SimoneAstarita/trilingual-no-bio-20251012-204304-t19

october-finetuning-more-variables-sweep-20251012-204304-t11

Slur reclamation binary classifier
Task: LGBTQ+ reclamation vs non-reclamation use of harmful words on social media text.

Trial timestamp (UTC): 2025-10-12 20:43:04

Data case: en-es-it

Configuration (trial hyperparameters)

Model: Alibaba-NLP/gte-multilingual-base

Hyperparameter	Value
LANGUAGES	en-es-it
LR	1e-05
EPOCHS	5
MAX_LENGTH	256
USE_BIO	False
USE_LANG_TOKEN	False
GATED_BIO	False
FOCAL_LOSS	True
FOCAL_GAMMA	2.5
USE_SAMPLER	True
R_DROP	True
R_KL_ALPHA	1.0
TEXT_NORMALIZE	True

Dev set results (summary)

Metric	Value
f1_macro_dev_0.5	0.6706772784019975
f1_weighted_dev_0.5	0.8127803080225442
accuracy_dev_0.5	0.7906458797327395
f1_macro_dev_best_global	0.722815063056707
f1_weighted_dev_best_global	0.8697842222606763
accuracy_dev_best_global	0.8752783964365256
f1_macro_dev_best_by_lang	0.7045809688296735
f1_weighted_dev_best_by_lang	0.8565143987671885
accuracy_dev_best_by_lang	0.8574610244988864
default_threshold	0.5
best_threshold_global	0.55
thresholds_by_lang	{"en": 0.5, "it": 0.55, "es": 0.55}

Thresholds

• Default: 0.5
• Best global: 0.55
• Best by language: { "en": 0.5, "it": 0.55, "es": 0.55 }

Detailed evaluation

Classification report @ 0.5

              precision    recall  f1-score   support

 no-recl (0)     0.9343    0.8130    0.8694       385
    recl (1)     0.3684    0.6562    0.4719        64

    accuracy                         0.7906       449
   macro avg     0.6514    0.7346    0.6707       449
weighted avg     0.8537    0.7906    0.8128       449

Classification report @ best global threshold (t=0.55)

              precision    recall  f1-score   support

 no-recl (0)     0.9144    0.9429    0.9284       385
    recl (1)     0.5769    0.4688    0.5172        64

    accuracy                         0.8753       449
   macro avg     0.7456    0.7058    0.7228       449
weighted avg     0.8663    0.8753    0.8698       449

Classification report @ best per-language thresholds

              precision    recall  f1-score   support

 no-recl (0)     0.9147    0.9195    0.9171       385
    recl (1)     0.5000    0.4844    0.4921        64

    accuracy                         0.8575       449
   macro avg     0.7074    0.7019    0.7046       449
weighted avg     0.8556    0.8575    0.8565       449

Per-language metrics (at best-by-lang)

lang	n	acc	f1_macro	f1_weighted	prec_macro	rec_macro	prec_weighted	rec_weighted
en	154	0.8442	0.4959	0.8442	0.4959	0.4959	0.8442	0.8442
it	163	0.8773	0.7903	0.8740	0.8077	0.7761	0.8722	0.8773
es	132	0.8485	0.7169	0.8514	0.7091	0.7259	0.8548	0.8485

Data

• Train/Dev: private multilingual splits with ~15% stratified Dev (by (lang,label)).
• Source: merged EN/IT/ES data with bios retained (ignored if unused by model).

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification, AutoConfig
import torch, numpy as np

repo = "SimoneAstarita/october-finetuning-more-variables-sweep-20251012-204304-t11"
tok = AutoTokenizer.from_pretrained(repo)
cfg = AutoConfig.from_pretrained(repo)
model = AutoModelForSequenceClassification.from_pretrained(repo)

texts = ["example text ..."]
langs = ["en"]

mode = "best_global"  # or "0.5", "by_lang"

enc = tok(texts, truncation=True, padding=True, max_length=256, return_tensors="pt")
with torch.no_grad():
    logits = model(**enc).logits
probs = torch.softmax(logits, dim=-1)[:, 1].cpu().numpy()

if mode == "0.5":
    th = 0.5
    preds = (probs >= th).astype(int)
elif mode == "best_global":
    th = getattr(cfg, "best_threshold_global", 0.5)
    preds = (probs >= th).astype(int)
elif mode == "by_lang":
    th_by_lang = getattr(cfg, "thresholds_by_lang", {})
    preds = np.zeros_like(probs, dtype=int)
    for lg in np.unique(langs):
        t = th_by_lang.get(lg, getattr(cfg, "best_threshold_global", 0.5))
        preds[np.array(langs) == lg] = (probs[np.array(langs) == lg] >= t).astype(int)
print(list(zip(texts, preds, probs)))

Additional files

reports.json: all metrics (macro/weighted/accuracy) for @0.5, @best_global, and @best_by_lang. config.json: stores thresholds: default_threshold, best_threshold_global, thresholds_by_lang. postprocessing.json: duplicate threshold info for external tools.