Views
No views yet
meta-llama/Llama-3.1-8B-Instruct, trained with the
bidirectional anchor reward (p(y|x) anchor coefficient
0.1).1from transformers import AutoModelForCausalLM, AutoTokenizer
2model = AutoModelForCausalLM.from_pretrained("just1nseo/llama31-tulu3-8b-dpo-if-rlvr-judge-only", subfolder="global_step_91")
3tok = AutoTokenizer.from_pretrained("just1nseo/llama31-tulu3-8b-dpo-if-rlvr-judge-only", subfolder="global_step_91")| subfolder | epoch | uploaded (UTC) |
|---|---|---|
global_step_91 | 1 | 2026-08-12 12:11:57 |
global_step_182 | 2 | 2026-08-13 04:08:20 |
global_step_273 | 3 | 2026-08-13 21:01:41 |
global_step_364 | 4 | 2026-08-14 13:44:05 |
| experiment | llama31_tulu3_8b_dpo_grpo_nonthink_llmverifier_gptoss120b_bonus01_threshold5_b1024_c1_t1_2k |
| anchor cache | loaded for dataset parity only (93,993 cached rows -> 91 steps/epoch), coefficient 0 |
| steps / epoch | 91 |
| train batch size | 1024 |
| rollout n | 8 |
| max prompt / response | 2048 / 2048 |
| wandb | ifif/verl_if_rlvr |