metadata
license: apache-2.0
pretty_name: Luminia
model_type: llama2
tags:
- llama-factory
- lora
- generated_from_trainer
- llama2
- llama
- instruct
- finetune
- gpt4
- synthetic data
- stable diffusion
- alpaca
- llm
datasets:
- Nekochu/discord-unstable-diffusion-SD-prompts
- Nekochu/Luminia-mixture
- AstraMindAI/RLAIF-Nectar
- hiyouga/DPO-En-Zh-20k
Training resume from Luminia-13B-v3.
Luminia-v4 Lora only
Luminia-13B-v4-QLora-sft (rank 32) barelly can handle new Luminia-mixture and ExtendedPrompts should give more flexible when ask prompt, e.g.:
### Instruction:
Create stable diffusion prompt based on the given english description.
### Input:
City street, night, raining, drone shot, cyberpunk
### Response:
And so Stage-B DPO: I do NOT recommend use QLora-orpo, poor lora failed to learn more . :<