model.layers.7.mhc_ffn.bias": "model-00007-of-00008.safetensors",
481
+ "model.language_model.layers.7.mhc_ffn.weight": "model-00007-of-00008.safetensors",
482
+ "model.language_model.layers.7.mlp.down_proj.weight": "model-00007-of-00008.safetensors",
483
+ "model.language_model.layers.7.mlp.gate_proj.weight": "model-00007-of-00008.safetensors",
484
+ "model.language_model.layers.7.mlp.up_proj.weight": "model-00008-of-00008.safetensors",
485
+ "model.language_model.layers.7.post_attention_layernorm.weight": "model-00008-of-00008.safetensors",
486
+ "model.language_model.layers.7.self_attn.k_norm.weight": "model-00008-of-00008.safetensors",
487
+ "model.language_model.layers.7.self_attn.k_proj.weight": "model-00008-of-00008.safetensors",
488
+ "model.language_model.layers.7.self_attn.o_proj.weight": "model-00008-of-00008.safetensors",
489
+ "model.language_model.layers.7.self_attn.q_norm.weight": "model-00008-of-00008.safetensors",
490
+ "model.language_model.layers.7.self_attn.q_proj.weight": "model-00008-of-00008.safetensors",
491
+ "model.language_model.layers.7.self_attn.v_proj.weight": "model-00008-of-00008.safetensors",
492
+ "model.language_model.layers.8.input_layernorm.weight": "model-00008-of-00008.safetensors",
493
+ "model.language_model.layers.8.linear_attn.A_log": "model-00008-of-00008.safetensors",
494
+ "model.language_model.layers.8.linear_attn.conv1d.weight": "model-00008-of-00008.safetensors",
495
+ "model.language_model.layers.8.linear_attn.dt_bias": "model-00008-of-00008.safetensors",
496
+ "model.language_model.layers.8.linear_attn.in_proj_a.weight": "model-00008-of-00008.safetensors",
497
+ "model.language_model.layers.8.linear_attn.in_proj_b.weight": "model-00008-of-00008.safetensors",
498
+ "model.language_model.layers.8.linear_attn.in_proj_qkv.weight": "model-00008-of-00008.safetensors",
499
+ "model.language_model.layers.8.linear_attn.in_proj_z.weight": "model-00008-of-00008.safetensors",
500
+ "model.language_model.layers.8.linear_attn.norm.weight": "model-00008-of-00008.safetensors",
501
+ "model.language_model.layers.8.linear_attn.out_proj.weight": "model-00008-of-00008.safetensors",
502
+ "model.language_model.layers.8.mhc_attn.bias": "model-00008-of-00008.safetensors",
503
+ "model.language_model.layers.8.mhc_attn.weight": "model-00008-of-00008.safetensors",
504
+ "model.language_model.layers.8.mhc_ffn.bias": "model-00008-of-00008.safetensors",
505
+ "model.language_model.layers.8.mhc_ffn.weight": "model-00008-of-00008.safetensors",
506
+ "model.language_model.layers.8.mlp.down_proj.weight": "model-00008-of-00008.safetensors",
507
+ "model.language_model.layers.8.mlp.gate_proj.weight": "model-00008-of-00008.safetensors",
508
+ "model.language_model.layers.8.mlp.up_proj.weight": "model-00008-of-00008.safetensors",
509
+ "model.language_model.layers.8.post_attention_layernorm.weight": "model-00008-of-00008.safetensors",
510
+ "model.language_model.layers.9.input_layernorm.weight": "model-00008-of-00008.safetensors",
511
+ "model.language_model.layers.9.linear_attn.A_log": "model-00008-of-00008.safetensors",
512
+ "model.language_model.layers.9.linear_attn.conv1d.weight": "model-00008-of-00008.safetensors",
513
+ "model.language_model.layers.9.linear_attn.dt_bias": "model-00008-of-00008.safetensors",
514
+ "model.language_model.layers.9.linear_attn.in_proj_a.weight": "model-00008-of-00008.safetensors",
515
+ "model.language_model.layers.9.linear_attn.in_proj_b.weight": "model-00008-of-00008.safetensors",
516
+ "model.language_model.layers.9.linear_attn.in_proj_qkv.weight": "model-00008-of-00008.safetensors",
517
+ "model.language_model.layers.9.linear_attn.in_proj_z.weight": "model-00008-of-00008.safetensors",
518
+ "model.language_model.layers.9.linear_attn.norm.weight": "model-00008-of-00008.safetensors",
519
+ "model.language_model.layers.9.linear_attn.out_proj.weight": "model-00008-of-00008.safetensors",
520
+ "model.language_model.layers.9.mhc_attn.bias": "model-00008-of-00008.safetensors",
521
+ "model.language_model.layers.9.mhc_attn.weight": "model-00008-of-00008.safetensors",
522
+ "model.language_model.layers.9.mhc_ffn.bias": "model-00008-of-00008.safetensors",
523
+ "model.language_model.layers.9.mhc_ffn.weight": "model-00008-of-00008.safetensors",
524
+ "model.language_model.layers.9.mlp.down_proj.weight": "model-00008-of-00008.safetensors",
525
+ "model.language_model.layers.9.mlp.gate_proj.weight": "model-00008-of-00008.safetensors",
526
+ "model.language_model.layers.9.mlp.up_proj.weight": "model-00008-of-00008.safetensors",
527
+ "model.language_model.layers.9.post_attention_layernorm.weight": "model-00008-of-00008.safetensors",
528
+ "model.language_model.norm.weight": "model-00008-of-00008.safetensors",
529
+ "model.visual.blocks.0.attn.proj.bias": "model-00008-of-00008.safetensors",
530
+ "model.visual.blocks.0.attn.proj.weight": "model-00008-of-00008.safetensors",
531
+ "model.visual.blocks.0.attn.qkv.bias": "model-00008-of-00008.safetensors",
532
+ "model.visual.blocks.0.attn.qkv.weight": "model-00008-of-00008.safetensors",
533
+ "model.visual.blocks.0.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
534
+ "model.visual.blocks.0.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
535
+ "model.visual.blocks.0.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
536
+ "model.visual.blocks.0.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
537
+ "model.visual.blocks.0.norm1.bias": "model-00008-of-00008.safetensors",
538
+ "model.visual.blocks.0.norm1.weight": "model-00008-of-00008.safetensors",
539
+ "model.visual.blocks.0.norm2.bias": "model-00008-of-00008.safetensors",
540
+ "model.visual.blocks.0.norm2.weight": "model-00008-of-00008.safetensors",
541
+ "model.visual.blocks.1.attn.proj.bias": "model-00008-of-00008.safetensors",
542
+ "model.visual.blocks.1.attn.proj.weight": "model-00008-of-00008.safetensors",
543
+ "model.visual.blocks.1.attn.qkv.bias": "model-00008-of-00008.safetensors",
544
+ "model.visual.blocks.1.attn.qkv.weight": "model-00008-of-00008.safetensors",
545
+ "model.visual.blocks.1.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
546
+ "model.visual.blocks.1.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
547
+ "model.visual.blocks.1.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
548
+ "model.visual.blocks.1.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
549
+ "model.visual.blocks.1.norm1.bias": "model-00008-of-00008.safetensors",
550
+ "model.visual.blocks.1.norm1.weight": "model-00008-of-00008.safetensors",
551
+ "model.visual.blocks.1.norm2.bias": "model-00008-of-00008.safetensors",
552
+ "model.visual.blocks.1.norm2.weight": "model-00008-of-00008.safetensors",
553
+ "model.visual.blocks.10.attn.proj.bias": "model-00008-of-00008.safetensors",
554
