SimoneAstarita/trilingual-no-bio-20251012-205606-t21

october-finetuning-more-variables-sweep-20251012-205606-t13

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

Trial timestamp (UTC): 2025-10-12 20:56:06

Data case: en-es-it

Configuration (trial hyperparameters)

Model: Alibaba-NLP/gte-multilingual-base

Hyperparameter	Value
LANGUAGES	en-es-it
LR	2e-05
EPOCHS	3
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	0.5
TEXT_NORMALIZE	True

Dev set results (summary)

Metric	Value
f1_macro_dev_0.5	0.7148663237442052
f1_weighted_dev_0.5	0.8397177643197989
accuracy_dev_0.5	0.821826280623608
f1_macro_dev_best_global	0.7349841198459766
f1_weighted_dev_best_global	0.864951679724489
accuracy_dev_best_global	0.8596881959910914
f1_macro_dev_best_by_lang	0.7336960726166224
f1_weighted_dev_best_by_lang	0.8575109332646288
accuracy_dev_best_by_lang	0.8463251670378619
default_threshold	0.5
best_threshold_global	0.6
thresholds_by_lang	{"en": 0.5, "it": 0.55, "es": 0.6}

Thresholds

• Default: 0.5
• Best global: 0.6
• Best by language: { "en": 0.5, "it": 0.55, "es": 0.6 }

Detailed evaluation

Classification report @ 0.5

              precision    recall  f1-score   support

 no-recl (0)     0.9499    0.8364    0.8895       385
    recl (1)     0.4273    0.7344    0.5402        64

    accuracy                         0.8218       449
   macro avg     0.6886    0.7854    0.7149       449
weighted avg     0.8754    0.8218    0.8397       449

Classification report @ best global threshold (t=0.60)

              precision    recall  f1-score   support

 no-recl (0)     0.9328    0.9013    0.9168       385
    recl (1)     0.5065    0.6094    0.5532        64

    accuracy                         0.8597       449
   macro avg     0.7196    0.7553    0.7350       449
weighted avg     0.8720    0.8597    0.8650       449

Classification report @ best per-language thresholds

              precision    recall  f1-score   support

 no-recl (0)     0.9438    0.8727    0.9069       385
    recl (1)     0.4731    0.6875    0.5605        64

    accuracy                         0.8463       449
   macro avg     0.7085    0.7801    0.7337       449
weighted avg     0.8767    0.8463    0.8575       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	0.8117	0.6081	0.8429	0.5925	0.6877	0.8910	0.8117
it	163	0.8773	0.8056	0.8787	0.7987	0.8132	0.8805	0.8773
es	132	0.8485	0.7533	0.8601	0.7255	0.8080	0.8827	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-20251012-205606-t13"
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.