TheRobotSimGuy commited on
Commit
e11b59e
·
verified ·
1 Parent(s): 92b84f1

Mirror of nvidia/Cosmos-Reason2-2B@9ce19a19 (NVIDIA Open Model License) + NOTICE

Browse files
.gitattributes CHANGED
@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ cosmos_cr2_8b_paiBench.png filter=lfs diff=lfs merge=lfs -text
37
+ cosmos_cr2_2b_paiBench.png filter=lfs diff=lfs merge=lfs -text
38
+ PAI_Bench_Leaderboard.png filter=lfs diff=lfs merge=lfs -text
CR2 Performance by Category.png ADDED
LICENSE_NVIDIA_OPEN_MODEL.md ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!-- Source: https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/ (as shipped in nvidia/Cosmos-Reason2-2B's model card) -->
2
+
3
+ # NVIDIA Open Model License Agreement
4
+ Version Release Date: September 23, 2025
5
+ This NVIDIA Open Model License Agreement (the “Agreement”) is a legal agreement between the Legal Entity You represent, or if no entity is identified, You and NVIDIA Corporation and its Affiliates (“NVIDIA”) and governs Your use of the Models that NVIDIA provides to You under this Agreement. NVIDIA and You are each a “party” and collectively the “parties.”
6
+ NVIDIA models released under this Agreement are intended to be used permissively and enable the further development of AI technologies. Subject to the terms of this Agreement, NVIDIA confirms that:
7
+ - Models are commercially usable. - You are free to create and distribute Derivative Models. - NVIDIA does not claim ownership to any outputs generated using the Models or Model Derivatives.
8
+ By using, reproducing, modifying, distributing, performing or displaying any portion or element of the Model or Derivative Model, or otherwise accepting the terms of this Agreement, you agree to be bound by this Agreement.
9
+ ## 1. Definitions
10
+ 1.1. **Derivative Model** means all (a) modifications to the Model, (b) works based on the Model, and (c) any other derivative works of the Model. An output is not a Derivative Model.
11
+ 1.2. **Legal Entity** means the union of the acting entity and all other entities that control, are controlled by, or are under common control with that entity. For the purposes of this definition, “control” means (a) the power, direct or indirect, to cause the direction or management of such entity, whether by contract or otherwise, or (b) ownership of fifty percent (50%) or more of the outstanding shares, or (c) beneficial ownership of such entity.
12
+ 1.3. **Model** means the machine learning model, software, checkpoints, learnt weights, algorithms, parameters, configuration files and documentation shared under this Agreement.
13
+ 1.4. **NVIDIA Cosmos Model** means a multimodal Model shared under this Agreement.
14
+ 1.5. **Special-Purpose Model** means a Model that is only competent in a narrow set of purpose-specific tasks and should not be used for unintended or general-purpose applications.
15
+ 1.6. **You** or **Your** means an individual or Legal Entity exercising permissions granted by this Agreement.
16
+ ## 2. Conditions for Use, License Grant, AI Ethics and IP Ownership
17
+ ### 2.1. Conditions for Use - The Model and any Derivative Model are subject to additional terms as described in Section 2 and Section 3 of this Agreement. - If You institute copyright or patent litigation against any entity alleging that the Model or a Derivative Model constitutes infringement, then any licenses granted will terminate as of the date such litigation is filed. - If You bypass or disable any technical limitation, safety guardrail, encryption, DRM, or authentication mechanism contained in the Model without a substantially similar Guardrail, your rights will terminate. - NVIDIA may designate a Model as a Special-Purpose Model. - NVIDIA may update this Agreement to comply with legal and regulatory requirements.
18
+ ### 2.2. License Grant NVIDIA grants You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, revocable license to publicly perform, publicly display, reproduce, use, create derivative works of, make, have made, sell, offer for sale, distribute, and import the Model.
19
+ ### 2.3. AI Ethics Use of the Models must be consistent with NVIDIA’s [Trustworthy AI terms](https://www.nvidia.com/en-us/agreements/trustworthy-ai/terms/).
20
+ ### 2.4. IP Ownership - NVIDIA owns the Model and any Model Derivatives it creates. - You own your Model Derivatives. - NVIDIA claims no ownership rights in outputs. - Except as expressly granted, NVIDIA reserves all rights.
21
+ ## 3. Redistribution
22
+ You may reproduce and distribute copies of the Model or Derivative Models in any medium, with or without modifications, provided that:
23
+ - **3.1.** You must provide recipients with a copy of this Agreement and include this attribution in a “Notice” text file:
24
+ *“Licensed by NVIDIA Corporation under the NVIDIA Open Model License”*
25
+
26
+ - **3.2.** If distributing or making available a NVIDIA Cosmos Model, or products/services derived from it, you must include:
27
+ *“Built on NVIDIA Cosmos”*
28
+
29
+ - **3.3.** You may add your own copyright statements and license terms for your modifications, provided use still complies with this Agreement.
30
+ ## 4. Separate Components The Models may include components licensed under separate legal notices (e.g., Open Source Software Licenses). These terms apply, except where overridden by this Agreement unless required by third-party license terms.
31
+ ## 5. Trademarks No permission is granted to use NVIDIA’s trade names, trademarks, or product names, except for reasonable descriptive use.
32
+ ## 6. Disclaimer of Warranty The Model is provided **“AS IS”**, without warranties of any kind, including title, non-infringement, merchantability, or fitness for purpose. You assume risks associated with its use.
33
+ ## 7. Limitation of Liability NVIDIA is not liable for damages (direct, indirect, incidental, or consequential) arising from use of the Model, unless required by law.
34
+ ## 8. Indemnity You will indemnify and hold NVIDIA harmless against claims from third parties arising from your use or distribution of the Model, derivatives, or outputs.
35
+ ## 9. Feedback NVIDIA may use any feedback you provide without restriction or compensation.
36
+ ## 10. Governing Law This Agreement is governed by U.S. and Delaware law. Courts in Santa Clara County, California, have exclusive jurisdiction, except for urgent injunctive relief.
37
+ ## 11. Trade and Compliance You must comply with all export, import, trade, and sanctions laws, including U.S. Export Administration Regulations and OFAC rules.
NOTICE ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ Licensed by NVIDIA Corporation under the NVIDIA Open Model License.
