Input: Tired Nature's sweet restorer, balmy Sleep!
Output: IY1 P This is a line-level labeling task (not stanza generation and not full combined-structure prediction).

Output Format

A short phonological ending sequence, e.g.:

IY1 P
EY1 Z
OW1
AY1 N D

Training Data Format

The rhyme-only training set uses:

input: poetic line text
target: rhyme phonology string only Example:
input: He, like the world, his ready visit pays
target: EY1 Z

Model Details

Developer: KemiOm
Base model: google/flan-t5-large
Method: LoRA fine-tuning
Framework: Hugging Face Transformers + PEFT

Best Run Configuration

epochs: 5.0
learning rate: 5e-05 (round2 setting)
batch size per device: 8
gradient accumulation: 2
max input length: 384
max target length: 128
LoRA rank (r): 32
LoRA alpha: 64
LoRA dropout: 0.05
LoRA target modules: q,k,v,o
seed: 42

Intended Use

Rhyme-phonology annotation for poetry lines
Rhyme diagnostics in constrained poetry pipelines
Preprocessing/evaluation for poetry generation systems

Limitations

Depends on learned orthography-to-phonology mapping; rare/archaic spellings may fail.
Output reflects dataset conventions and may not capture all dialectal pronunciations.
Best used with human review in literary-critical workflows.

Usage

python

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
model_id = "KemiOm/poetry-rhyme-best"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSeq2SeqLM.from_pretrained(model_id)
line = "Tired Nature's sweet restorer, balmy Sleep!"
inputs = tokenizer(line, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=12, do_sample=False)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
# expected format: "IY1 P"