llama3-7b-en-ar-v2

LLaMA-3 7B pretrained on English-Arabic bilingual data, 134k steps (injection fix v2).

Validation Results

Final checkpoint: step 133,600

Validation set	Cross-entropy loss (nats)	Perplexity
English (en)	2.2718	9.70
AR (ar)	2.8939	18.06

Note: Rerun with corrected injection config (Buckwalter transliteration fix + injection count fix).

Evaluation Results

EEE-format evaluation results are stored under eval_results/eee/ in this repository. Tasks: Global MMLU (EN/AR), PIQA, ECLeKTic, Fictive Entity (2-rate mix).

Citation

Part of the The-CoLab multilingual-transfer collection.

Training Curves

Per-step training and validation loss curves are stored in training_curves/:

File	Description
train_loss.csv	Training loss every 10 steps (step,train_loss)
val_loss.csv	Validation loss per language (step,lang,val_loss)
injection_counts_*.json	Per-entity injection counts at step 133,600