Modelpublic

SimoneAstarita/trilingual-no-bio-20251014-t02

sourceHugging Faceapache-2.0updated 1y agoView on Hugging Face

0likes5downloads

Model Card

trilingual-no-bio-20251014-t02

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

Trial timestamp (UTC): 2025-10-14 08:26:10 Data case: en-es-it

Configuration (trial hyperparameters)

Model: Alibaba-NLP/gte-multilingual-base

Hyperparameter	Value
LANGUAGES	en-es-it
LR	3e-05
EPOCHS	3
MAX_LENGTH	256
USE_BIO	False
USELANGTOKEN	False
GATED_BIO	False
FOCAL_LOSS	True
FOCAL_GAMMA	1.5
USE_SAMPLER	False
R_DROP	True
RKLALPHA	1.0
TEXT_NORMALIZE	True

Dev set results (summary)

Metric	Value
f1macrodev_0.5	0.7268144285093439
f1weighteddev_0.5	0.8707930801849492
accuracydev0.5	0.8752783964365256
f1macrodevbestglobal	0.7420086491063598
f1weighteddevbestglobal	0.8734607329938144
accuracydevbest_global	0.8730512249443207
f1macrodevbestby_lang	0.7452849905948625
f1weighteddevbestby_lang	0.8742565667778062
accuracydevbestbylang	0.8730512249443207
default_threshold	0.5
bestthresholdglobal	0.45000000000000007
thresholdsbylang	{"en": 0.4, "it": 0.45000000000000007, "es": 0.45000000000000007}

Thresholds

—Default: 0.5
—Best global: 0.45000000000000007
—Best by language: { "en": 0.4, "it": 0.45000000000000007, "es": 0.45000000000000007 }

Detailed evaluation

Classification report @ 0.5

text

              precision    recall  f1-score   support

 no-recl (0)     0.9165    0.9403    0.9282       385
    recl (1)     0.5741    0.4844    0.5254        64

    accuracy                         0.8753       449
   macro avg     0.7453    0.7123    0.7268       449
weighted avg     0.8677    0.8753    0.8708       449

Classification report @ best global threshold (t=0.45)

text

              precision    recall  f1-score   support

 no-recl (0)     0.9271    0.9247    0.9259       385
    recl (1)     0.5538    0.5625    0.5581        64

    accuracy                         0.8731       449
   macro avg     0.7405    0.7436    0.7420       449
weighted avg     0.8739    0.8731    0.8735       449

Classification report @ best per-language thresholds

text

              precision    recall  f1-score   support

 no-recl (0)     0.9293    0.9221    0.9257       385
    recl (1)     0.5522    0.5781    0.5649        64

    accuracy                         0.8731       449
   macro avg     0.7408    0.7501    0.7453       449
weighted avg     0.8756    0.8731    0.8743       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	0.8701	0.5481	0.8652	0.5524	0.5450	0.8605	0.8701
it	163	0.9018	0.8407	0.9018	0.8407	0.8407	0.9018	0.9018
es	132	0.8409	0.7187	0.8479	0.7026	0.7420	0.8578	0.8409

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

python

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

repo = "SimoneAstarita/trilingual-no-bio-20251014-t02"
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, @bestglobal, and @bestbylang. config.json: stores thresholds: defaultthreshold, bestthresholdglobal, thresholdsbylang. postprocessing.json: duplicate threshold info for external tools.