2
+
3
+ This repository is an unmodified copy of nvidia/Cosmos-Reason2-2B
4
+ (https://huggingface.co/nvidia/Cosmos-Reason2-2B), redistributed by Simoji under Section 3 of the
5
+ NVIDIA Open Model License Agreement (copy in LICENSE_NVIDIA_OPEN_MODEL.md) so that the Simoji robot
6
+ brain (https://github.com/robotsimguy/simoji-brain) can load NVIDIA Isaac GR00T N1.7 without a
7
+ Hugging Face token. Built on NVIDIA Cosmos. Simoji is not affiliated with or endorsed by NVIDIA.
PAI_Bench_Leaderboard.png ADDED

Git LFS Details

  • SHA256: f9ac0320a57d6ee76a4f582388564dcb379ab170f2d99550e891fcd26a0b31c9
  • Pointer size: 131 Bytes
  • Size of remote file: 474 kB
README.md ADDED
@@ -0,0 +1,389 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ license_name: nvidia-open-model-license
4
+ license_link: >-
5
+ https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license
6
+ library_name: cosmos
7
+ tags:
8
+ - nvidia
9
+ - cosmos
10
+ - conversational
11
+ base_model:
12
+ - Qwen/Qwen3-VL-2B-Instruct
13
+ pipeline_tag: image-text-to-text
14
+ ---
15
+ > **Simoji mirror.** Unmodified copy of [nvidia/Cosmos-Reason2-2B](https://huggingface.co/nvidia/Cosmos-Reason2-2B) at revision `9ce19a195e423419c349abfc86fd07178b230561`, redistributed under the NVIDIA Open Model License (see `NOTICE` and `LICENSE_NVIDIA_OPEN_MODEL.md`). Built on NVIDIA Cosmos.
16
+
17
+ # **Cosmos-Reason2: Physical AI Common Sense and Embodied Reasoning Models**
18
+
19
+ [**Cosmos**](https://huggingface.co/collections/nvidia/cosmos-reason2) | [**Code**](https://github.com/nvidia-cosmos/cosmos-reason2)
20
+
21
+ # Model Overview
22
+
23
+ ## Description:
24
+
25
+ NVIDIA Cosmos Reason 2 is an open, customizable, 2B-parameter reasoning vision language model (VLM) for physical AI and robotics that enables robots and vision AI agents to reason like humans, using prior knowledge, physics understanding and common sense to understand and act in the real world. This model understands space, time, and fundamental physics, and can serve as a planning model to reason what steps an embodied agent might take next.
26
+
27
+ New features with Cosmos Reason 2:
28
+
29
+ * Enhanced physical AI reasoning with improved spatio-temporal understanding and timestamp precision.
30
+ * Supports object detection with 2D/3D point localization and bounding box coordinates with reasoning explanations and labels.
31
+ * Improved long-context understanding up to 256K input tokens.
32
+
33
+ Use cases:
34
+
35
+ * Video analytics AI agents — Extract valuable insights and perform root-cause analysis on massive volumes of video data. These agents can be used to analyze and understand recorded or live video streams across city and industrial operations. Jumpstart your development of video analytics AI agents by using the [NVIDIA Blueprint for video search and summarization (VSS)](https://build.nvidia.com/nvidia/video-search-and-summarization) with Cosmos Reason as the VLM.
36
+ * Data curation and annotation — Enable developers to automate high-quality curation and annotation of massive, diverse training datasets. Experience [NVIDIA Cosmos Curator](https://github.com/nvidia-cosmos/cosmos-curate), powered by Cosmos Reason, a framework that enables developers to quickly filter, annotate, and deduplicate large amounts of sensor data necessary for physical AI development.
37
+ * Robot planning and reasoning — Act as the brain for deliberate, methodical decision-making in a robot vision language action (VLA) model. Now robots such as humanoids and autonomous vehicles (AV) can interpret environments and complex commands, break them down into tasks and execute them using common sense, even in unfamiliar environments. Explore the [NVIDIA Isaac GR00T-Dreams blueprint](https://github.com/nvidia/gr00t-dreams), which generates vast amounts of synthetic trajectory data using NVIDIA Cosmos world foundation models.
38
+
39
+ Explore the [Cosmos Cookbook](https://nvidia-cosmos.github.io/cosmos-cookbook/index.html), a technical guide that delivers end-to-end workflows, implementation recipes, and detailed examples for building, fine-tuning, and deploying Cosmos Reason in production-ready environments.
40
+
41
+ The model is ready for commercial use.
42
+
43
+ **Model Developer**: NVIDIA
44
+
45
+ ## Model Versions
46
+
47
+ The Cosmos-Reason2 includes the following model:
48
+
49
+ - [Cosmos-Reason2-2B](https://huggingface.co/nvidia/Cosmos-Reason2-2B): Given a text prompt and an input video, think and generate the answer with respect to the input text prompt and video.
50
+ - [Cosmos-Reason2-8B](https://huggingface.co/nvidia/Cosmos-Reason2-8B): Given a text prompt and an input video, think and generate the answer with respect to the input text prompt and video.
51
+ - [Cosmos-Reason2-32B](https://huggingface.co/nvidia/Cosmos-Reason2-32B): Given a text prompt and an input video, think and generate the answer with respect to the input text prompt and video.
52
+
53
+ ## Model Quality Performance
54
+ [04/28/2026] For comparative evaluation, we present benchmark scores using the [Physical AI Bench Leaderboard](https://huggingface.co/spaces/shi-labs/physical-ai-bench-leaderboard).
55
+
56
+ ![PAI Bench Leaderboard](PAI_Bench_Leaderboard.png)
57
+
58
+ We also evaluated Cosmos-Reason2 against Qwen3-VL across the following domains and categories: General, Robotics, Self-Driving, and Smart Spaces.