+ "model.visual.blocks.10.attn.proj.weight": "model-00008-of-00008.safetensors",
555
+ "model.visual.blocks.10.attn.qkv.bias": "model-00008-of-00008.safetensors",
556
+ "model.visual.blocks.10.attn.qkv.weight": "model-00008-of-00008.safetensors",
557
+ "model.visual.blocks.10.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
558
+ "model.visual.blocks.10.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
559
+ "model.visual.blocks.10.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
560
+ "model.visual.blocks.10.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
561
+ "model.visual.blocks.10.norm1.bias": "model-00008-of-00008.safetensors",
562
+ "model.visual.blocks.10.norm1.weight": "model-00008-of-00008.safetensors",
563
+ "model.visual.blocks.10.norm2.bias": "model-00008-of-00008.safetensors",
564
+ "model.visual.blocks.10.norm2.weight": "model-00008-of-00008.safetensors",
565
+ "model.visual.blocks.11.attn.proj.bias": "model-00008-of-00008.safetensors",
566
+ "model.visual.blocks.11.attn.proj.weight": "model-00008-of-00008.safetensors",
567
+ "model.visual.blocks.11.attn.qkv.bias": "model-00008-of-00008.safetensors",
568
+ "model.visual.blocks.11.attn.qkv.weight": "model-00008-of-00008.safetensors",
569
+ "model.visual.blocks.11.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
570
+ "model.visual.blocks.11.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
571
+ "model.visual.blocks.11.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
572
+ "model.visual.blocks.11.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
573
+ "model.visual.blocks.11.norm1.bias": "model-00008-of-00008.safetensors",
574
+ "model.visual.blocks.11.norm1.weight": "model-00008-of-00008.safetensors",
575
+ "model.visual.blocks.11.norm2.bias": "model-00008-of-00008.safetensors",
576
+ "model.visual.blocks.11.norm2.weight": "model-00008-of-00008.safetensors",
577
+ "model.visual.blocks.12.attn.proj.bias": "model-00008-of-00008.safetensors",
578
+ "model.visual.blocks.12.attn.proj.weight": "model-00008-of-00008.safetensors",
579
+ "model.visual.blocks.12.attn.qkv.bias": "model-00008-of-00008.safetensors",
580
+ "model.visual.blocks.12.attn.qkv.weight": "model-00008-of-00008.safetensors",
581
+ "model.visual.blocks.12.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
582
+ "model.visual.blocks.12.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
583
+ "model.visual.blocks.12.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
584
+ "model.visual.blocks.12.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
585
+ "model.visual.blocks.12.norm1.bias": "model-00008-of-00008.safetensors",
586
+ "model.visual.blocks.12.norm1.weight": "model-00008-of-00008.safetensors",
587
+ "model.visual.blocks.12.norm2.bias": "model-00008-of-00008.safetensors",
588
+ "model.visual.blocks.12.norm2.weight": "model-00008-of-00008.safetensors",
589
+ "model.visual.blocks.13.attn.proj.bias": "model-00008-of-00008.safetensors",
590
+ "model.visual.blocks.13.attn.proj.weight": "model-00008-of-00008.safetensors",
591
+ "model.visual.blocks.13.attn.qkv.bias": "model-00008-of-00008.safetensors",
592
+ "model.visual.blocks.13.attn.qkv.weight": "model-00008-of-00008.safetensors",
593
+ "model.visual.blocks.13.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
594
+ "model.visual.blocks.13.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
595
+ "model.visual.blocks.13.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
596
+ "model.visual.blocks.13.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
597
+ "model.visual.blocks.13.norm1.bias": "model-00008-of-00008.safetensors",
598
+ "model.visual.blocks.13.norm1.weight": "model-00008-of-00008.safetensors",
599
+ "model.visual.blocks.13.norm2.bias": "model-00008-of-00008.safetensors",
600
+ "model.visual.blocks.13.norm2.weight": "model-00008-of-00008.safetensors",
601
+ "model.visual.blocks.14.attn.proj.bias": "model-00008-of-00008.safetensors",
602
+ "model.visual.blocks.14.attn.proj.weight": "model-00008-of-00008.safetensors",
603
+ "model.visual.blocks.14.attn.qkv.bias": "model-00008-of-00008.safetensors",
604
+ "model.visual.blocks.14.attn.qkv.weight": "model-00008-of-00008.safetensors",
605
+ "model.visual.blocks.14.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
606
+ "model.visual.blocks.14.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
607
+ "model.visual.blocks.14.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
608
+ "model.visual.blocks.14.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
609
+ "model.visual.blocks.14.norm1.bias": "model-00008-of-00008.safetensors",
610
+ "model.visual.blocks.14.norm1.weight": "model-00008-of-00008.safetensors",
611
+ "model.visual.blocks.14.norm2.bias": "model-00008-of-00008.safetensors",
612
+ "model.visual.blocks.14.norm2.weight": "model-00008-of-00008.safetensors",
613
+ "model.visual.blocks.15.attn.proj.bias": "model-00008-of-00008.safetensors",
614
+ "model.visual.blocks.15.attn.proj.weight": "model-00008-of-00008.safetensors",
615
+ "model.visual.blocks.15.attn.qkv.bias": "model-00008-of-00008.safetensors",
616
+ "model.visual.blocks.15.attn.qkv.weight": "model-00008-of-00008.safetensors",
617
+ "model.visual.blocks.15.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
618
+ "model.visual.blocks.15.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
619
+ "model.visual.blocks.15.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
620
+ "model.visual.blocks.15.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
621
+ "model.visual.blocks.15.norm1.bias": "model-00008-of-00008.safetensors",
622
+ "model.visual.blocks.15.norm1.weight": "model-00008-of-00008.safetensors",
623
+ "model.visual.blocks.15.norm2.bias": "model-00008-of-00008.safetensors",
624
+ "model.visual.blocks.15.norm2.weight": "model-00008-of-00008.safetensors",
625
+ "model.visual.blocks.16.attn.proj.bias": "model-00008-of-00008.safetensors",
626
+ "model.visual.blocks.16.attn.proj.weight": "model-00008-of-00008.safetensors",
627
+ "model.visual.blocks.16.attn.qkv.bias": "model-00008-of-00008.safetensors",
628
+ "model.visual.blocks.16.attn.qkv.weight": "model-00008-of-00008.safetensors",
629
+ "model.visual.blocks.16.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
630
+ "model.visual.blocks.16.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
631
+ "model.visual.blocks.16.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
632
+ "model.visual.blocks.16.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
633
+ "model.visual.blocks.16.norm1.bias": "model-00008-of-00008.safetensors",
634
+ "model.visual.blocks.16.norm1.weight": "model-00008-of-00008.safetensors",
635
+ "model.visual.blocks.16.norm2.bias": "model-00008-of-00008.safetensors",
636
