teknium's picture
Fixed incorrect numbers
74a4098
metadata
license: mit

Base Mode: Llama 7B

Llama DEUS v3 is the largest dataset I've trained on yet, including:

GPTeacher - General Instruct - Code Instruct - Roleplay Instruct
My unreleased Roleplay V2 Instruct
GPT4-LLM Uncensored + Unnatural Instructions
WizardLM Uncensored
CamelAI's 20k Biology, 20k Physics, 20k Chemistry, and 50k Math GPT4 Datasets
CodeAlpaca

This model was trained for 4 epochs over 1 day of training, it's a rank 128 LORA that targets attention heads, LM_Head, and MLP layers

Prompt format:

### Instruction:
<prompt>
  
### Response:
  

or

### Instruction:
<prompt>

### Input:
<input>
  
### Response: