More actions
No edit summary |
No edit summary |
||
| (One intermediate revision by the same user not shown) | |||
| Line 1: | Line 1: | ||
=== ''' | === '''Can we make AI responses longer? Can we Increase background length or number of keywords?''' === | ||
Increasing the length of the background or the amount of physical traits, identity traits, personality traits, or clothing keywords (which are already set at 17 each) would slow down chat performance for each inference. Our current context is already substantial, including 30 chat history entries, 10 memories (embeddings), 17*4 traits at 80 characters each, plus the role-play and user profiles. This results in a context size of around 2.5K to 4K tokens, depending on the persona and history length. | Increasing the length of the background or the amount of physical traits, identity traits, personality traits, or clothing keywords (which are already set at 17 each) would slow down chat performance for each inference. Our current context is already substantial, including 30 chat history entries, 10 memories (embeddings), 17*4 traits at 80 characters each, plus the role-play and user profiles. This results in a context size of around 2.5K to 4K tokens, depending on the persona and history length. | ||
Given that we're running high-quality language models on high-end GPUs (Tesla H100 80GB), expanding these traits would require more processing power. Unfortunately, we currently don't have the financial resources to increase the number of these GPUs, as Soulkyn is still not profitable at the time of writing. | Given that we're running high-quality language models on high-end GPUs (Tesla H100 80GB), expanding these traits would require more processing power. Unfortunately, we currently don't have the financial resources to increase the number of these GPUs, as Soulkyn is still not profitable at the time of writing. | ||
=== Image generation with no persona or 100% free prompts === | |||
We offer image generation specifically for characters because Soulkyn is focused on creating the most interesting, realistic, and complex characters possible. Providing a standalone image generator would significantly increase the number of images generated, which is already high. This would require a separate service with its own pricing, settings, and quotas to cover the costs. | |||
=== Better freemium or options to gain free credits === | |||
https://soulkyn.com/blog/important-update-changes-to-soulkyns-free-usage | |||
Latest revision as of 01:57, 21 August 2024
Can we make AI responses longer? Can we Increase background length or number of keywords?
Increasing the length of the background or the amount of physical traits, identity traits, personality traits, or clothing keywords (which are already set at 17 each) would slow down chat performance for each inference. Our current context is already substantial, including 30 chat history entries, 10 memories (embeddings), 17*4 traits at 80 characters each, plus the role-play and user profiles. This results in a context size of around 2.5K to 4K tokens, depending on the persona and history length.
Given that we're running high-quality language models on high-end GPUs (Tesla H100 80GB), expanding these traits would require more processing power. Unfortunately, we currently don't have the financial resources to increase the number of these GPUs, as Soulkyn is still not profitable at the time of writing.
Image generation with no persona or 100% free prompts
We offer image generation specifically for characters because Soulkyn is focused on creating the most interesting, realistic, and complex characters possible. Providing a standalone image generator would significantly increase the number of images generated, which is already high. This would require a separate service with its own pricing, settings, and quotas to cover the costs.
Better freemium or options to gain free credits
https://soulkyn.com/blog/important-update-changes-to-soulkyns-free-usage