october-finetuning-more-variables-sweep-20251012-202343-t08
Slur reclamation binary classifier
Task: LGBTQ+ reclamation vs non-reclamation use of harmful words on social media text.
Trial timestamp (UTC): 2025-10-12 20:23:43
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 | 1.5 |
| 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 | 0.6966881607229021 |
| f1_weighted_dev_0.5 | 0.8396412101658748 |
| accuracy_dev_0.5 | 0.8285077951002228 |
| f1_macro_dev_best_global | 0.7184728843416901 |
| f1_weighted_dev_best_global | 0.861020415912312 |
| accuracy_dev_best_global | 0.8596881959910914 |
| f1_macro_dev_best_by_lang | 0.7297459973516311 |
| f1_weighted_dev_best_by_lang | 0.8726099195059952 |
| accuracy_dev_best_by_lang | 0.8775055679287305 |
| default_threshold | 0.5 |
| best_threshold_global | 0.65 |
| thresholds_by_lang | {"en": 0.65, "it": 0.6, "es": 0.8} |
Thresholds
- Default:
0.5 - Best global:
0.65 - Best by language:
{ "en": 0.65, "it": 0.6, "es": 0.8 }
Detailed evaluation
Classification report @ 0.5
precision recall f1-score support
no-recl (0) 0.9278 0.8675 0.8966 385
recl (1) 0.4270 0.5938 0.4967 64
accuracy 0.8285 449
macro avg 0.6774 0.7306 0.6967 449
weighted avg 0.8564 0.8285 0.8396 449
Classification report @ best global threshold (t=0.65)
precision recall f1-score support
no-recl (0) 0.9215 0.9143 0.9179 385
recl (1) 0.5075 0.5312 0.5191 64
accuracy 0.8597 449
macro avg 0.7145 0.7228 0.7185 449
weighted avg 0.8625 0.8597 0.8610 449
Classification report @ best per-language thresholds
precision recall f1-score support
no-recl (0) 0.9167 0.9429 0.9296 385
recl (1) 0.5849 0.4844 0.5299 64
accuracy 0.8775 449
macro avg 0.7508 0.7136 0.7297 449
weighted avg 0.8694 0.8775 0.8726 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 | 0.8636 | 0.5429 | 0.8612 | 0.5446 | 0.5415 | 0.8587 | 0.8636 |
| it | 163 | 0.8834 | 0.7980 | 0.8794 | 0.8216 | 0.7799 | 0.8779 | 0.8834 |
| es | 132 | 0.8864 | 0.7530 | 0.8795 | 0.7906 | 0.7277 | 0.8770 | 0.8864 |
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-20251012-202343-t08"
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.