distilgpt2-magicprompt-SD

Generate/augment your prompt, stable diffusion style.

This model is a fine-tuned version of distilgpt2 on the Gustavosta/Stable-Diffusion-Prompts dataset. It achieves the following results on the evaluation set:

• Loss: 1.3089
• eval_steps_per_second = 17.201
• perplexity = 3.7022

example

Results in (DALL-E, but you get the idea):

this distilgpt2 version is probably small/fast enough to be used locally on CPU!

basic usage

install transformers as needed:

pip install -U transformers

load and query through a pipeline object:

from transformers import pipeline

model_tag = "pszemraj/distilgpt2-magicprompt-SD"
generator = pipeline(
    "text-generation",
    model=model_tag,
)

prompt = "The Answer to Why"
result = generator(
    prompt,
    max_new_tokens=24,
)  # generate, adjust/add kwargs as needed
print(result[0]["generated_text"])

Training and evaluation data

refer to the Gustavosta/Stable-Diffusion-Prompts dataset.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.001
• train_batch_size: 16
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 8
• total_train_batch_size: 256
• total_eval_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 10.0

Training results

Training Loss	Epoch	Step	Validation Loss
2.7061	0.99	33	2.5859
2.08	1.99	66	1.9965
1.7623	2.99	99	1.7248
1.5408	3.99	132	1.5449
1.4147	4.99	165	1.4437
1.3593	5.99	198	1.3768
1.2703	6.99	231	1.3362
1.2528	7.99	264	1.3175
1.1981	8.99	297	1.3091
1.2117	9.99	330	1.3089

Framework versions

• Transformers 4.25.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.6.1
• Tokenizers 0.13.1