+ "model.visual.blocks.16.norm2.weight": "model-00008-of-00008.safetensors",
637
+ "model.visual.blocks.17.attn.proj.bias": "model-00008-of-00008.safetensors",
638
+ "model.visual.blocks.17.attn.proj.weight": "model-00008-of-00008.safetensors",
639
+ "model.visual.blocks.17.attn.qkv.bias": "model-00008-of-00008.safetensors",
640
+ "model.visual.blocks.17.attn.qkv.weight": "model-00008-of-00008.safetensors",
641
+ "model.visual.blocks.17.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
642
+ "model.visual.blocks.17.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
643
+ "model.visual.blocks.17.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
644
+ "model.visual.blocks.17.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
645
+ "model.visual.blocks.17.norm1.bias": "model-00008-of-00008.safetensors",
646
+ "model.visual.blocks.17.norm1.weight": "model-00008-of-00008.safetensors",
647
+ "model.visual.blocks.17.norm2.bias": "model-00008-of-00008.safetensors",
648
+ "model.visual.blocks.17.norm2.weight": "model-00008-of-00008.safetensors",
649
+ "model.visual.blocks.18.attn.proj.bias": "model-00008-of-00008.safetensors",
650
+ "model.visual.blocks.18.attn.proj.weight": "model-00008-of-00008.safetensors",
651
+ "model.visual.blocks.18.attn.qkv.bias": "model-00008-of-00008.safetensors",
652
+ "model.visual.blocks.18.attn.qkv.weight": "model-00008-of-00008.safetensors",
653
+ "model.visual.blocks.18.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
654
+ "model.visual.blocks.18.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
655
+ "model.visual.blocks.18.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
656
+ "model.visual.blocks.18.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
657
+ "model.visual.blocks.18.norm1.bias": "model-00008-of-00008.safetensors",
658
+ "model.visual.blocks.18.norm1.weight": "model-00008-of-00008.safetensors",
659
+ "model.visual.blocks.18.norm2.bias": "model-00008-of-00008.safetensors",
660
+ "model.visual.blocks.18.norm2.weight": "model-00008-of-00008.safetensors",
661
+ "model.visual.blocks.19.attn.proj.bias": "model-00008-of-00008.safetensors",
662
+ "model.visual.blocks.19.attn.proj.weight": "model-00008-of-00008.safetensors",
663
+ "model.visual.blocks.19.attn.qkv.bias": "model-00008-of-00008.safetensors",
664
+ "model.visual.blocks.19.attn.qkv.weight": "model-00008-of-00008.safetensors",
665
+ "model.visual.blocks.19.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
666
+ "model.visual.blocks.19.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
667
+ "model.visual.blocks.19.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
668
+ "model.visual.blocks.19.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
669
+ "model.visual.blocks.19.norm1.bias": "model-00008-of-00008.safetensors",
670
+ "model.visual.blocks.19.norm1.weight": "model-00008-of-00008.safetensors",
671
+ "model.visual.blocks.19.norm2.bias": "model-00008-of-00008.safetensors",
672
+ "model.visual.blocks.19.norm2.weight": "model-00008-of-00008.safetensors",
673
+ "model.visual.blocks.2.attn.proj.bias": "model-00008-of-00008.safetensors",
674
+ "model.visual.blocks.2.attn.proj.weight": "model-00008-of-00008.safetensors",
675
+ "model.visual.blocks.2.attn.qkv.bias": "model-00008-of-00008.safetensors",
676
+ "model.visual.blocks.2.attn.qkv.weight": "model-00008-of-00008.safetensors",
677
+ "model.visual.blocks.2.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
678
+ "model.visual.blocks.2.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
679
+ "model.visual.blocks.2.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
680
+ "model.visual.blocks.2.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
681
+ "model.visual.blocks.2.norm1.bias": "model-00008-of-00008.safetensors",
682
+ "model.visual.blocks.2.norm1.weight": "model-00008-of-00008.safetensors",
683
+ "model.visual.blocks.2.norm2.bias": "model-00008-of-00008.safetensors",
684
+ "model.visual.blocks.2.norm2.weight": "model-00008-of-00008.safetensors",
685
+ "model.visual.blocks.20.attn.proj.bias": "model-00008-of-00008.safetensors",
686
+ "model.visual.blocks.20.attn.proj.weight": "model-00008-of-00008.safetensors",
687
+ "model.visual.blocks.20.attn.qkv.bias": "model-00008-of-00008.safetensors",
688
+ "model.visual.blocks.20.attn.qkv.weight": "model-00008-of-00008.safetensors",
689
+ "model.visual.blocks.20.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
690
+ "model.visual.blocks.20.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
691
+ "model.visual.blocks.20.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
692
+ "model.visual.blocks.20.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
693
+ "model.visual.blocks.20.norm1.bias": "model-00008-of-00008.safetensors",
694
+ "model.visual.blocks.20.norm1.weight": "model-00008-of-00008.safetensors",
695
+ "model.visual.blocks.20.norm2.bias": "model-00008-of-00008.safetensors",
696
+ "model.visual.blocks.20.norm2.weight": "model-00008-of-00008.safetensors",
697
+ "model.visual.blocks.21.attn.proj.bias": "model-00008-of-00008.safetensors",
698
+ "model.visual.blocks.21.attn.proj.weight": "model-00008-of-00008.safetensors",
699
+ "model.visual.blocks.21.attn.qkv.bias": "model-00008-of-00008.safetensors",
700
+ "model.visual.blocks.21.attn.qkv.weight": "model-00008-of-00008.safetensors",
701
+ "model.visual.blocks.21.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
702
+ "model.visual.blocks.21.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
703
+ "model.visual.blocks.21.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
704
+ "model.visual.blocks.21.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
705
+ "model.visual.blocks.21.norm1.bias": "model-00008-of-00008.safetensors",
706
+ "model.visual.blocks.21.norm1.weight": "model-00008-of-00008.safetensors",
707
+ "model.visual.blocks.21.norm2.bias": "model-00008-of-00008.safetensors",
708
+ "model.visual.blocks.21.norm2.weight": "model-00008-of-00008.safetensors",
709
+ "model.visual.blocks.22.attn.proj.bias": "model-00008-of-00008.safetensors",
710
+ "model.visual.blocks.22.attn.proj.weight": "model-00008-of-00008.safetensors",
711
+ "model.visual.blocks.22.attn.qkv.bias": "model-00008-of-00008.safetensors",
712
+ "model.visual.blocks.22.attn.qkv.weight": "model-00008-of-00008.safetensors",
713
+ "model.visual.blocks.22.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
714
+ "model.visual.blocks.22.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
715
+ "model.visual.blocks.22.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
716
+ "model.visual.blocks.22.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
717
+ "model.visual.blocks.22.norm1.bias": "model-00008-of-00008.safetensors",
718
+ "model.visual.blocks.22.norm1.weight": "model-00008-of-00008.safetensors",
