Stoicheia -- macronization and metrical scansion (joint)
Stoicheia is a 405M-parameter character-level masked-diffusion encoder for Ancient Greek
(d_model 1024, depth 32, banded attention: three of every four blocks attend within a
256-character window, the fourth globally). Its input is factored into five aligned planes --
letters, word/sentence boundaries, diacritics, capitalization, punctuation -- each of which can
be masked independently to an explicit unknown state at inference. That is what lets one model
read an edited text, scriptio continua, and a lacuna of unknown length without changing
anything but its input.
Anonymous release accompanying a paper under review.
Two per-letter heads on Stoicheia-doc_clean, trained jointly: macronization (long versus
short at ambiguous bare α/ι/υ -- Greek orthography never marks vowel length) and scansion
(none / heavy-end / light-end / verse-end). An optional Viterbi decoder constrains the scansion
output to valid paths through a set of metre automata.
Use this checkpoint when you want both tasks from one model; use Stoicheia-macronizer when you
want vowel length alone, where a dedicated head does better.
Usage
import torch
from transformers import AutoModel
from huggingface_hub import hf_hub_download
REPO = "Ericu950/Stoicheia-meter"
model = AutoModel.from_pretrained(REPO, trust_remote_code=True).eval()
hf_hub_download(repo_id=REPO, filename="processing_char_bert_meter.py", local_dir=".")
from processing_char_bert_meter import CharBertMeterProcessor
proc = CharBertMeterProcessor()
batch = proc("ἄνδρα μοι ἔννεπε, μοῦσα, πολύτροπον, ὃς μάλα πολλὰ")
with torch.no_grad():
out = model(**{k: v for k, v in batch.items() if not k.startswith("_")})
print(proc.decode_macronization(out, batch)) # _ long, ^ short
print(proc.decode_scansion(out, batch)) # [heavy] {light}