mrm8488's picture
Update README.md
ab1ebef
metadata
license: bigscience-bloom-rail-1.0
tags:
  - generated_from_trainer
  - stable-diffusion
  - diffusion
model-index:
  - name: bloom-560m-finetuned-sd-prompts
    results: []
datasets:
  - Gustavosta/Stable-Diffusion-Prompts
widget:
  - text: '<s>Prompt: young, curly haired, redhead Natalie Portman as a'
  - text: '<s>Prompt: a powerful energy woman, by alexander fedosav'
inference:
  parameters:
    eos_token_id: 2
    max_length: 128

bloom-560m-finetuned-sd-prompts

This model is a fine-tuned version of bigscience/bloom-560m on the Gustavosta/Stable-Diffusion-Prompts dataset. It achieves the following results on the evaluation set:

  • Loss: 0.8742

Example of usage

import torch
from transformers import BloomTokenizerFast, BloomForCausalLM

device = 'cuda' if torch.cuda.is_available() else 'cpu'
ckpt = 'mrm8488/bloom-560m-finetuned-sd-prompts' 

tokenizer = BloomTokenizerFast.from_pretrained(ckpt)
model = BloomForCausalLM.from_pretrained(ckpt).to(device)

def generate_prompt(text):
    inputs = tokenizer(text, return_tensors='pt')
    input_ids = inputs.input_ids.to(device)
    attention_mask = inputs.attention_mask.to(device)
    output = model.generate(input_ids, attention_mask=attention_mask, repetition_penalty=1.05, max_length=2048, eos_token_id=tokenizer.eos_token_id)

    return tokenizer.decode(output[0], skip_special_tokens=False)
    
text = "<s>Prompt: pikachu dinning in the eiffel tower"

generate_prompt(text)

# Output: <s>Prompt: pikachu dinning in the eiffel tower, intricate, elegant, highly detailed, digital painting, artstation, concept art, smooth, sharp focus, illustration, art by artgerm and greg rutkowski and alphonse mucha</s>

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 5e-05
  • train_batch_size: 1
  • eval_batch_size: 1
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 4
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 2
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss
2.6743 0.17 100 2.0891
1.8919 0.33 200 1.7191
1.5907 0.5 300 1.4454
1.3865 0.67 400 1.3247
1.2487 0.83 500 1.2150
1.1565 1.0 600 1.1031
0.896 1.17 700 1.0612
0.8389 1.33 800 0.9994
0.8071 1.5 900 0.9530
0.7628 1.67 1000 0.9206
0.7423 1.83 1100 0.8883
0.7155 2.0 1200 0.8742

Framework versions

  • Transformers 4.22.1
  • Pytorch 1.12.1+cu113
  • Datasets 2.5.1
  • Tokenizers 0.12.1