719
+ "model.visual.blocks.22.norm2.bias": "model-00008-of-00008.safetensors",
720
+ "model.visual.blocks.22.norm2.weight": "model-00008-of-00008.safetensors",
721
+ "model.visual.blocks.23.attn.proj.bias": "model-00008-of-00008.safetensors",
722
+ "model.visual.blocks.23.attn.proj.weight": "model-00008-of-00008.safetensors",
723
+ "model.visual.blocks.23.attn.qkv.bias": "model-00008-of-00008.safetensors",
724
+ "model.visual.blocks.23.attn.qkv.weight": "model-00008-of-00008.safetensors",
725
+ "model.visual.blocks.23.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
726
+ "model.visual.blocks.23.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
727
+ "model.visual.blocks.23.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
728
+ "model.visual.blocks.23.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
729
+ "model.visual.blocks.23.norm1.bias": "model-00008-of-00008.safetensors",
730
+ "model.visual.blocks.23.norm1.weight": "model-00008-of-00008.safetensors",
731
+ "model.visual.blocks.23.norm2.bias": "model-00008-of-00008.safetensors",
732
+ "model.visual.blocks.23.norm2.weight": "model-00008-of-00008.safetensors",
733
+ "model.visual.blocks.24.attn.proj.bias": "model-00008-of-00008.safetensors",
734
+ "model.visual.blocks.24.attn.proj.weight": "model-00008-of-00008.safetensors",
735
+ "model.visual.blocks.24.attn.qkv.bias": "model-00008-of-00008.safetensors",
736
+ "model.visual.blocks.24.attn.qkv.weight": "model-00008-of-00008.safetensors",
737
+ "model.visual.blocks.24.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
738
+ "model.visual.blocks.24.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
739
+ "model.visual.blocks.24.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
740
+ "model.visual.blocks.24.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
741
+ "model.visual.blocks.24.norm1.bias": "model-00008-of-00008.safetensors",
742
+ "model.visual.blocks.24.norm1.weight": "model-00008-of-00008.safetensors",
743
+ "model.visual.blocks.24.norm2.bias": "model-00008-of-00008.safetensors",
744
+ "model.visual.blocks.24.norm2.weight": "model-00008-of-00008.safetensors",
745
+ "model.visual.blocks.25.attn.proj.bias": "model-00008-of-00008.safetensors",
746
+ "model.visual.blocks.25.attn.proj.weight": "model-00008-of-00008.safetensors",
747
+ "model.visual.blocks.25.attn.qkv.bias": "model-00008-of-00008.safetensors",
748
+ "model.visual.blocks.25.attn.qkv.weight": "model-00008-of-00008.safetensors",
749
+ "model.visual.blocks.25.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
750
+ "model.visual.blocks.25.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
751
+ "model.visual.blocks.25.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
752
+ "model.visual.blocks.25.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
753
+ "model.visual.blocks.25.norm1.bias": "model-00008-of-00008.safetensors",
754
+ "model.visual.blocks.25.norm1.weight": "model-00008-of-00008.safetensors",
755
+ "model.visual.blocks.25.norm2.bias": "model-00008-of-00008.safetensors",
756
+ "model.visual.blocks.25.norm2.weight": "model-00008-of-00008.safetensors",
757
+ "model.visual.blocks.26.attn.proj.bias": "model-00008-of-00008.safetensors",
758
+ "model.visual.blocks.26.attn.proj.weight": "model-00008-of-00008.safetensors",
759
+ "model.visual.blocks.26.attn.qkv.bias": "model-00008-of-00008.safetensors",
760
+ "model.visual.blocks.26.attn.qkv.weight": "model-00008-of-00008.safetensors",
761
+ "model.visual.blocks.26.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
762
+ "model.visual.blocks.26.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
763
+ "model.visual.blocks.26.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
764
+ "model.visual.blocks.26.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
765
+ "model.visual.blocks.26.norm1.bias": "model-00008-of-00008.safetensors",
766
+ "model.visual.blocks.26.norm1.weight": "model-00008-of-00008.safetensors",
767
+ "model.visual.blocks.26.norm2.bias": "model-00008-of-00008.safetensors",
768
+ "model.visual.blocks.26.norm2.weight": "model-00008-of-00008.safetensors",
769
+ "model.visual.blocks.3.attn.proj.bias": "model-00008-of-00008.safetensors",
770
+ "model.visual.blocks.3.attn.proj.weight": "model-00008-of-00008.safetensors",
771
+ "model.visual.blocks.3.attn.qkv.bias": "model-00008-of-00008.safetensors",
772
+ "model.visual.blocks.3.attn.qkv.weight": "model-00008-of-00008.safetensors",
773
+ "model.visual.blocks.3.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
774
+ "model.visual.blocks.3.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
775
+ "model.visual.blocks.3.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
776
+ "model.visual.blocks.3.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
777
+ "model.visual.blocks.3.norm1.bias": "model-00008-of-00008.safetensors",
778
+ "model.visual.blocks.3.norm1.weight": "model-00008-of-00008.safetensors",
779
+ "model.visual.blocks.3.norm2.bias": "model-00008-of-00008.safetensors",
780
+ "model.visual.blocks.3.norm2.weight": "model-00008-of-00008.safetensors",
781
+ "model.visual.blocks.4.attn.proj.bias": "model-00008-of-00008.safetensors",
782
+ "model.visual.blocks.4.attn.proj.weight": "model-00008-of-00008.safetensors",
783
+ "model.visual.blocks.4.attn.qkv.bias": "model-00008-of-00008.safetensors",
784
+ "model.visual.blocks.4.attn.qkv.weight": "model-00008-of-00008.safetensors",
785
+ "model.visual.blocks.4.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
786
+ "model.visual.blocks.4.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
787
+ "model.visual.blocks.4.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
788
+ "model.visual.blocks.4.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
789
+ "model.visual.blocks.4.norm1.bias": "model-00008-of-00008.safetensors",
790
+ "model.visual.blocks.4.norm1.weight": "model-00008-of-00008.safetensors",
791
+ "model.visual.blocks.4.norm2.bias": "model-00008-of-00008.safetensors",
792
+ "model.visual.blocks.4.norm2.weight": "model-00008-of-00008.safetensors",
793
+ "model.visual.blocks.5.attn.proj.bias": "model-00008-of-00008.safetensors",
794
+ "model.visual.blocks.5.attn.proj.weight": "model-00008-of-00008.safetensors",
795
+ "model.visual.blocks.5.attn.qkv.bias": "model-00008-of-00008.safetensors",
796
+ "model.visual.blocks.5.attn.qkv.weight": "model-00008-of-00008.safetensors",
797
+ "model.visual.blocks.5.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
798
+ "model.visual.blocks.5.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
799
+ "model.visual.blocks.5.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
800
+ "model.visual.blocks.5.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
801
