You can deploy this model and start earning money today!

arielcerdap/
modernbert-disfluency-expC-large-realonly

arielcerdap/disfluency-fluencybank

Views

No views yet

ModernBERT-large Disfluency Detection — Exp C (Real Only)

Fine-tuned from answerdotai/ModernBERT-large on real-only FluencyBank Timestamped data. Identical setup to Exp A (base) to isolate architecture effect.

Dataset

Config: real_only de arielcerdap/disfluency-fluencybank
Train: 2737 segmentos (100% reales)
Val/Test: idénticos a Exp A y B para comparación directa

Labels

O · FP (filled pause) · RP (repetition) · RV (revision) · PW (partial word)

Test Results vs Exp A (base)

Label	Exp A (base)	Exp C (large)
FP	0.9944	0.9944
RP	0.8022	0.8964
RV	0.3145	0.4974
PW	0.8879	0.9451
Macro (dis)	0.7497	0.8333
Binary F1	0.8902	0.9250

Per-class Detail

Label	P	R	F1	Support
O	0.9892	0.9846	0.9869	3704
FP	0.9888	1.0000	0.9944	176
RP	0.8743	0.9195	0.8964	174
RV	0.4563	0.5465	0.4974	86
PW	0.9685	0.9227	0.9451	233

Hyperparameters

learning_rate: 5e-05
batch_size effective: 32 (8 × 4 grad_accum)
epochs: 15
warmup_steps: 191
weight_decay: 0.1
classifier_dropout: 0.3
focal_loss_gamma: 3.0 (adaptive)
class_weights: O=1.0, FP=3.0, RP=6.0, RV=20.0, PW=5.0

arielcerdap/modernbert-disfluency-expC-large-realonly