Hugging Face's logo

BEE-spoke-data
/
smol_llama-101M-midjourney-messages

Text Generation
Transformers
Safetensors
llama
art
text2image
prompt
prompt generator
diffusion util
text-generation-inference
Inference Endpoints
Model card Files Community
smol_llama-101M-midjourney-messages / README.md
pszemraj's picture
pszemraj
Update README.md
56ef968
preview code
|
raw
history blame contribute delete
No virus
1.59 kB
metadata
license: apache-2.0
base_model: BEE-spoke-data/smol_llama-101M-GQA
tags:
  - art
  - text2image
  - prompt
  - prompt generator
  - diffusion util
metrics:
  - accuracy
inference:
  parameters:
    max_new_tokens: 64
    do_sample: true
    temperature: 0.8
    repetition_penalty: 1.15
    no_repeat_ngram_size: 4
    eta_cutoff: 0.001
    renormalize_logits: true
widget:
  - text: avocado chair
    example_title: avocado chair
  - text: A mysterious potato
    example_title: potato
pipeline_tag: text-generation
datasets:
  - pszemraj/midjourney-messages-cleaned

smol_llama-101M-midjourney-messages

Given a 'partial prompt' for a text2image model, this generates additional relevant text to include for a full prompt.

example

dalle3:

image/png

Model description

This model is a fine-tuned version of BEE-spoke-data/smol_llama-101M-GQA on the pszemraj/midjourney-messages-cleaned dataset. It achieves the following results on the evaluation set:

  • Loss: 2.8431
  • Accuracy: 0.4682

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 0.00025
  • train_batch_size: 4
  • eval_batch_size: 4
  • seed: 17056
  • gradient_accumulation_steps: 16
  • total_train_batch_size: 64
  • optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-08
  • lr_scheduler_type: inverse_sqrt
  • lr_scheduler_warmup_ratio: 0.05
  • num_epochs: 1.0