+ "model.visual.blocks.5.norm1.bias": "model-00008-of-00008.safetensors",
802
+ "model.visual.blocks.5.norm1.weight": "model-00008-of-00008.safetensors",
803
+ "model.visual.blocks.5.norm2.bias": "model-00008-of-00008.safetensors",
804
+ "model.visual.blocks.5.norm2.weight": "model-00008-of-00008.safetensors",
805
+ "model.visual.blocks.6.attn.proj.bias": "model-00008-of-00008.safetensors",
806
+ "model.visual.blocks.6.attn.proj.weight": "model-00008-of-00008.safetensors",
807
+ "model.visual.blocks.6.attn.qkv.bias": "model-00008-of-00008.safetensors",
808
+ "model.visual.blocks.6.attn.qkv.weight": "model-00008-of-00008.safetensors",
809
+ "model.visual.blocks.6.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
810
+ "model.visual.blocks.6.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
811
+ "model.visual.blocks.6.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
812
+ "model.visual.blocks.6.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
813
+ "model.visual.blocks.6.norm1.bias": "model-00008-of-00008.safetensors",
814
+ "model.visual.blocks.6.norm1.weight": "model-00008-of-00008.safetensors",
815
+ "model.visual.blocks.6.norm2.bias": "model-00008-of-00008.safetensors",
816
+ "model.visual.blocks.6.norm2.weight": "model-00008-of-00008.safetensors",
817
+ "model.visual.blocks.7.attn.proj.bias": "model-00008-of-00008.safetensors",
818
+ "model.visual.blocks.7.attn.proj.weight": "model-00008-of-00008.safetensors",
819
+ "model.visual.blocks.7.attn.qkv.bias": "model-00008-of-00008.safetensors",
820
+ "model.visual.blocks.7.attn.qkv.weight": "model-00008-of-00008.safetensors",
821
+ "model.visual.blocks.7.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
822
+ "model.visual.blocks.7.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
823
+ "model.visual.blocks.7.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
824
+ "model.visual.blocks.7.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
825
+ "model.visual.blocks.7.norm1.bias": "model-00008-of-00008.safetensors",
826
+ "model.visual.blocks.7.norm1.weight": "model-00008-of-00008.safetensors",
827
+ "model.visual.blocks.7.norm2.bias": "model-00008-of-00008.safetensors",
828
+ "model.visual.blocks.7.norm2.weight": "model-00008-of-00008.safetensors",
829
+ "model.visual.blocks.8.attn.proj.bias": "model-00008-of-00008.safetensors",
830
+ "model.visual.blocks.8.attn.proj.weight": "model-00008-of-00008.safetensors",
831
+ "model.visual.blocks.8.attn.qkv.bias": "model-00008-of-00008.safetensors",
832
+ "model.visual.blocks.8.attn.qkv.weight": "model-00008-of-00008.safetensors",
833
+ "model.visual.blocks.8.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
834
+ "model.visual.blocks.8.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
835
+ "model.visual.blocks.8.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
836
+ "model.visual.blocks.8.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
837
+ "model.visual.blocks.8.norm1.bias": "model-00008-of-00008.safetensors",
838
+ "model.visual.blocks.8.norm1.weight": "model-00008-of-00008.safetensors",
839
+ "model.visual.blocks.8.norm2.bias": "model-00008-of-00008.safetensors",
840
+ "model.visual.blocks.8.norm2.weight": "model-00008-of-00008.safetensors",
841
+ "model.visual.blocks.9.attn.proj.bias": "model-00008-of-00008.safetensors",
842
+ "model.visual.blocks.9.attn.proj.weight": "model-00008-of-00008.safetensors",
843
+ "model.visual.blocks.9.attn.qkv.bias": "model-00008-of-00008.safetensors",
844
+ "model.visual.blocks.9.attn.qkv.weight": "model-00008-of-00008.safetensors",
845
+ "model.visual.blocks.9.mlp.linear_fc1.bias": "model-00008-of-00008.safetensors",
846
+ "model.visual.blocks.9.mlp.linear_fc1.weight": "model-00008-of-00008.safetensors",
847
+ "model.visual.blocks.9.mlp.linear_fc2.bias": "model-00008-of-00008.safetensors",
848
+ "model.visual.blocks.9.mlp.linear_fc2.weight": "model-00008-of-00008.safetensors",
849
+ "model.visual.blocks.9.norm1.bias": "model-00008-of-00008.safetensors",
850
+ "model.visual.blocks.9.norm1.weight": "model-00008-of-00008.safetensors",
851
+ "model.visual.blocks.9.norm2.bias": "model-00008-of-00008.safetensors",
852
+ "model.visual.blocks.9.norm2.weight": "model-00008-of-00008.safetensors",
853
+ "model.visual.merger.linear_fc1.bias": "model-00008-of-00008.safetensors",
854
+ "model.visual.merger.linear_fc1.weight": "model-00008-of-00008.safetensors",
855
+ "model.visual.merger.linear_fc2.bias": "model-00008-of-00008.safetensors",
856
+ "model.visual.merger.linear_fc2.weight": "model-00008-of-00008.safetensors",
857
+ "model.visual.merger.norm.bias": "model-00008-of-00008.safetensors",
858
+ "model.visual.merger.norm.weight": "model-00008-of-00008.safetensors",
859
+ "model.visual.patch_embed.proj.bias": "model-00008-of-00008.safetensors",
860
+ "model.visual.patch_embed.proj.weight": "model-00008-of-00008.safetensors",
861
+ "model.visual.pos_embed.weight": "model-00008-of-00008.safetensors",
862
+ "mtp.layers.0.self_attn.k_norm.weight": "mtp.safetensors",
863
+ "mtp.layers.0.self_attn.q_proj.weight": "mtp.safetensors",
864
+ "mtp.norm.weight": "mtp.safetensors",
865
+ "mtp.pre_fc_norm_embedding.weight": "mtp.safetensors",
866
+ "mtp.layers.0.mlp.down_proj.weight": "mtp.safetensors",
867
+ "mtp.layers.0.self_attn.o_proj.weight": "mtp.safetensors",
868
+ "mtp.layers.0.post_attention_layernorm.weight": "mtp.safetensors",
869
+ "mtp.layers.0.self_attn.k_proj.weight": "mtp.safetensors",
870
+ "mtp.layers.0.mlp.up_proj.weight": "mtp.safetensors",
871
+ "mtp.layers.0.input_layernorm.weight": "mtp.safetensors",
872
+ "mtp.layers.0.self_attn.q_norm.weight": "mtp.safetensors",
873
+ "mtp.pre_fc_norm_hidden.weight": "mtp.safetensors",
874
+ "mtp.fc.weight": "mtp.safetensors",
875
+ "mtp.layers.0.self_attn.v_proj.weight": "mtp.safetensors",
876
+ "mtp.layers.0.mlp.gate_proj.weight": "mtp.safetensors"
877
+ }
878
+ }
modeling_sensenova_loop.py ADDED
@@ -0,0 +1,342 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Hugging Face remote-code support for shared-weight SenseNova Loop and MHC.
2
+
3
+ This single file contains configuration, plan validation, MHC routing and the
4
+ model implementation. The original physical layer weights are reused on every
5
+ Loop visit; autoregressive cache states are independent per visit.
6
+ """
7
+ import copy
8
+ import math
9
+
10
+ import torch
11
+ from torch import nn
12
+ from transformers import Qwen3_5Config
13
+ from transformers.cache_utils import DynamicCache
14
+ from transformers.models.qwen3_5 import modeling_qwen3_5 as hf
15
+
16
+
17
+ class SenseNovaLoopConfig(Qwen3_5Config):
18
+ model_type = "sensenova_loop"
19
+
20
+
21
+ def execution_plan(config):
22
+ """Return physical layer indices in the order they execute during inference."""
