File size: 477 Bytes
4fb5c1f 99adea6 4fb5c1f 99adea6 4fb5c1f 99adea6 4fb5c1f 99adea6 4fb5c1f 99adea6 4fb5c1f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 | # payelb/UltraFeedback_openbmb_Llama-3.2-1B_aligned_with_baseline_deberta_RM
Base model: meta-llama/Llama-3.2-1B-Instruct
Alignment dataset: openbmb/UltraFeedback
Reward model: payelb/UltraFeedback_openbmb_reward-model-deberta-v3-base_1k_fixed_baseline
Method: PPO alignment with LoRA adapters.
Notes:
- Reward normalization and clipping enabled
- KL control enabled
- pad_token_id/eos_token_id explicitly set
- DeBERTa RM loaded on a single device (no device_map='auto')
|