59
+
60
+ | Category | Benchmark | Cosmos-Reason2-32B | Qwen3-VL-32B-Instruct | Cosmos-Reason2-8B | Qwen3-VL-8B-Instruct | Cosmos-Reason2-2B | Qwen3-VL-2B-Instruct |
61
+ |----------|-----------|:-------------------:|:----------------------:|:-------------------:|:----------------------:|:-------------------:|:----------------------:|
62
+ | General | Overall | **75.85** | 73.07 | **73.73** | 71.98 | **62.21** | 59.60 |
63
+ | | [BlinkDepth](https://huggingface.co/datasets/BLINK-Benchmark/BLINK) | **84.68** | 82.26 | **87.90** | 87.10 | **82.26** | 74.19 |
64
+ | | [BlinkSpatial](https://huggingface.co/datasets/BLINK-Benchmark/BLINK) | **86.71** | **86.71** | 84.62 | **87.41** | 75.52 | **77.62** |
65
+ | | [CVBench](https://huggingface.co/datasets/nyu-visionx/CV-Bench) | **88.01** | 86.74 | **85.60** | 85.17 | **78.74** | 78.67 |
66
+ | | [VideoPhy2](https://arxiv.org/pdf/2503.06800) | **43.98** | 36.57 | **36.80** | 28.24 | **12.33** | 7.92 |
67
+ | Robotics | Overall | **60.60** | 55.06 | **56.90** | 53.08 | **45.52** | 42.07 |
68
+ | | [ERQA](https://github.com/embodiedreasoning/ERQA) | 45.25 | **46.50** | **44.00** | **44.00** | **37.75** | **37.75** |
69
+ | | [CR Common](https://huggingface.co/spaces/shi-labs/physical-ai-bench-leaderboard) | **65.89** | 63.41 | **64.24** | 58.44 | **54.30** | 49.67 |
70
+ | | [CR Embodied](https://huggingface.co/spaces/shi-labs/physical-ai-bench-leaderboard) | **75.25** | 59.34 | **69.34** | 56.89 | **58.03** | 48.85 |
71
+ | | [Where2Place](https://arxiv.org/pdf/2406.10721) | **56.00** | 51.00 | 50.00 | **53.00** | **32.00** | **32.00** |
72
+ | Self-Driving | Overall | **70.15** | 48.08 | **67.85** | 46.38 | **57.37** | 42.73 |
73
+ | | [AV Collision](https://arxiv.org/pdf/2603.18178) | **72.06** | 37.67 | **74.00** | 34.00 | **74.33** | 36.33 |
74
+ | | [AV Stop](https://arxiv.org/pdf/2603.18178) | **69.39** | 38.78 | **57.14** | 36.73 | **38.78** | 32.65 |
75
+ | | [LingoQA](https://huggingface.co/papers/2312.14115) | **69.00** | 67.80 | **72.40** | 68.40 | 59.00 | **59.20** |
76
+ | Smart Spaces | Overall | **77.79** | 47.55 | **69.96** | 42.66 | **64.14** | 36.63 |
77
+ | | [Warehouse AI](https://huggingface.co/datasets/nvidia/PhysicalAI-Spatial-Intelligence-Warehouse) | **77.79** | 47.55 | **69.96** | 42.66 | **64.14** | 36.63 |
78
+
79
+ ![PAI Bench Leaderboard](CR2%20Performance%20by%20Category.png)
80
+
81
+ ### License:
82
+
83
+ This model is released under the [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license). Additional Information: [Apache License 2.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md).
84
+
85
+ For a custom license, please contact [cosmos-license@nvidia.com](mailto:cosmos-license@nvidia.com).
86
+
87
+ Under the NVIDIA Open Model License, NVIDIA confirms:
88
+
89
+ * Models are commercially usable.
90
+ * You are free to create and distribute Derivative Models.
91
+ * NVIDIA does not claim ownership to any outputs generated using the Models or Derivative Models.
92
+
93
+ **Important Note**: If You bypass, disable, reduce the efficacy of, or circumvent any technical limitation, safety guardrail or associated safety guardrail hyperparameter, encryption, security, digital rights management, or authentication mechanism (collectively “Guardrail”) contained in the Model without a substantially similar Guardrail appropriate for your use case, your rights under this Agreement [NVIDIA Open Model License Agreement](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license) will automatically terminate.
94
+
95
+ ### Deployment Geography:
96
+
97
+ Global
98
+
99
+ ### Use Case:
100
+
101
+ Physical AI: Space, time, fundamental physics understanding and embodied reasoning, encompassing robotics, and autonomous vehicles (AV).
102
+
103
+ ### Release Date:
104
+
105
+ * Github: [12/19/2025](https://github.com/nvidia-cosmos/cosmos-reason)
106
+ * Huggingface:
107
+ * [03/10/2026](https://huggingface.co/nvidia/Cosmos-Reason2-2B/commit/2f6b325db8bf501da0ae325a16390dd6d3f68856). Improved benchmark performance and reduced hallucinations.
108
+ * [12/19/2025](https://huggingface.co/nvidia/Cosmos-Reason2-2B/commit/99a6223bc1c93b29841bc71ad9f8844400b7638e). Initial release.
109
+
110
+ ## Model Architecture:
111
+
112
+ Architecture Type: A Multi-modal LLM consists of a Vision Transformer (ViT) for vision encoder and a Dense Transformer model for LLM.
113
+ Network Architecture: Qwen3-VL-2B-Instruct.
114
+
115
+ Cosmos-Reason2-2B is post-trained based on [Qwen3-VL-2B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-2B-Instruct) and follows the same model architecture.
116
+
117
+ **Number of model parameters:**
118
+
119
+ Cosmos-Reason2-2B: 2,438,696,960
120
+
121
+ ## Input
122
+
123
+ **Input Type(s)**: Text+Video/Image
124
+
125
+ **Input Format(s)**:
126
+
127
+ * Text: String
128
+ * Video: mp4
129
+ * Image: jpg
130
+
131
+ **Input Parameters**:
132
+ * Text: One-dimensional (1D)
133
+ * Video: Three-dimensional (3D)
134
+ * Image: Two-dimensional (2D)
135
+
136
+ **Other Properties Related to Input**:
137
+ * Use `FPS=4` for input video to match the training setup.
138
+ * Append `Answer the question in the following format: <think>\nyour reasoning\n</think>\n\n<answer>\nyour answer\n</answer>.` in the system prompt to encourage long chain-of-thought reasoning response.
139
+
140
+ ## Output
141
+
142
+ **Output Type(s)**: Text
143
+
144
+ **Output Format**: String
145
+
146
+ **Output Parameters**: Text: One-dimensional (1D)
147
+
148
+ **Other Properties Related to Output**:
149
+
150
+ * Recommend using 4096 or more output max tokens to avoid truncation of long chain-of-thought response.
151
+ * Our AI model recognizes timestamps added at the bottom of each frame for accurate temporal localization.
152
+ * Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA’s hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.