23
+ if config.get("loop_expanded"):
24
+ raise ValueError("Use the canonical checkpoint, not an expanded serving checkpoint")
25
+ loop = config.get("loop_settings")
26
+ if not isinstance(loop, dict) or loop.get("use_gate", True):
27
+ raise ValueError("This implementation requires explicit, ungated loop_settings")
28
+
29
+ text_config = config["text_config"]
30
+ num_layers = text_config["num_hidden_layers"]
31
+ start, end, count = loop.get("start"), loop.get("end"), loop.get("eval_count")
32
+ if any(type(value) is not int for value in (num_layers, start, end, count)) or not (
33
+ 0 <= start < end <= num_layers and count >= 1
34
+ ):
35
+ raise ValueError("Invalid Loop interval or eval_count")
36
+ if loop.get("counts") != [count] or loop.get("probabilities") != [1.0]:
37
+ raise ValueError("Only fixed-count Loop checkpoints are supported")
38
+
39
+ layer_types = text_config["layer_types"]
40
+ if len(layer_types) != num_layers or set(layer_types) - {"full_attention", "linear_attention"}:
41
+ raise ValueError("Unexpected physical layer types")
42
+ if "full_attention" not in layer_types:
43
+ raise ValueError("At least one full-attention layer is required for sequence-length tracking")
44
+
45
+ mhc = config.get("mhc")
46
+ if mhc is not None:
47
+ mhc_start, mhc_end, streams, iterations = (
48
+ mhc.get(key) for key in ("start", "end", "streams", "iterations")
49
+ )
50
+ if (
51
+ any(type(value) is not int for value in (mhc_start, mhc_end, streams, iterations))
52
+ or not 0 <= mhc_start < mhc_end <= num_layers
53
+ or streams not in (2, 4)
54
+ or iterations < 1
55
+ ):
56
+ raise ValueError("Invalid MHC range, stream count, or iterations")
57
+ if type(mhc.get("fixed_identity")) is not bool:
58
+ raise ValueError("MHC fixed_identity must be a boolean")
59
+ scale = mhc.get("pre_scale", 1.0)
60
+ if not isinstance(scale, (int, float)) or not math.isfinite(scale) or scale <= 0:
61
+ raise ValueError("Invalid MHC pre_scale")
62
+ allowed = {"start", "end", "streams", "iterations", "fixed_identity", "pre_scale"}
63
+ if set(mhc) - allowed:
64
+ raise ValueError("Unknown MHC configuration fields")
65
+ phase_boundaries = (0, start, end, num_layers)
66
+ if not any(
67
+ left <= mhc_start < mhc_end <= right
68
+ for left, right in zip(phase_boundaries, phase_boundaries[1:])
69
+ ):
70
+ raise ValueError("MHC cannot cross a Loop phase boundary")
71
+
72
+ return tuple(range(start)) + tuple(range(start, end)) * count + tuple(range(end, num_layers))
73
+
74
+
75
+ def sinkhorn(logits, iterations):
76
+ """Normalize routing logits across rows and columns in log space."""
77
+ logits = logits.float()
78
+ for _ in range(iterations):
79
+ logits = logits - torch.logsumexp(logits, dim=-1, keepdim=True)
80
+ logits = logits - torch.logsumexp(logits, dim=-2, keepdim=True)
81
+ return logits.exp()
82
+
83
+
84
+ class MHCRouter(nn.Module):
85
+ def __init__(self, dim, settings):
86
+ super().__init__()
87
+ self.streams = settings["streams"]
88
+ self.iterations = settings["iterations"]
89
+ self.fixed_identity = settings["fixed_identity"]
90
+ self.pre_scale = settings.get("pre_scale", 1.0)
91
+ streams = self.streams
92
+ output_size = 2 * streams + (0 if self.fixed_identity else streams * streams)
93
+ self.weight = nn.Parameter(torch.zeros(output_size, streams * dim, dtype=torch.float32))
94
+ self.bias = nn.Parameter(torch.zeros(output_size, dtype=torch.float32))
95
+ with torch.no_grad():
96
+ initial_probabilities = {2: (0.25, 0.75), 4: (0.0625, 0.1875, 0.25, 0.5)}[streams]
97
+ self.bias[:streams].copy_(torch.logit(torch.tensor(initial_probabilities, device=self.bias.device)))
98
+ if not self.fixed_identity:
99
+ diagonal = math.log((1 - (streams - 1) / 16) * 16)
100
+ self.bias[2 * streams:].copy_((torch.eye(streams, device=self.bias.device) * diagonal).flatten())
101
+
102
+ def _apply(self, fn, recurse=True):
103
+ # Keep the original FP32 values when the surrounding model is cast to BF16.
104
+ def keep_fp32(tensor):
105
+ result = fn(tensor)
106
+ if tensor.is_floating_point() and result.dtype != torch.float32:
107
+ if tensor.device.type == "meta" and result.device.type != "meta":
108
+ return torch.empty_like(result, dtype=torch.float32)
109
+ return tensor.to(device=result.device, dtype=torch.float32)
110
+ return result
111
+
112
+ return super()._apply(keep_fp32, recurse=recurse)
113
+
114
+ def forward(self, residual):
115
+ if self.weight.dtype != torch.float32 or self.bias.dtype != torch.float32:
116
+ raise RuntimeError("mHC routers require FP32 parameters; configure FSDP per-module precision")
117
+ with torch.autocast(device_type=residual.device.type, enabled=False):
118
+ flat = residual.flatten(-2).float()
119
+ normalized = flat * torch.rsqrt(flat.square().mean(-1, keepdim=True) + 1e-6)
120
+ logits = torch.nn.functional.linear(normalized, self.weight, self.bias)
121
+ streams = self.streams
122
+ pre = logits[..., :streams].sigmoid()
123
+ post = 2 * logits[..., streams:2 * streams].sigmoid()
124
+ if self.fixed_identity:
125
+ mix = torch.eye(streams, device=flat.device, dtype=torch.float32).expand(
126
+ *flat.shape[:-1], streams, streams
127
+ )
128
+ else:
129
+ mix = sinkhorn(logits[..., 2 * streams:].unflatten(-1, (streams, streams)), self.iterations)
130
+ branch = torch.matmul(pre.to(residual.dtype).unsqueeze(-2), residual).squeeze(-2)
131
+ if self.pre_scale != 1.0:
132
+ branch = branch * self.pre_scale
133
+ return branch, post, mix
134
+
135
+ @staticmethod
136
+ def merge(residual, branch, post, mix):
137
+ """Mix residual streams and add the branch output to each stream."""
138
+ with torch.autocast(device_type=residual.device.type, enabled=False):
139
+ mixed_residual = torch.matmul(mix.to(residual.dtype), residual)
140
+ branch_update = post.to(branch.dtype).unsqueeze(-1) * branch.unsqueeze(-2)
141
+ return mixed_residual + branch_update
142
+
143
+
144
+ def enter(hidden, physical, settings):
145
+ """Expand to residual streams at the first mHC layer, or restore their shape."""
146
+ if physical == settings["start"]:
147
+ return hidden.unsqueeze(-2).expand(
148
+ *hidden.shape[:-1], settings["streams"], hidden.shape[-1]
149
+ ).contiguous()
150
+ return hidden.unflatten(-1, (settings["streams"], -1))
151
+
152
+
153
+ def leave(residual, physical, settings):
154
+ """Reduce streams at the last mHC layer; otherwise carry them forward."""
155
+ return residual.mean(-2) if physical == settings["end"] - 1 else residual.flatten(-2)
156
+
157
+
158
+ class _CacheLayersView:
159
+ """Expose the execution slot when a layer accesses its physical index."""
160
+ def __init__(self, cache, physical, virtual):
161
+ self.cache = cache
162
+ self.physical = physical
163
+ self.virtual = virtual
164
+
165
+ def __getitem__(self, index):
166
+ if index != self.physical:
167
+ raise IndexError("A physical layer may only access its own virtual cache slot")
168
+ return self.cache.layers[self.virtual]
169
+
170
+
171
+ class _VisitCache:
172
+ """Map a physical attention module's layer_idx to one execution slot.
