Model Type | |
Use Cases |
Areas: | Research, General AI applications |
|
Limitations: | Potential for producing inaccurate, biased, or other objectionable responses., Testing primarily conducted in English., Limited testing in other languages. |
|
|
Additional Notes | Uses quantized formats (e.g., GGML) for varied inference settings. |
|
Supported Languages | English (full support), German (limited), Spanish (limited), French (limited), Italian (limited), Portuguese (limited), Polish (limited), Dutch (limited), Romanian (limited), Czech (limited), Swedish (limited) |
|
Training Details |
Data Sources: | rombodawg/LosslessMegaCodeTrainingV2_1m_Evol_Uncensored, OpenAssistant/oasst1, shahules786/orca-best, argilla/databricks-dolly-15k-curated-multilingual |
|
Methodology: | Fine-tuned in two stages: first on synthetic instructions/coding tasks, then on top human demonstrations. |
|
Context Length: | |
Model Architecture: | Causal decoder-only transformer |
|
|
Responsible Ai Considerations |
Fairness: | Testing conducted primarily in English with limited testing in other languages. Limited coverage of all scenarios. |
|
Mitigation Strategies: | Developers should perform safety testing tailored to specific use cases. |
|
|
Input Output |
Input Format: | Uses OpenAI's chatml standard prompt format |
|
Accepted Modalities: | |
Performance Tips: | For GPU offloading, consider VRAM capacity. |
|
|