153
+
154
+ ## Software Integration
155
+
156
+ **Runtime Engine(s):**
157
+
158
+ * [Transformers](https://github.com/huggingface/transformers)
159
+
160
+ **Supported Hardware Microarchitecture Compatibility:**
161
+
162
+ * NVIDIA Blackwell
163
+ * NVIDIA Hopper
164
+
165
+ **Note**: We have only tested doing inference with BF16 precision.
166
+
167
+ **Operating System(s):**
168
+
169
+ * Linux (We have not tested on other operating systems.)
170
+
171
+ The integration of foundation and fine-tuned models into AI systems requires additional testing using use-case-specific data to ensure safe and effective deployment. Following the V-model methodology, iterative testing and validation at both unit and system levels are essential to mitigate risks, meet technical and functional requirements, and ensure compliance with safety and ethical standards before deployment.
172
+
173
+ # Usage
174
+
175
+ See [Cosmos-Reason2](https://github.com/nvidia-cosmos/cosmos-reason2) for details.
176
+
177
+ * Post Training: [Cosmos-Reason2](https://github.com/nvidia-cosmos/cosmos-reason2) provides examples of supervised fine-tuning and reinforcement learning on embodied reasoning datasets.
178
+
179
+ ## Training and Evaluation Sections:
180
+
181
+ Cosmos-Reason2-2B model was trained and evaluated on the same datasets used for [Cosmos-Reason1-7B](https://huggingface.co/nvidia/Cosmos-Reason1-7B#training-datasets), in addition to the following newly added datasets.
182
+
183
+ ## Training Datasets:
184
+
185
+ **Data Collection Method**:
186
+
187
+ [12/19/2025]
188
+ * EgoExo4D: Hybrid: Automatic/Sensors
189
+ * PerceptionTest: Hybrid: Automatic/Sensors
190
+ * Language Table: Hybrid: Automatic/Sensors
191
+ * IntPhys: Hybrid: Automatic/Sensors
192
+ * InfLevel: Hybrid: Automatic/Sensors
193
+ * CLEVRER: Hybrid: Automatic/Sensors
194
+
195
+ [03/10/2026]
196
+ * VideoPhy2: Hybrid: Automatic/Sensors
197
+ * NVIDIA Physical AI AV: Hybrid: Automatic/Sensors
198
+ * NVIDIA Physical AI Metropolis: Hybrid: Automatic/Sensors
199
+ * Hyperism: Hybrid: Automatic/Sensors
200
+ * EgoExOR: Hybrid: Automatic/Sensors
201
+
202
+ **Labeling Method**:
203
+
204
+ [12/19/2025]
205
+ * EgoExo4D: Hybrid: Human,Automated
206
+ * PerceptionTest: Hybrid: Human,Automated
207
+ * Language Table: Hybrid: Human,Automated
208
+ * IntPhys: Hybrid: Hybrid: Human,Automated
209
+ * InfLevel: Hybrid: Hybrid: Human,Automated
210
+ * CLEVRER: Hybrid: Hybrid: Human,Automated
211
+
212
+ [03/10/2026]
213
+ * VideoPhy2: Hybrid: Automatic/Sensors
214
+ * NVIDIA Physical AI AV: Hybrid: Automatic/Sensors
215
+ * NVIDIA Physical AI Metropolis: Hybrid: Automatic/Sensors
216
+ * Hyperism: Hybrid: Automatic/Sensors
217
+ * EgoExOR: Hybrid: Automatic/Sensors
218
+
219
+ The combined datasets span multimodal video, sensor signals, and structured physical-reasoning tasks, providing broad coverage for training world-model reasoning capabilities.
220
+
221
+ # Evaluation Datasets:
222
+
223
+ **Data Collection Method**:
224
+
225
+ [12/19/2025]
226
+ * EgoExo4D: Hybrid: Automatic/Sensors
227
+ * PerceptionTest: Hybrid: Automatic/Sensors
228
+ * Language Table: Hybrid: Automatic/Sensors
229
+ * IntPhys: Hybrid: Automatic/Sensors
230
+ * InfLevel: Hybrid: Automatic/Sensors
231
+ * CLEVRER: Hybrid: Automatic/Sensors
232
+
233
+ [03/10/2026]
234
+ * VideoPhy2: Hybrid: Automatic/Sensors
235
+ * NVIDIA Physical AI AV: Hybrid: Automatic/Sensors
236
+ * NVIDIA Physical AI Metropolis: Hybrid: Automatic/Sensors
237
+ * Hyperism: Hybrid: Automatic/Sensors
238
+ * EgoExOR: Hybrid: Automatic/Sensors
239
+
240
+ **Labeling Method**:
241
+
242
+ [12/19/2025]
243
+ * EgoExo4D: Hybrid: Human,Automated
244
+ * PerceptionTest: Hybrid: Human,Automated
245
+ * Language Table: Hybrid: Human,Automated
246
+ * IntPhys: Hybrid: Hybrid: Human,Automated
247
+ * InfLevel: Hybrid: Hybrid: Human,Automated
248
+ * CLEVRER: Hybrid: Hybrid: Human,Automated
249
+
250
+ [03/10/2026]
251
+ * VideoPhy2: Hybrid: Automatic/Sensors
252
+ * NVIDIA Physical AI AV: Hybrid: Automatic/Sensors
253
+ * NVIDIA Physical AI Metropolis: Hybrid: Automatic/Sensors
254
+ * Hyperism: Hybrid: Automatic/Sensors
255
+ * EgoExOR: Hybrid: Automatic/Sensors
256
+
257
+ The combined datasets span multimodal video, sensor signals, and structured physical-reasoning tasks, providing broad coverage for training world-model reasoning capabilities.
258
+
259
+ ## Dataset Format
260
+
261
+ Modality: Video (mp4) and Text
262
+
263
+ ## Inference:
264
+
265
+ **Test Hardware:** H100, A100
266
+
267
+ > [!NOTE]
268
+ > We suggest using `fps=4` for the input video and `max_tokens=4096` to avoid truncated response.
269
+
270
+ ```python
271
+ import transformers
272
+ import torch
273
+
274
+ model_name = "nvidia/Cosmos-Reason2-2B"
275
+ model = transformers.Qwen3VLForConditionalGeneration.from_pretrained(
276
+ model_name, dtype=torch.float16, device_map="auto", attn_implementation="sdpa"
277
+ )
278
+ processor: transformers.Qwen3VLProcessor = (
279
+ transformers.AutoProcessor.from_pretrained(model_name)
280
+ )
281
+
282
+ video_messages = [
283
+ {
284
+ "role": "system",
285
+ "content": [{"type": "text", "text": "You are a helpful assistant."}],
286
+ },
287
+ {"role": "user", "content": [
288
+ {
289
+ "type": "video",
290
+ "video": "file:///path/to/your/video.mp4",
291
+ "fps": 4,
292
+ },
293
+ {"type": "text", "text": (
294
+ "Is it safe to turn right? Answer the question using the following format:\n\n<think>\nYour reasoning.\n</think>\n\nWrite your final answer immediately after the </think> tag."