173
+
174
+ Mapping is local to this forward call: shared modules and their layer_idx are
175
+ never mutated. `layers` supports Transformers 5.12; state accessors support
176
+ newer official backbone cache APIs as well.
177
+ """
178
+ def __init__(self, cache, physical, virtual):
179
+ self.cache = cache
180
+ self.virtual = virtual
181
+ self.layers = _CacheLayersView(cache, physical, virtual)
182
+
183
+ def update(self, key_states, value_states, layer_idx, **kwargs):
184
+ return self.cache.update(key_states, value_states, self.virtual, **kwargs)
185
+
186
+ def has_previous_state(self, layer_idx=None, **kwargs):
187
+ return self.cache.has_previous_state(self.virtual, **kwargs)
188
+
189
+ def update_conv_state(self, state, layer_idx, **kwargs):
190
+ return self.cache.update_conv_state(state, self.virtual, **kwargs)
191
+
192
+ def update_recurrent_state(self, state, layer_idx, **kwargs):
193
+ return self.cache.update_recurrent_state(state, self.virtual, **kwargs)
194
+
195
+ def get_conv_state(self, layer_idx, **kwargs):
196
+ return self.cache.get_conv_state(self.virtual, **kwargs)
197
+
198
+ def get_recurrent_state(self, layer_idx, **kwargs):
199
+ return self.cache.get_recurrent_state(self.virtual, **kwargs)
200
+
201
+ def is_recording_past(self, layer_idx):
202
+ return self.cache.is_recording_past(self.virtual)
203
+
204
+
205
+ class SenseNovaLoopTextModel(hf.Qwen3_5TextModel):
206
+ def forward(self, input_ids=None, attention_mask=None, position_ids=None,
207
+ past_key_values=None, inputs_embeds=None, use_cache=None, **kwargs):
208
+ if (input_ids is None) == (inputs_embeds is None):
209
+ raise ValueError("Specify exactly one of input_ids or inputs_embeds")
210
+ if inputs_embeds is None:
211
+ inputs_embeds = self.embed_tokens(input_ids)
212
+ use_cache = self.config.use_cache if use_cache is None else use_cache
213
+ if self.gradient_checkpointing and self.training and use_cache:
214
+ raise ValueError("Disable use_cache when training with gradient checkpointing")
215
+ if not use_cache and past_key_values is not None:
216
+ raise ValueError("past_key_values requires use_cache=True")
217
+ if past_key_values is not None and not isinstance(past_key_values, DynamicCache):
218
+ raise TypeError("Loop generation currently supports DynamicCache only")
219
+ if use_cache:
220
+ if past_key_values is None:
221
+ past_key_values = DynamicCache(config=self._loop_cache_config)
222
+ past_key_values._loop_execution_plan = self._loop_plan
223
+ elif getattr(past_key_values, "_loop_execution_plan", None) != self._loop_plan:
224
+ # GenerationMixin may allocate an empty physical-depth cache.
225
+ # A populated foreign cache cannot be reinterpreted safely.
226
+ if past_key_values.get_seq_length() != 0 or past_key_values.has_previous_state():
227
+ raise ValueError("A populated cache from another execution plan cannot be reused")
228
+ past_key_values = DynamicCache(config=self._loop_cache_config)
229
+ past_key_values._loop_execution_plan = self._loop_plan
230
+
231
+ if position_ids is None:
232
+ seen = past_key_values.get_seq_length() if past_key_values is not None else 0
233
+ position_ids = torch.arange(inputs_embeds.shape[1], device=inputs_embeds.device) + seen
234
+ position_ids = position_ids.view(1, 1, -1).expand(4, inputs_embeds.shape[0], -1)
235
+ elif position_ids.ndim == 2:
236
+ position_ids = position_ids.unsqueeze(0).expand(4, -1, -1)
237
+ if position_ids.ndim == 3 and position_ids.shape[0] == 4:
238
+ text_positions, position_ids = position_ids[0], position_ids[1:]
239
+ else:
240
+ text_positions = None
241
+
242
+ if isinstance(attention_mask, dict):
243
+ masks = attention_mask
244
+ else:
245
+ mask_args = dict(
246
+ config=self._loop_cache_config,
247
+ inputs_embeds=inputs_embeds,
248
+ attention_mask=attention_mask,
249
+ past_key_values=past_key_values,
250
+ position_ids=text_positions,
251
+ )
252
+ if hasattr(hf, "create_recurrent_attention_mask"):
253
+ linear_mask = hf.create_recurrent_attention_mask(**mask_args)
254
+ else:
255
+ linear_mask = self._update_linear_attn_mask(attention_mask, past_key_values)
256
+ masks = {
257
+ "full_attention": hf.create_causal_mask(**mask_args),
258
+ "linear_attention": linear_mask,
259
+ }
260
+
261
+ hidden = inputs_embeds
262
+ position_embeddings = self.rotary_emb(hidden, position_ids)
263
+ # Repeated visits share weights, but each visit has its own cache slot.
264
+ for execution_index, physical_index in enumerate(self._loop_plan):
265
+ cache = (
266
+ _VisitCache(past_key_values, physical_index, execution_index)
267
+ if use_cache else None
268
+ )
269
+ hidden = self.layers[physical_index](
270
+ hidden, position_embeddings=position_embeddings,
271
+ attention_mask=masks[self.config.layer_types[physical_index]],
272
+ position_ids=text_positions, past_key_values=cache,
273
+ use_cache=use_cache, **kwargs,
274
+ )
275
+ return hf.Qwen3_5ModelOutputWithPast(
276
+ last_hidden_state=self.norm(hidden), past_key_values=past_key_values)
277
+
278
+
279
+ class MHCDecoderLayer(hf.Qwen3_5DecoderLayer):
280
+ def forward(self, hidden_states, position_embeddings, attention_mask=None,
281
+ position_ids=None, past_key_values=None, **kwargs):
282
+ residual = enter(hidden_states, self._mhc_physical, self._mhc_settings)
283
+ branch, post, mix = self.mhc_attn(residual)
284
+ branch = self.input_layernorm(branch)
285
+ kind = self.layer_type if hasattr(self, "layer_type") else self.block_type
286
+ if kind == "linear_attention":
287
+ branch = self.linear_attn(hidden_states=branch, cache_params=past_key_values,
288
+ attention_mask=attention_mask, **kwargs)
289
+ else:
290
+ branch, _ = self.self_attn(
291
+ hidden_states=branch, position_embeddings=position_embeddings,
292
+ attention_mask=attention_mask, position_ids=position_ids,
293
+ past_key_values=past_key_values, **kwargs)
294
+ residual = MHCRouter.merge(residual, branch, post, mix)
295
+ branch, post, mix = self.mhc_ffn(residual)
296
+ branch = self.mlp(self.post_attention_layernorm(branch))
297
+ return leave(MHCRouter.merge(residual, branch, post, mix),
298
+ self._mhc_physical, self._mhc_settings)
299
+
300
+
301
+ def _configure_text_model(text, config):
302
+ raw = config.to_dict()
303
+ plan = execution_plan(raw)
304
+ # Change only these new model instances, preserving canonical state_dict
305
+ # names and avoiding a second allocation of the full backbone.
