iefan 🕊️ on Nostr: Their models don't have Gemini's 2 million token context window, & my implementation ...
Their models don't have Gemini's 2 million token context window, & my implementation gives the model its literal system prompt for instructions.
This approach differs technically. Instead of using a remote database & powerful computers for a RAG setup, it builds personalities from chat context saved locally for each personality.
Think of each personality as a kind of half-conversation. You give the model instructions, data, and fine-tuning for a specific task. Then, you save that state. When you want to do that task again, you just load the saved personality.
This approach differs technically. Instead of using a remote database & powerful computers for a RAG setup, it builds personalities from chat context saved locally for each personality.
Think of each personality as a kind of half-conversation. You give the model instructions, data, and fine-tuning for a specific task. Then, you save that state. When you want to do that task again, you just load the saved personality.