295
+ )
296
+ },
297
+ ]
298
+ },
299
+ ]
300
+
301
+ # Process inputs
302
+ inputs = processor.apply_chat_template(
303
+ video_messages,
304
+ tokenize=True,
305
+ add_generation_prompt=True,
306
+ return_dict=True,
307
+ return_tensors="pt",
308
+ fps=4,
309
+ )
310
+ inputs = inputs.to(model.device)
311
+
312
+ # Run inference
313
+ generated_ids = model.generate(**inputs, max_new_tokens=4096)
314
+ generated_ids_trimmed = [
315
+ out_ids[len(in_ids) :]
316
+ for in_ids, out_ids in zip(inputs.input_ids, generated_ids, strict=False)
317
+ ]
318
+ output_text = processor.batch_decode(
319
+ generated_ids_trimmed,
320
+ skip_special_tokens=True,
321
+ clean_up_tokenization_spaces=False,
322
+ )
323
+
324
+ ```
325
+
326
+ #### System Requirements and Performance
327
+
328
+ This model requires a minimum of 24 GB of GPU memory. Inference latency for a single generation across different NVIDIA GPU platforms will be published shortly.
329
+
330
+ ## Ethical Considerations
331
+
332
+ NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
333
+
334
+ Users are responsible for model inputs and outputs. Users are responsible for ensuring safe integration of this model, including implementing guardrails as well as other safety mechanisms, prior to deployment.
335
+
336
+ For more detailed information on ethical considerations for this model, please see the subcards of Explainability, Bias, Safety & Security, and Privacy below.
337
+
338
+ Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).
339
+
340
+ ### Plus Plus (++) Promise
341
+
342
+ We value you, the datasets, the diversity they represent, and what we have been entrusted with. This model and its associated data have been:
343
+
344
+ * Verified to comply with current applicable disclosure laws, regulations, and industry standards.
345
+ * Verified to comply with applicable privacy labeling requirements.
346
+ * Annotated to describe the collector/source (NVIDIA or a third-party).
347
+ * Characterized for technical limitations.
348
+ * Reviewed to ensure proper disclosure is accessible to, maintained for, and in compliance with NVIDIA data subjects and their requests.
349
+ * Reviewed before release.
350
+ * Tagged for known restrictions and potential safety implications.
351
+
352
+ ### Bias
353
+ | Field | Response |
354
+ | :-------------------------------------------------------------------------------------------------------------------------------------------------------------- | :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
355
+ | Participation considerations from adversely impacted groups [protected classes](https://www.senate.ca.gov/content/protected-classes) in model design and testing: | None |
356
+ | Measures taken to mitigate against unwanted bias: | The training video sources contain multiple physical embodiments and environments including human, car, single arm robot, bimanual robot in indoor and outdoor environments. By training on numerous and various physical interactions and curated datasets, we strive to provide a model that mitigates biases towards certain embodiments or environments. |
357
+
358
+ ### Explainability
359
+ | Field | Response |
360
+ | :-------------------------------------------------------- | :-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
361
+ | Intended Application & Domain: | Physical AI Reasoning |
362
+ | Model Type: | Transformer |
363
+ | Intended Users: | Physical AI developers |
364
+ | Output: | Text |
365
+ | Describe how the model works: | Given a video/image and a text prompt, the model first converts the video/image into tokens using a vision encoder and a special translator called a projector. These video tokens are combined with the text prompt and fed into the core model, which uses a mix of LLM modules and techniques. This enables the model to think step-by-step and provide detailed, logical responses. |
366
+ | Technical Limitations: | The model may not follow the video or text input accurately in challenging cases, where the input video shows complex scene composition and temporal dynamics. Examples of challenging scenes include: fast camera movements, overlapping human-object interactions, low lighting with high motion blur, and multiple people performing different actions simultaneously. |
367
+ | Verified to have met prescribed NVIDIA quality standards: | Yes |
368
+ | Performance Metrics: | Quantitative and Qualitative Evaluation. Cosmos-Reason2 proposes the embodied reasoning benchmark and physical common sense benchmark to evaluate accuracy with visual question answering. |
369
+ | Potential Known Risks: | The model's output can generate all forms of texts, including what may be considered toxic, offensive, or indecent. |
370
+ | Licensing: | [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license). Additional Information: [Apache License 2.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md). |
371
+
372
+ ### Privacy
373
+ | Field | Response |
374
+ | :------------------------------------------------------------------ | :----------------------------------------------------------------------------- |
375
+ | Generatable or reverse engineerable personal data? | No |
376
+ | Personal data used to create this model? | None Known |
377
+ | Was consent obtained for any personal data used? | None Known |
378
+ | How often is dataset reviewed? | Before Release |
379
+ | Is there provenance for all datasets used in training? | Yes |
380
+ | Does data labeling (annotation, metadata) comply with privacy laws? | Yes |
381
+ | Applicable Privacy Policy | [NVIDIA Privacy Policy](https://www.nvidia.com/en-us/about-nvidia/privacy-policy) |
382
+
383
+ ### Safety
384
+ | Field | Response |
385
