N-Bot-Int/OpenElla3-Llama3.2B-V2

Llama3.2B-V2 - OpenElla3B

• Introducing OpenElla3-Llama3.2B-V2, This Model B has an improved prompting and output generation through the use of DPO, We used 5 DPO datasets, aiming to force OpenElla3-Llama3.2B to output more interactive, better Output, and allowing the model to use Emoji, to activate this, please use emojis on your prompts!

• OpenElla3B-V2 Excells in Outputting RAW and UNCENSORED Output And Acknowledges OpenElla3B's weakness for Poor Performance and Roleplaying capabilities, Due to this, the model is re-finetuned with DPO, which solves the issue with OpenElla3B's Bad and Plain Roleplay scenarios, This also allows the Model to Engage in Uncensored response and with appropriate responses, rivaling its older models

• OpenElla3B-V2 contains more Fine-tuned Dataset so please Report any issues found through our email

  [email protected], about any overfitting, or improvements for the future Model **C**, Once again feel free to Modify the LORA to your likings, However please consider Adding this Page for credits and if you'll increase its **Dataset**, then please handle it with care and ethical considerations

• OpenElla3B-V2 is

  • Developed by: N-Bot-Int
  • License: apache-2.0
  • Parent Model from model: unsloth/llama-3.2-3b-instruct-unsloth-bnb-4bit
  • Sequential Trained from Model: N-Bot-Int/OpenElla3-Llama3.2A
  • Dataset Combined Using: Mosher-R1(Propietary Software)

• OpenElla3B Official Metric Score

  • Metrics Made By ItsMeDevRoland Which compares:

    • Deepseek R1 3B GGUF
    • Dolphin 3B GGUF
    • Hermes 3b Llama GGUFF
    • OpenElla3-Llama3.2B GGUFF Which are All Ranked with the Same Prompt, Same Temperature, Same Hardware(Google Colab), To Properly Showcase the differences and strength of the Models

  • THIS MODEL EXCELLS IN LONGER PROMPT AND STAYING IN CHARACTER BUT LAGS BEHIND DEEPSEEK-R1

• Notice

  • For a Good Experience, Please use
    • Low temperature 1.5, min_p = 0.1 and max_new_tokens = 128

• Detail card:

  • Parameter

    • 3 Billion Parameters
    • (Please visit your GPU Vendor if you can Run 3B models)

  • Training

    • 500 steps
      • Mixed-RP Startup Dataset
    • 200 steps
      • PIPPA-ShareGPT for Increased Roleplaying capabilities
    • 150 steps(Re-fining)
      • PIPPA-ShareGPT to further increase weight of PIPPA and to override the noises
    • 500 steps(Lower LR)
      • Character-roleplay-DO to further encourage the Model to respond appropriately with the RP scenario

  • Finetuning tool:

  • Unsloth AI

    • This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.

  • Fine-tuned Using:

  • Google Colab