| Model Type | | text-to-text, decoder-only, large language model |
|
| Use Cases |
| Areas: | | Various industries and domains |
|
| Applications: | | Content Creation and Communication, Research and Education |
|
| Primary Use Cases: | | Text Generation, Chatbots and Conversational AI, Text Summarization, NLP Research, Language Learning, Knowledge Exploration |
|
| Limitations: | | Bias and Fairness, Misinformation and Misuse, Lack of Common Sense, Factual Inaccuracy |
|
| Considerations: | | LLMs performance is heavily dependent on the quality of input prompts and the context length. |
|
|
| Additional Notes | | This description is based on the specified version, other iteration details can be found in technical documents. |
|
| Supported Languages | | English (available for text generation, question answering, summarization, and reasoning tasks) |
|
| Training Details |
| Data Sources: | | Web Documents, Code, Mathematics |
|
| Data Volume: | |
| Methodology: | | Rigorous CSAM filtering, Sensitive Data Filtering, filtering based on content quality and safety |
|
| Context Length: | |
| Hardware Used: | |
| Model Architecture: | |
|
| Safety Evaluation |
| Methodologies: | | Red-teaming, Structured evaluations |
|
| Risk Categories: | | Text-to-Text Content Safety, Text-to-Text Representational Harms, Memorization, Large-scale harm |
|
| Ethical Considerations: | | Models evaluated against a number of different categories relevant to ethics and safety, include Text-to-Text Content Safety, Representational Harms, potential data memorization, and dangerous capability tests. |
|
|
| Responsible Ai Considerations |
| Fairness: | | These models underwent careful scrutiny and input data pre-processing with posterior evaluations reported in this card. |
|
| Transparency: | | The model card summarizes details on the models' architecture, capabilities, limitations, and evaluation processes. |
|
| Accountability: | | Google is accountable for the use of the model under its terms of service and policies. |
|
| Mitigation Strategies: | | Developers are encouraged to monitor and report misuse, employ de-biasing techniques, implement content safety safeguards, and adhere to privacy regulations. |
|
|
| Input Output |
| Input Format: | |
| Accepted Modalities: | |
| Output Format: | | Generated English-language text |
|
| Performance Tips: | | Ensure to use correct input formats for fine-tuning or inference, use optimizations for specific hardware and quantization methods. |
|
|
| Release Notes |
| Version: | |
| Notes: | | Contains updates and new numbers for the IT version models, surpasses previous versions across various benchmarks. |
|
|
|