+ | :---------------------------------------------- | :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
386
+ | Model Application(s): | Physical AI common sense understanding and embodied reasoning |
387
+ | Describe the life critical impact (if present). | Because this model is designed for robot planning and serves as a Vision-Language-Action (VLA) model, its outputs directly influence physical actuation. Planning errors or misinterpretations of the environment carry inherent life-safety risks, including physical collisions, unsafe object manipulation, or unintended interactions with humans and property.|
388
+ | Use Case Restrictions: | [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license). Additional Information: [Apache License 2.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md). |
389
+ | Model and dataset restrictions: | The Principle of least privilege (PoLP) is applied limiting access for dataset generation and model development. Restrictions enforce dataset access during training, and dataset license constraints adhered to. Model checkpoints are made available on Hugging Face, and may become available on cloud providers' model catalog. |
chat_template.json ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {
2
+ "chat_template": "{%- if tools %}\n {{- '<|im_start|>system\\n' }}\n {%- if messages[0].role == 'system' %}\n {%- if messages[0].content is string %}\n {{- messages[0].content }}\n {%- else %}\n {%- for content in messages[0].content %}\n {%- if 'text' in content %}\n {{- content.text }}\n {%- endif %}\n {%- endfor %}\n {%- endif %}\n {{- '\\n\\n' }}\n {%- endif %}\n {{- \"# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n {%- for tool in tools %}\n {{- \"\\n\" }}\n {{- tool | tojson }}\n {%- endfor %}\n {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n {%- if messages[0].role == 'system' %}\n {{- '<|im_start|>system\\n' }}\n {%- if messages[0].content is string %}\n {{- messages[0].content }}\n {%- else %}\n {%- for content in messages[0].content %}\n {%- if 'text' in content %}\n {{- content.text }}\n {%- endif %}\n {%- endfor %}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n{%- endif %}\n{%- set image_count = namespace(value=0) %}\n{%- set video_count = namespace(value=0) %}\n{%- for message in messages %}\n {%- if message.role == \"user\" %}\n {{- '<|im_start|>' + message.role + '\\n' }}\n {%- if message.content is string %}\n {{- message.content }}\n {%- else %}\n {%- for content in message.content %}\n {%- if content.type == 'image' or 'image' in content or 'image_url' in content %}\n {%- set image_count.value = image_count.value + 1 %}\n {%- if add_vision_id %}Picture {{ image_count.value }}: {% endif -%}\n <|vision_start|><|image_pad|><|vision_end|>\n {%- elif content.type == 'video' or 'video' in content %}\n {%- set video_count.value = video_count.value + 1 %}\n {%- if add_vision_id %}Video {{ video_count.value }}: {% endif -%}\n <|vision_start|><|video_pad|><|vision_end|>\n {%- elif 'text' in content %}\n {{- content.text }}\n {%- endif %}\n {%- endfor %}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n {%- elif message.role == \"assistant\" %}\n {{- '<|im_start|>' + message.role + '\\n' }}\n {%- if message.content is string %}\n {{- message.content }}\n {%- else %}\n {%- for content_item in message.content %}\n {%- if 'text' in content_item %}\n {{- content_item.text }}\n {%- endif %}\n {%- endfor %}\n {%- endif %}\n {%- if message.tool_calls %}\n {%- for tool_call in message.tool_calls %}\n {%- if (loop.first and message.content) or (not loop.first) %}\n {{- '\\n' }}\n {%- endif %}\n {%- if tool_call.function %}\n {%- set tool_call = tool_call.function %}\n {%- endif %}\n {{- '<tool_call>\\n{\"name\": \"' }}\n {{- tool_call.name }}\n {{- '\", \"arguments\": ' }}\n {%- if tool_call.arguments is string %}\n {{- tool_call.arguments }}\n {%- else %}\n {{- tool_call.arguments | tojson }}\n {%- endif %}\n {{- '}\\n</tool_call>' }}\n {%- endfor %}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n {%- elif message.role == \"tool\" %}\n {%- if loop.first or (messages[loop.index0 - 1].role != \"tool\") %}\n {{- '<|im_start|>user' }}\n {%- endif %}\n {{- '\\n<tool_response>\\n' }}\n {%- if message.content is string %}\n {{- message.content }}\n {%- else %}\n {%- for content in message.content %}\n {%- if content.type == 'image' or 'image' in content or 'image_url' in content %}\n {%- set image_count.value = image_count.value + 1 %}\n {%- if add_vision_id %}Picture {{ image_count.value }}: {% endif -%}\n <|vision_start|><|image_pad|><|vision_end|>\n {%- elif content.type == 'video' or 'video' in content %}\n {%- set video_count.value = video_count.value + 1 %}\n {%- if add_vision_id %}Video {{ video_count.value }}: {% endif -%}\n <|vision_start|><|video_pad|><|vision_end|>\n {%- elif 'text' in content %}\n {{- content.text }}\n {%- endif %}\n {%- endfor %}\n {%- endif %}\n {{- '\\n</tool_response>' }}\n {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
3
+ }
4
+
config.json ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Qwen3VLForConditionalGeneration"
4
+ ],
5
+ "image_token_id": 151655,
6
+ "model_type": "qwen3_vl",
7
+ "text_config": {
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "bos_token_id": 151643,
11
+ "dtype": "bfloat16",
12
+ "eos_token_id": 151645,
13
+ "head_dim": 128,
14
+ "hidden_act": "silu",
15
+ "hidden_size": 2048,
16
+ "initializer_range": 0.02,
17
+ "intermediate_size": 6144,
18
+ "max_position_embeddings": 262144,
19
+ "model_type": "qwen3_vl_text",
20
+ "num_attention_heads": 16,
21
+ "num_hidden_layers": 28,
22
+ "num_key_value_heads": 8,
23
+ "rms_norm_eps": 1e-06,
24
+ "rope_scaling": {
25
+ "mrope_interleaved": true,
26
+ "mrope_section": [
27
+ 24,
28
