llama3-7b-en-ru-v2
LLaMA-3 7B pretrained on English-Russian bilingual data, 134k steps (injection fix v2).
Validation Results
Final checkpoint: step 133,600
(Validation losses below are from step 70,808 — the last checkpoint before a training run restart caused log rotation.)
| Validation set | Cross-entropy loss (nats) | Perplexity |
|---|---|---|
| English (en) | 1.273 | 3.57 |
RU (ru) |
0.9411 | 2.56 |
Note: Rerun with corrected injection config (Buckwalter transliteration fix + injection count fix). The training run was restarted from a mid-run checkpoint; the final step-133600 validation loss was not captured in logs due to log rotation during the restart. Reported validation losses are from the last logged checkpoint (step 70,808).
Evaluation Results
EEE-format evaluation results are stored under eval_results/eee/ in this repository.
Tasks: Global MMLU (EN/RU), PIQA, ECLeKTic, Fictive Entity (2-rate mix).
Citation
Part of the The-CoLab multilingual-transfer collection.
- Downloads last month
- 14