306
+ text.__class__ = SenseNovaLoopTextModel
307
+ text._loop_plan = plan
308
+ text._loop_cache_config = copy.deepcopy(text.config)
309
+ text._loop_cache_config.num_hidden_layers = len(plan)
310
+ text._loop_cache_config.layer_types = [text.config.layer_types[i] for i in plan]
311
+ if raw.get("mhc") is not None:
312
+ settings = raw["mhc"]
313
+ for i in range(settings["start"], settings["end"]):
314
+ layer = text.layers[i]
315
+ layer.__class__ = MHCDecoderLayer
316
+ layer._mhc_physical, layer._mhc_settings = i, dict(settings)
317
+ layer.mhc_attn = MHCRouter(text.config.hidden_size, settings)
318
+ layer.mhc_ffn = MHCRouter(text.config.hidden_size, settings)
319
+
320
+
321
+ class SenseNovaLoopModel(hf.Qwen3_5Model):
322
+ config_class = SenseNovaLoopConfig
323
+ _keep_in_fp32_modules = ["mhc_attn", "mhc_ffn"]
324
+ _no_split_modules = ["Qwen3_5DecoderLayer", "MHCDecoderLayer", "Qwen3_5VisionBlock"]
325
+ _keys_to_ignore_on_load_unexpected = [r"model\.language_model\.loop_gate\.logits", r"^mtp\."]
326
+
327
+ def __init__(self, config):
328
+ execution_plan(config.to_dict())
329
+ super().__init__(config)
330
+ _configure_text_model(self.language_model, config)
331
+
332
+
333
+ class SenseNovaLoopForConditionalGeneration(hf.Qwen3_5ForConditionalGeneration):
334
+ config_class = SenseNovaLoopConfig
335
+ _keep_in_fp32_modules = ["mhc_attn", "mhc_ffn"]
336
+ _no_split_modules = ["Qwen3_5DecoderLayer", "MHCDecoderLayer", "Qwen3_5VisionBlock"]
337
+ _keys_to_ignore_on_load_unexpected = [r"model\.language_model\.loop_gate\.logits", r"^mtp\."]
338
+
339
+ def __init__(self, config):
340
+ execution_plan(config.to_dict())
341
+ super().__init__(config)
342
+ _configure_text_model(self.model.language_model, config)
mtp.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fc870cc5953e18abce3664f29bd47e174656fcd3dbf50226998a64f79c83a178
3
+ size 486582880
preprocessor_config.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_convert_rgb": true,
3
+ "do_normalize": true,
4
+ "do_rescale": true,
5
+ "do_resize": true,
6
+ "image_mean": [
7
+ 0.5,
8
+ 0.5,
9
+ 0.5
10
+ ],
11
+ "image_processor_type": "Qwen2VLImageProcessor",
12
+ "image_std": [
13
+ 0.5,
14
+ 0.5,
15
+ 0.5
16
+ ],
17
+ "merge_size": 2,
18
+ "patch_size": 16,
19
+ "resample": 3,
20
+ "rescale_factor": 0.00392156862745098,
21
+ "size": {
22
+ "longest_edge": 16777216,
23
+ "shortest_edge": 65536
24
+ },
25
+ "temporal_patch_size": 2
26
+ }
processor_config.json ADDED
@@ -0,0 +1,60 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "image_processor": {
3
+ "do_convert_rgb": true,
4
+ "do_normalize": true,
5
+ "do_rescale": true,
6
+ "do_resize": true,
7
+ "image_mean": [
8
+ 0.5,
9
+ 0.5,
10
+ 0.5
11
+ ],
12
+ "image_processor_type": "Qwen2VLImageProcessor",
13
+ "image_std": [
14
+ 0.5,
15
+ 0.5,
16
+ 0.5
17
+ ],
18
+ "merge_size": 2,
19
+ "patch_size": 16,
20
+ "resample": 3,
21
+ "rescale_factor": 0.00392156862745098,
22
+ "size": {
23
+ "longest_edge": 16777216,
24
+ "shortest_edge": 65536
25
+ },
26
+ "temporal_patch_size": 2
27
+ },
28
+ "processor_class": "Qwen3VLProcessor",
29
+ "video_processor": {
30
+ "do_convert_rgb": true,
31
+ "do_normalize": true,
32
+ "do_rescale": true,
33
+ "do_resize": true,
34
+ "do_sample_frames": true,
35
+ "fps": 2,
36
+ "image_mean": [
37
+ 0.5,
38
+ 0.5,
39
+ 0.5
40
+ ],
41
+ "image_std": [
42
+ 0.5,
43
+ 0.5,
44
+ 0.5
45
+ ],
46
+ "max_frames": 768,
47
+ "merge_size": 2,
48
+ "min_frames": 4,
49
+ "patch_size": 16,
50
+ "resample": 3,
51
+ "rescale_factor": 0.00392156862745098,
52
+ "return_metadata": false,
53
+ "size": {
54
+ "longest_edge": 25165824,
55
+ "shortest_edge": 4096
56
+ },
57
+ "temporal_patch_size": 2,
58
+ "video_processor_type": "Qwen3VLVideoProcessor"
59
+ }
60
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523
3
+ size 19989325
tokenizer_config.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "audio_bos_token": "<|audio_start|>",
4
+ "audio_eos_token": "<|audio_end|>",
5
+ "audio_token": "<|audio_pad|>",
6
+ "backend": "tokenizers",
7
+ "bos_token": null,
8
+ "clean_up_tokenization_spaces": false,
9
+ "eos_token": "<|im_end|>",
10
+ "errors": "replace",
11
+ "image_token": "<|image_pad|>",
12
+ "is_local": true,
13
+ "local_files_only": false,
14
+ "model_max_length": 262144,
15
+ "model_specific_special_tokens": {
16
+ "audio_bos_token": "<|audio_start|>",
17
+ "audio_eos_token": "<|audio_end|>",
18
+ "audio_token": "<|audio_pad|>",
19
+ "image_token": "<|image_pad|>",
20
+ "video_token": "<|video_pad|>",
21
+ "vision_bos_token": "<|vision_start|>",
22
+ "vision_eos_token": "<|vision_end|>"
23
+ },
24
+ "pad_token": "<|endoftext|>",
25
+ "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
26
+ "processor_class": "Qwen3VLProcessor",
27
+ "split_special_tokens": false,
28
+ "tokenizer_class": "Qwen2Tokenizer",
29
+ "unk_token": null,
30
+ "video_token": "<|video_pad|>",
31
+ "vision_bos_token": "<|vision_start|>",
32
+ "vision_eos_token": "<|vision_end|>"
33
+ }
video_preprocessor_config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_convert_rgb": true,
3
+ "do_normalize": true,
4
+ "do_rescale": true,
5
+ "do_resize": true,
6
+ "do_sample_frames": true,
7
+ "fps": 2,
8
+ "image_mean": [
9
+ 0.5,
10
+ 0.5,
11
+ 0.5
12
+ ],
13
+ "image_std": [
14
+ 0.5,
15
+ 0.5,
16
+ 0.5
17
+ ],
18
+ "max_frames": 768,
19
+ "merge_size": 2,
20
+ "min_frames": 4,
21
+ "patch_size": 16,
22
+ "resample": 3,
23
+ "rescale_factor": 0.00392156862745098,
24
+ "return_metadata": false,
25
+ "size": {
26
+ "longest_edge": 25165824,
27
+ "shortest_edge": 4096
28
+ },
29
+ "temporal_patch_size": 2,
30
+ "video_processor_type": "Qwen3VLVideoProcessor"
31
+ }