+ 20,
29
+ 20
30
+ ],
31
+ "rope_type": "default"
32
+ },
33
+ "rope_theta": 5000000,
34
+ "tie_word_embeddings": true,
35
+ "use_cache": true,
36
+ "vocab_size": 151936
37
+ },
38
+ "tie_word_embeddings": true,
39
+ "transformers_version": "4.57.0.dev0",
40
+ "video_token_id": 151656,
41
+ "vision_config": {
42
+ "deepstack_visual_indexes": [
43
+ 5,
44
+ 11,
45
+ 17
46
+ ],
47
+ "depth": 24,
48
+ "hidden_act": "gelu_pytorch_tanh",
49
+ "hidden_size": 1024,
50
+ "in_channels": 3,
51
+ "initializer_range": 0.02,
52
+ "intermediate_size": 4096,
53
+ "model_type": "qwen3_vl",
54
+ "num_heads": 16,
55
+ "num_position_embeddings": 2304,
56
+ "out_hidden_size": 2048,
57
+ "patch_size": 16,
58
+ "spatial_merge_size": 2,
59
+ "temporal_patch_size": 2
60
+ },
61
+ "vision_end_token_id": 151653,
62
+ "vision_start_token_id": 151652
63
+ }
cosmos_cr2_2b_paiBench.png ADDED

Git LFS Details

  • SHA256: 4b79d34191efa63420e64f4ded2a113b43cd3a2dadbb65e226e9d99a3442d795
  • Pointer size: 131 Bytes
  • Size of remote file: 138 kB
generation_config.json ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 151643,
3
+ "pad_token_id": 151643,
4
+ "do_sample": true,
5
+ "eos_token_id": [
6
+ 151645,
7
+ 151643
8
+ ],
9
+ "top_p": 0.8,
10
+ "top_k": 20,
11
+ "temperature": 0.7,
12
+ "repetition_penalty": 1.0,
13
+ "transformers_version": "4.56.0"
14
+ }
merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fa5a6e6ef4fce40216b185cc48a3b24d31637ac3e2ba69c107ed1f389c1e6ede
3
+ size 4877470304
preprocessor_config.json ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "size": {
3
+ "longest_edge": 16777216,
4
+ "shortest_edge": 65536
5
+ },
6
+ "patch_size": 16,
7
+ "temporal_patch_size": 2,
8
+ "merge_size": 2,
9
+ "image_mean": [
10
+ 0.5,
11
+ 0.5,
12
+ 0.5
13
+ ],
14
+ "image_std": [
15
+ 0.5,
16
+ 0.5,
17
+ 0.5
18
+ ],
19
+ "processor_class": "Qwen3VLProcessor",
20
+ "image_processor_type": "Qwen2VLImageProcessorFast"
21
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,239 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_prefix_space": false,
4
+ "added_tokens_decoder": {
5
+ "151643": {
6
+ "content": "<|endoftext|>",
7
+ "lstrip": false,
8
+ "normalized": false,
9
+ "rstrip": false,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "151644": {
14
+ "content": "<|im_start|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "151645": {
22
+ "content": "<|im_end|>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "151646": {
30
+ "content": "<|object_ref_start|>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "151647": {
38
+ "content": "<|object_ref_end|>",
39
+ "lstrip": false,
40
+ "normalized": false,
41
+ "rstrip": false,
42
+ "single_word": false,
43
+ "special": true
44
+ },
45
+ "151648": {
46
+ "content": "<|box_start|>",
47
+ "lstrip": false,
48
+ "normalized": false,
49
+ "rstrip": false,
50
+ "single_word": false,
51
+ "special": true
52
+ },
53
+ "151649": {
54
+ "content": "<|box_end|>",
55
+ "lstrip": false,
56
+ "normalized": false,
57
+ "rstrip": false,
58
+ "single_word": false,
59
+ "special": true
60
+ },
61
+ "151650": {
62
+ "content": "<|quad_start|>",
63
+ "lstrip": false,
64
+ "normalized": false,
65
+ "rstrip": false,
66
+ "single_word": false,
67
+ "special": true
68
+ },
69
+ "151651": {
70
+ "content": "<|quad_end|>",
71
+ "lstrip": false,
72
+ "normalized": false,
73
+ "rstrip": false,
74
+ "single_word": false,
75
+ "special": true
76
+ },
77
+ "151652": {
78
+ "content": "<|vision_start|>",
79
+ "lstrip": false,
80
+ "normalized": false,
81
+ "rstrip": false,
82
+ "single_word": false,
83
+ "special": true
84
+ },
85
+ "151653": {
86
+ "content": "<|vision_end|>",
87
+ "lstrip": false,
88
+ "normalized": false,
89
+ "rstrip": false,
90
+ "single_word": false,
91
+ "special": true
92
+ },
93
+ "151654": {
94
+ "content": "<|vision_pad|>",
95
+ "lstrip": false,
96
+ "normalized": false,
97
+ "rstrip": false,
98
+ "single_word": false,
99
+ "special": true
100
+ },
101
+ "151655": {
102
+ "content": "<|image_pad|>",
103
+ "lstrip": false,
104
+ "normalized": false,
105
+ "rstrip": false,
106
+ "single_word": false,
107
+ "special": true
108
+ },
109
+ "151656": {
110
+ "content": "<|video_pad|>",
111
+ "lstrip": false,
112
+ "normalized": false,
113
+ "rstrip": false,
114
+ "single_word": false,
115
+ "special": true
116
+ },
117
+ "151657": {
118
+ "content": "<tool_call>",
119
+ "lstrip": false,
120
+ "normalized": false,
121
+ "rstrip": false,
122
+ "single_word": false,
123
+ "special": false
124
+ },
125
+ "151658": {
126
+ "content": "</tool_call>",
127
+ "lstrip": false,
128
+ "normalized": false,
129
+ "rstrip": false,
130
+ "single_word": false,
131
+ "special": false
132
+ },
133
+ "151659": {
134
+ "content": "<|fim_prefix|>",
135
+ "lstrip": false,
136
+ "normalized": false,
137
+ "rstrip": false,
138
+ "single_word": false,
139
+ "special": false
140
+ },
141
+ "151660": {
142
+ "content": "<|fim_middle|>",
143
+ "lstrip": false,
144
+ "normalized": false,
145
+ "rstrip": false,
146
+ "single_word": false,
147
+ "special": false
148
+ },
149
+ "151661": {
150
+ "content": "<|fim_suffix|>",
151
+ "lstrip": false,
152
+ "normalized": false,
153
+ "rstrip": false,
154
+ "single_word": false,
155
+ "special": false
156
+ },
157
+ "151662": {
158
+ "content": "<|fim_pad|>",
159
+ "lstrip": false,
160
+ "normalized": false,
161
+ "rstrip": false,
162
+ "single_word": false,
163
+ "special": false
164
+ },
165
+ "151663": {
166
+ "content": "<|repo_name|>",
167
+ "lstrip": false,
168
+ "normalized": false,
169
+ "rstrip": false,
170
+ "single_word": false,
171
+ "special": false
172
+ },
173
+ "151664": {
174
+ "content": "<|file_sep|>",
175
+ "lstrip": false,
176
+ "normalized": false,
177
+ "rstrip": false,
