CoolFace
Apppublic

TomatitoToho/nanbeige-dc-trainer

sourceHugging Faceapache-2.0updated 4mo agoView on Hugging Face
0likes
App README

Nanbeige-DC Training Space

Fine-tune Nanbeige4.1-3B on Discord bot scenarios using QLoRA for the Zelin DC bot.

Usage

  1. 1.Enter your HuggingFace write token
  2. 2.Set target repository (default: TomatitoToho/Nanbeige-DC-v1)
  3. 3.Adjust training parameters if needed
  4. 4.Click Start Training
  5. 5.Wait for training to complete (~15-20 min)

Technical Details

  • —Base Model: Nanbeige/Nanbeige4.1-3B (3B params, LLaMA-based, Apache 2.0)
  • —Method: QLoRA (4-bit NF4 quantization + LoRA adapters)
  • —Training Data: 3000+ Discord bot scenarios (personality, moderation, commands, tech knowledge)
  • —Format: ChatML (<|im_start|>user\n...<|im_end|>)