README.md

metadata

model_creator: Nekochu
quantized_by: Nekochu
model_name: Luminia 13B v3
pretty_name: Luminia
model_type: llama2
prompt_template: >-
  Below is an instruction that describes a task. Write a response that
  appropriately completes the request. ### Instruction: {Instruction} {summary}
  ### input: {category} ### Response: {prompt}
base_model: meta-llama/Llama-2-13b-chat-hf
library_name: peft
license: apache-2.0
datasets:
  - Nekochu/discord-unstable-diffusion-SD-prompts
  - glaiveai/glaive-function-calling-v2
  - TIGER-Lab/MathInstruct
  - Open-Orca/SlimOrca
  - GAIR/lima
  - sahil2801/CodeAlpaca-20k
  - garage-bAInd/Open-Platypus
language:
  - en
pipeline_tag: text-generation
task_categories:
  - question-answering
  - text2text-generation
  - conversational
inference: true
widget:
  - example_title: prompt assistant
    messages:
      - role: system
        content: >-
          Below is an instruction that describes a task, paired with an input
          that provides further context. Write a response that appropriately
          completes the request.
      - role: user
        content: >
          ### Instruction:

          Create stable diffusion metadata based on the given english
          description. Luminia

          ### Input:

          favorites and popular SFW

          ### Response:
    output:
      text: Luminia, 1girl, solo, blonde hair, long hair,
tags:
  - llama-factory
  - lora
  - generated_from_trainer
  - llama2
  - llama
  - instruct
  - finetune
  - gpt4
  - synthetic data
  - stable diffusion
  - alpaca
  - llm
model-index:
  - name: Luminia-13B-v3
    results: []

Luminia v3 is good at reasoning to enhance Stable Diffusion prompt from short summary description, may output NSFW content.

LoRa is include and Quants: exllamav2 2.4bpw-h6, 4.25bpw-h6, 8.0bpw-h8 | GGUF Q4_K_M, IQ4_NL |

Prompt template: Alpaca

Output example tested In text-generation-webui

Input	base llama-2-chat	QLoRa
[question]: Create stable diffusion metadata based on the given english description. Luminia \n### Input:\n favorites and popular SFW	Answer: Luminia, a mystical world of wonder and magic 🧝‍♀️✨ A place where technology and nature seamlessly blend together ...	Answer! < lora:Luminari-10:0.8> Luminari, 1girl, solo, blonde hair, long hair, blue eyes, (black dress), looking at viewer, night sky, starry sky, constellation, smile, upper body, outdoors, forest, moon, tree, mountain, light particle ....

Output prompt from QLoRa to A1111/SD-WebUI:

Training Details

Click to see details

Model Description

• Train by: Nekochu, Model type: Llama, Finetuned from model Llama-2-13b-chat

• Continue from the base of LoRA Luminia-13B-v2-QLora

  Know issue: [issue]

Trainer

• hiyouga/LLaMA-Efficient-Tuning

  Hardware: QLoRA training OS Windows, Python 3.10.8, CUDA 12.1 on 24GB VRAM.

Training hyperparameters

The following hyperparameters were used during training:

• num_epochs: 1.0
• finetuning_type: lora
• quantization_bit: 4
• stage: sft
• learning_rate: 5e-05
• cutoff_len: 4096
• num_train_epochs: 3.0
• max_samples: 100000
• warmup_steps: 0
• train_batch_size: 1
• distributed_type: single-GPU
• num_devices: 1
• warmup_steps: 0
• rope_scaling: linear
• lora_rank: 32
• lora_target: all
• lora_dropout: 0.15
• bnb_4bit_compute_dtype: bfloat16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine

training_loss:

Framework versions

• PEFT 0.9.0
• Transformers 4.38.1
• Pytorch 2.1.2+cu121
• Datasets 2.14.5
• Tokenizers 0.15.0