Model Type | Causal decoder-only, Instruct model |
|
Use Cases |
Areas: | Research, Commercial applications |
|
Applications: | Text generation, Chatbot systems |
|
Primary Use Cases: | Ready-to-use chat model, Instruction generation |
|
Limitations: | Mostly trained on English data, May carry stereotypes and biases from web data |
|
Considerations: | Develop guardrails for production use. |
|
|
Additional Notes | This is a specialized instruct model. For fine-tuning, consider starting from the base Falcon-40B model. |
|
Supported Languages | English (High), French (Moderate) |
|
Training Details |
Data Sources: | |
Data Volume: | |
Methodology: | Finetuned on chat dataset with 5% RefinedWeb data |
|
Context Length: | |
Hardware Used: | 64 A100 40GB GPUs in P4d instances |
|
Model Architecture: | Causal decoder-only with optimized architecture |
|
|
Input Output |
Input Format: | |
Accepted Modalities: | |
Output Format: | |
Performance Tips: | Ensure sufficient VRAM and optimize inference settings accordingly. |
|
|
Release Notes |
Notes: | Includes various quantized model files for different inference needs. |
|
|
|