Commit ·
9f0ebb3
1
Parent(s): de07ad9
added subcards
Browse files- bias.md +7 -0
- explainability.md +18 -0
- privacy.md +20 -0
- safety.md +11 -0
bias.md
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Bias Subcard
|
| 2 |
+
|
| 3 |
+
|
| 4 |
+
| Field | Response |
|
| 5 |
+
| ---------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ |
|
| 6 |
+
| Participation considerations from adversely impacted groups protected classes in model design and testing: | None |
|
| 7 |
+
| Measures taken to mitigate against unwanted bias: | None |
|
explainability.md
ADDED
|
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Explainability Subcard
|
| 2 |
+
|
| 3 |
+
|
| 4 |
+
| Field | Description |
|
| 5 |
+
| ----------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
|
| 6 |
+
| Intended Domain | Multimodal Content Safety |
|
| 7 |
+
| Model Type | Safety Classifier |
|
| 8 |
+
| Intended Users | AI/ML Engineers, LLM Developers, Safety Assurance Teams |
|
| 9 |
+
| Output | Text |
|
| 10 |
+
| Describe how the model works: | Type: Finetuned Transformer (Decoder-only) working as a classifier. Backbone: Google Gemma-3-4B-it Parameters: 4B (Billion) |
|
| 11 |
+
| Name the adversely impacted groups this has been tested to deliver comparable outcomes regardless of: | Not Applicable |
|
| 12 |
+
| Technical Limitations: | • The model only accepts a single text input along with an optional image. The model does not accept more than one image. |
|
| 13 |
+
| Verified to have met prescribed NVIDIA quality standards: | Yes |
|
| 14 |
+
| Performance Metrics: | Accuracy • F-1 Score • Throughput/Latency |
|
| 15 |
+
| Potential Known Risks: | The model may struggle to classify synthetically generated images. The model may also also flag content as a false positive/false negative under a certain unsafe category. |
|
| 16 |
+
| Terms of Use: | Use of the model is governed by the [OpenMDW License Agreement, version 1.1](https://raw.githubusercontent.com/OpenMDW/OpenMDW/refs/heads/main/1.1/LICENSE.OpenMDW-1.1) (OpenMDW-1.1), [Gemma Terms of Use](https://ai.google.dev/gemma/terms) and [Gemma Prohibited Use Policy](https://ai.google.dev/gemma/prohibited_use_policy). |
|
| 17 |
+
|
| 18 |
+
|
privacy.md
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Privacy Subcard
|
| 2 |
+
|
| 3 |
+
|
| 4 |
+
| Field | Response |
|
| 5 |
+
| ----------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------- |
|
| 6 |
+
| Generatable or reverse engineerable personal data? | No |
|
| 7 |
+
| Personal data used to create this model? | Yes |
|
| 8 |
+
| Was consent obtained for any personal data used? | Unknown (Externally-Sourced) |
|
| 9 |
+
| How often is dataset reviewed? | Before Release |
|
| 10 |
+
| Is a mechanism in place to honor data subject right of access or deletion of personal data? | No |
|
| 11 |
+
| If personal data was collected for the development of the model, was it collected directly by NVIDIA? | Yes |
|
| 12 |
+
| If personal data was collected for the development of the model by NVIDIA, do you maintain or have access to disclosures made to data subjects? | No |
|
| 13 |
+
| If personal data was collected for the development of this AI model, was it minimized to only what was required? | Yes |
|
| 14 |
+
| Was data from user interactions with the AI model (e.g. user input and prompts) used to train the model? | Yes |
|
| 15 |
+
| Is there provenance for all datasets used in training? | Yes |
|
| 16 |
+
| Does data labeling (annotation, metadata) comply with privacy laws? | Yes |
|
| 17 |
+
| Is data compliant with data subject requests for data correction or removal, if such a request was made? | Unknown (Externally-Sourced Data) |
|
| 18 |
+
| Applicable Privacy Policy | [https://www.nvidia.com/en-us/about-nvidia/privacy-policy/](https://www.nvidia.com/en-us/about-nvidia/privacy-policy/) |
|
| 19 |
+
|
| 20 |
+
|
safety.md
ADDED
|
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Safety & Security Subcard
|
| 2 |
+
|
| 3 |
+
|
| 4 |
+
| Field | Response |
|
| 5 |
+
| ----------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
|
| 6 |
+
| Model Application Field(s): | Large Language Model-based Content Safety & Moderation |
|
| 7 |
+
| Describe the life-critical impact (if present). | Not Applicable |
|
| 8 |
+
| Use Case Restrictions: | Use of the model is governed by the [OpenMDW License Agreement, version 1.1](https://raw.githubusercontent.com/OpenMDW/OpenMDW/refs/heads/main/1.1/LICENSE.OpenMDW-1.1) (OpenMDW-1.1), [Gemma Terms of Use](https://ai.google.dev/gemma/terms) and [Gemma Prohibited Use Policy](https://ai.google.dev/gemma/prohibited_use_policy). |
|
| 9 |
+
| 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. |
|
| 10 |
+
|
| 11 |
+
|