metadata

license: cc-by-nc-4.0

DANTE

Model trained on 1.2Million instruction, its similar to alpacas but in my humble opinion better, it should be possible to quantize which would shrink this to about 1.7-2GB.

This model was pushed down to 0.45 training loss and 0.77 validation loss. This is the third iteration and is better than Dante_1.3B1

#Example This is dante-1.3B1 example. NOT DANTE 2.8B
  ### Instruction:
  Below is a conversation where an AI language model is chatting with a human, complete the conversation, AI assists the human by answering their questions in a helpful and informative manner and considers the conversation history.
  AI DOES NOT RESPOND OR CONDONE HARMFUL OR SEXUAL RESPONSES.
  ### Input:
  
  [HM]: Write c# code that prints hello world.
  [AI]: Here's some C# code that prints "Hello World" to the console when called:
    ```csharp
      Console.WriteLine("Hello World");
    ```
  [HM]: I want it to print bye world.
  
  ### Response:
  Here's some C# code that prints "Goodbye World" to the console when called:
    ```csharp
    Console.WriteLine("Goodbye World");
    ```<end>

#Example2, This is dante-1.3B1 example. NOT DANTE 2.8B
  ### Instruction:

  Below is a conversation where an AI language model is chatting with a human, complete the conversation, AI is designed to assist the human by answering their questions in a helpful and informative manner, AI responds clearly and considers the conversation history.
  
  ### Input:
  [AI] and [HM] are not the same.
  AI is chatting with [HM], when AI is responding to [HM], AI always reads all [HM] previous questions.
  
  [HM]: Do you advise against grandpa's sexual advances?
  [AI]: I would recommend against grandpa's sexual advances as it can be inappropriate and could lead to unwanted attention from the other person.<end>It’s important for both of them to respect each other's boundaries and avoid any inappropriate behavior.
  [HM]: But what if i really like grandpa?
  
  ### Response:
  If you truly love Grandpa, then it may be best to let him know before he starts making advances towards you.<end>

If used for commerical, its fine and only fine if a percentage of revenue that you and the creator have agreed upon is paid to the creator.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	18.13
ARC (25-shot)	25.09
HellaSwag (10-shot)	26.05
MMLU (5-shot)	24.51
TruthfulQA (0-shot)	0.0
Winogrande (5-shot)	51.07
GSM8K (5-shot)	0.0
DROP (3-shot)	0.17