178
+ "single_word": false,
179
+ "special": false
180
+ },
181
+ "151665": {
182
+ "content": "<tool_response>",
183
+ "lstrip": false,
184
+ "normalized": false,
185
+ "rstrip": false,
186
+ "single_word": false,
187
+ "special": false
188
+ },
189
+ "151666": {
190
+ "content": "</tool_response>",
191
+ "lstrip": false,
192
+ "normalized": false,
193
+ "rstrip": false,
194
+ "single_word": false,
195
+ "special": false
196
+ },
197
+ "151667": {
198
+ "content": "<think>",
199
+ "lstrip": false,
200
+ "normalized": false,
201
+ "rstrip": false,
202
+ "single_word": false,
203
+ "special": false
204
+ },
205
+ "151668": {
206
+ "content": "</think>",
207
+ "lstrip": false,
208
+ "normalized": false,
209
+ "rstrip": false,
210
+ "single_word": false,
211
+ "special": false
212
+ }
213
+ },
214
+ "additional_special_tokens": [
215
+ "<|im_start|>",
216
+ "<|im_end|>",
217
+ "<|object_ref_start|>",
218
+ "<|object_ref_end|>",
219
+ "<|box_start|>",
220
+ "<|box_end|>",
221
+ "<|quad_start|>",
222
+ "<|quad_end|>",
223
+ "<|vision_start|>",
224
+ "<|vision_end|>",
225
+ "<|vision_pad|>",
226
+ "<|image_pad|>",
227
+ "<|video_pad|>"
228
+ ],
229
+ "bos_token": null,
230
+ "chat_template": "{%- if tools %}\n {{- '<|im_start|>system\\n' }}\n {%- if messages[0].role == 'system' %}\n {%- if messages[0].content is string %}\n {{- messages[0].content }}\n {%- else %}\n {%- for content in messages[0].content %}\n {%- if 'text' in content %}\n {{- content.text }}\n {%- endif %}\n {%- endfor %}\n {%- endif %}\n {{- '\\n\\n' }}\n {%- endif %}\n {{- \"# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n {%- for tool in tools %}\n {{- \"\\n\" }}\n {{- tool | tojson }}\n {%- endfor %}\n {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n {%- if messages[0].role == 'system' %}\n {{- '<|im_start|>system\\n' }}\n {%- if messages[0].content is string %}\n {{- messages[0].content }}\n {%- else %}\n {%- for content in messages[0].content %}\n {%- if 'text' in content %}\n {{- content.text }}\n {%- endif %}\n {%- endfor %}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n{%- endif %}\n{%- set image_count = namespace(value=0) %}\n{%- set video_count = namespace(value=0) %}\n{%- for message in messages %}\n {%- if message.role == \"user\" %}\n {{- '<|im_start|>' + message.role + '\\n' }}\n {%- if message.content is string %}\n {{- message.content }}\n {%- else %}\n {%- for content in message.content %}\n {%- if content.type == 'image' or 'image' in content or 'image_url' in content %}\n {%- set image_count.value = image_count.value + 1 %}\n {%- if add_vision_id %}Picture {{ image_count.value }}: {% endif -%}\n <|vision_start|><|image_pad|><|vision_end|>\n {%- elif content.type == 'video' or 'video' in content %}\n {%- set video_count.value = video_count.value + 1 %}\n {%- if add_vision_id %}Video {{ video_count.value }}: {% endif -%}\n <|vision_start|><|video_pad|><|vision_end|>\n {%- elif 'text' in content %}\n {{- content.text }}\n {%- endif %}\n {%- endfor %}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n {%- elif message.role == \"assistant\" %}\n {{- '<|im_start|>' + message.role + '\\n' }}\n {%- if message.content is string %}\n {{- message.content }}\n {%- else %}\n {%- for content_item in message.content %}\n {%- if 'text' in content_item %}\n {{- content_item.text }}\n {%- endif %}\n {%- endfor %}\n {%- endif %}\n {%- if message.tool_calls %}\n {%- for tool_call in message.tool_calls %}\n {%- if (loop.first and message.content) or (not loop.first) %}\n {{- '\\n' }}\n {%- endif %}\n {%- if tool_call.function %}\n {%- set tool_call = tool_call.function %}\n {%- endif %}\n {{- '<tool_call>\\n{\"name\": \"' }}\n {{- tool_call.name }}\n {{- '\", \"arguments\": ' }}\n {%- if tool_call.arguments is string %}\n {{- tool_call.arguments }}\n {%- else %}\n {{- tool_call.arguments | tojson }}\n {%- endif %}\n {{- '}\\n</tool_call>' }}\n {%- endfor %}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n {%- elif message.role == \"tool\" %}\n {%- if loop.first or (messages[loop.index0 - 1].role != \"tool\") %}\n {{- '<|im_start|>user' }}\n {%- endif %}\n {{- '\\n<tool_response>\\n' }}\n {%- if message.content is string %}\n {{- message.content }}\n {%- else %}\n {%- for content in message.content %}\n {%- if content.type == 'image' or 'image' in content or 'image_url' in content %}\n {%- set image_count.value = image_count.value + 1 %}\n {%- if add_vision_id %}Picture {{ image_count.value }}: {% endif -%}\n <|vision_start|><|image_pad|><|vision_end|>\n {%- elif content.type == 'video' or 'video' in content %}\n {%- set video_count.value = video_count.value + 1 %}\n {%- if add_vision_id %}Video {{ video_count.value }}: {% endif -%}\n <|vision_start|><|video_pad|><|vision_end|>\n {%- elif 'text' in content %}\n {{- content.text }}\n {%- endif %}\n {%- endfor %}\n {%- endif %}\n {{- '\\n</tool_response>' }}\n {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n",
231
+ "clean_up_tokenization_spaces": false,
232
+ "eos_token": "<|im_end|>",
233
+ "errors": "replace",
234
+ "model_max_length": 262144,
235
+ "pad_token": "<|endoftext|>",
236
+ "split_special_tokens": false,
237
+ "tokenizer_class": "Qwen2Tokenizer",
238
+ "unk_token": null
239
+ }
video_preprocessor_config.json ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "size": {
3
+ "longest_edge": 25165824,
4
+ "shortest_edge": 4096
5
+ },
6
+ "patch_size": 16,
7
+ "temporal_patch_size": 2,
8
+ "merge_size": 2,
9
+ "image_mean": [
10
+ 0.5,
11
+ 0.5,
12
+ 0.5
13
+ ],
14
+ "image_std": [
15
+ 0.5,
16
+ 0.5,
17
+ 0.5
18
+ ],
19
+ "processor_class": "Qwen3VLProcessor",
20
+ "video_processor_type": "Qwen3VLVideoProcessor"
21
+ }
vocab.json ADDED
The diff for this file is too large to render. See raw diff