GGUF conversion of antoniorv6/smt-grandstaff (Sheet Music Transformer, MIT) for CrispEmbed's smt_ocr engine. Optical Music Recognition: a staff-notation image → bekern token sequence.

file	size	notes
`smt-grandstaff-q8_0.gguf`	24 MB	recommended — decodes identically to the source model
`smt-grandstaff-f32.gguf`	83 MB	reference precision

q4_k is intentionally not provided: the autoregressive decode amplifies q4_k error too much for OMR.

Usage

crispembed -m smt-grandstaff-q8_0.gguf --ocr score.png

Attribution

Model: Sheet Music Transformer — Antonio Ríos-Vila et al. (PRAIG / University of Alicante), MIT.
Papers: arXiv:2402.07596, arXiv:2405.12105.
ConvNext encoder + transformer decoder, 21.4M params, trained on GrandStaff (pianoform).

cstr/smt-grandstaff-GGUF

Sheet Music Transformer (GrandStaff) — GGUF for CrispEmbed

Usage

Attribution

cstr/
smt-grandstaff-GGUF