Hugging Face's logo

nvidia
/
Cosmos-Reason1-7B

Image-Text-to-Text
Transformers
Safetensors
Cosmos
English
qwen2_5_vl
nvidia
conversational
text-generation-inference
Model card Files Community
2
tsungyi commited on
Commit
01f631b
·
1 Parent(s): 4e92fc4

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +142 -0
README.md ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ datasets:
7
+ - nvidia/Cosmos-Reason1-SFT-Dataset-Sample
8
+ - nvidia/Cosmos-Reason1-RL-Dataset-Sample
9
+ - nvidia/Cosmos-Reason1-Benchmark-Sample
10
+ library_name: cosmos
11
+ language:
12
+ - en
13
+ base_model:
14
+ - Qwen/Qwen2.5-VL-7B-Instruct
15
+ tags:
16
+ - nvidia
17
+ - cosmos
18
+ ---
19
+
20
+
21
+ # **Cosmos-Reason1: Physical AI Common Sense and Embodied Reasoning Models**
22
+
23
+ [**Cosmos**](https://huggingface.co/collections/nvidia/cosmos-reason1-67c9e926206426008f1da1b7) | [**Code**](https://github.com/nvidia-cosmos/cosmos-reason1) | [**Paper**](https://arxiv.org/abs/2503.15558) | [**Paper Website**](https://research.nvidia.com/labs/dir/cosmos-reason1)
24
+
25
+ # Model Overview
26
+
27
+ ## Description:
28
+
29
+ **Cosmos-Reason1 Models**: Physical AI models understand the physical common sense and generate appropriate embodied decisions in natural language through long chain-of-thought reasoning processes.
30
+
31
+ The Cosmos-Reason1 models are post-trained with physical common sense and embodied reasoning data with supervised fine-tuning and RL. It can serve as a critic model to reason about AI-generated videos defying physical laws or a planning model to reason about the next action of an embodied agent. The models are ready for commercial use.
32
+
33
+ **Model Developer**: NVIDIA
34
+
35
+ ## Model Versions
36
+
37
+ The Cosmos-Reason1 includes the following model:
38
+
39
+ - [Cosmos-Reason1-7B](https://huggingface.co/nvidia/Cosmos-Reason1-7B)
40
+ - Given a text prompt and an input video, think and generate the answer with respect to the input text prompt and video.
41
+
42
+ ### License:
43
+
44
+ This model is released under the [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license). For a custom license, please contact [[email protected]](mailto:[email protected]).
45
+
46
+ Under the NVIDIA Open Model License, NVIDIA confirms:
47
+
48
+ * Models are commercially usable.
49
+ * You are free to create and distribute Derivative Models.
50
+ * NVIDIA does not claim ownership to any outputs generated using the Models or Derivative Models.
51
+
52
+ **Important Note**: If you bypass, disable, reduce the efficacy of, or circumvent any technical limitation, **safety guardrail** or
53
+ associated safety guardrail hyperparameter, encryption, security, digital rights management, or authentication mechanism contained
54
+ in the Model, your rights under [NVIDIA Open Model License Agreement](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license) will automatically terminate.
55
+
56
+ ### Deployment Geography:
57
+
58
+ Global
59
+
60
+ ### Use Case:
61
+
62
+ Physical AI: synthetic data evaluation, encompassing robotics, autonomous vehicles (AV), and more.
63
+
64
+ ### Release Date:
65
+
66
+ 05/17/2025
67
+
68
+ ## Model Architecture:
69
+
70
+ Cosmos-Reason-7B is developed based on [https://Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct) and follows the same model architecture.
71
+
72
+ ## Software Integration
73
+
74
+ # Usage
75
+
76
+ See [Cosmos-Reason1](https://github.com/nvidia-cosmos/cosmos-reason1) for details.
77
+ * Post Training: [Cosmos-Reason1](https://github.com/nvidia-cosmos/cosmos-reason1) provides examples of supervised fine-tuning and reinforcement learning on embodied reasoning datasets.
78
+
79
+ # Evaluation
80
+
81
+ Please see our [technical paper](https://arxiv.org/pdf/2503.15558) for detailed evaluations on physical common sense and embodied reasoning. Part of the evaluation datasets are released under [Cosmos-Reason1-Benchmark-Sample](https://huggingface.co/datasets/nvidia/Cosmos-Reason1-Benchmark-Sample)
82
+
83
+ ## Ethical Considerations
84
+
85
+ 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.
86
+
87
+ For more detailed information on ethical considerations for this model, please see the subcards of Explainability, Bias, Safety & Security, and Privacy below. Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).
88
+
89
+ ### Plus Plus (++) Promise
90
+
91
+ We value you, the datasets, the diversity they represent, and what we have been entrusted with. This model and its associated data have been:
92
+
93
+ * Verified to comply with current applicable disclosure laws, regulations, and industry standards.
94
+ * Verified to comply with applicable privacy labeling requirements.
95
+ * Annotated to describe the collector/source (NVIDIA or a third-party).
96
+ * Characterized for technical limitations.
97
+ * Reviewed to ensure proper disclosure is accessible to, maintained for, and in compliance with NVIDIA data subjects and their requests.
98
+ * Reviewed before release.
99
+ * Tagged for known restrictions and potential safety implications.
100
+
101
+ ### Bias
102
+
103
+ | Field | Response |
104
+ | :--------------------------------------------------------------------------------------------------------------------------------------------------------------- | :------- |
105
+ | Participation considerations from adversely impacted groups[protected classes](https://www.senate.ca.gov/content/protected-classes) in model design and testing: | None |
106
+ | Measures taken to mitigate against unwanted bias: | None |
107
+
108
+ ### Explainability
109
+
110
+ | Field | Response |
111
+ | :-------------------------------------------------------- | :------------------------------------------------------------------------------------------------------------------- |
112
+ | Intended Application & Domain: | Physical AI Reasoning |
113
+ | Model Type: | Transformer |
114
+ | Intended Users: | Physical AI developers |
115
+ | Output: | Text |
116
+ | Describe how the model works: | Generates text answers based on input text prompt and video |
117
+ | Technical Limitations: | The model may not follow the video or text input accurately in challenging cases |
118
+ | Verified to have met prescribed NVIDIA quality standards: | Yes |
119
+ | Performance Metrics: | Quantitative and Qualitative Evaluation |
120
+ | Potential Known Risks: | The model's output can generate all forms of texts, including what may be considered toxic, offensive, or indecent. |
121
+ | Licensing: | [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license) |
122
+
123
+ ### Privacy
124
+
125
+ | Field | Response |
126
+ | :------------------------------------------------------------------ | :------------- |
127
+ | Generatable or reverse engineerable personal information? | None Known |
128
+ | Protected class data used to create this model? | None Known |
129
+ | Was consent obtained for any personal data used? | None Known |
130
+ | How often is dataset reviewed? | Before Release |
131
+ | Is there provenance for all datasets used in training? | Yes |
132
+ | Does data labeling (annotation, metadata) comply with privacy laws? | Yes |
133
+
134
+ ### Safety
135
+
136
+ | Field | Response |
137
+ | :---------------------------------------------- | :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
138
+ | Model Application(s): | World Generation |
139
+ | Describe the life critical impact (if present). | None Known |
140
+ | Use Case Restrictions: | [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license) |
141
+ | 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. |
142
+