Model Type | text-generation-inference, question-answering |
|
Use Cases |
Areas: | |
Applications: | Natural language generation tasks |
|
Primary Use Cases: | Assistant-like chat, Natural language generation tasks |
|
Limitations: | Use in languages beyond those explicitly referenced as supported |
|
Considerations: | Additional languages support requires fine-tuning and compliance with the license terms. |
|
|
Additional Notes | Model training was sped up using Unsloth and Huggingface's TRL library. |
|
Supported Languages | English (High proficiency), German (High proficiency), French (High proficiency), Italian (High proficiency), Portuguese (High proficiency), Hindi (High proficiency), Spanish (High proficiency), Thai (High proficiency) |
|
Training Details |
Data Sources: | A new mix of publicly available online data |
|
Data Volume: | |
Methodology: | Supervised Fine tuning (SFT) and Reinforcement Learning with Human Feedback (RLHF) |
|
Context Length: | |
Model Architecture: | Llama 3.1 - auto-regressive language model with transformer architecture |
|
|
Safety Evaluation |
Methodologies: | red-teaming, adversarial tests |
|
Findings: | Potential outputs cannot be predicted in advance and may include inaccurate or biased responses. |
|
Risk Categories: | |
Ethical Considerations: | Developers should perform safety testing and tuning tailored to their specific model applications. |
|
|
Responsible Ai Considerations |
Fairness: | Efforts to avoid biases through diverse language support. |
|
Transparency: | Details on data sources and training methods provided. |
|
Accountability: | Developers encouraged to implement system safeguards before deploying. |
|
Mitigation Strategies: | Includes safety fine-tuning and synthesis of quality data. |
|
|
Input Output |
Input Format: | |
Accepted Modalities: | |
Output Format: | Multilingual Text and code |
|
Performance Tips: | Usage with the 'transformers' library as specified in instructions. |
|
|
Release Notes |
Version: | |
Date: | |
Notes: | New capabilities include a longer context window, multilingual inputs and outputs. |
|
|
|