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/llama31-8b-if-rlvr-anchor-pyx01", subfolder="global_step_91")
3tok = AutoTokenizer.from_pretrained("just1nseo/llama31-8b-if-rlvr-anchor-pyx01", subfolder="global_step_91")| subfolder | epoch | uploaded (UTC) |
|---|---|---|
global_step_91 | 1 | 2026-08-04 21:52:20 |
global_step_182 | 2 | 2026-08-05 06:32:30 |
global_step_273 | 3 | 2026-08-05 15:07:39 |
global_step_364 | 4 | 2026-08-05 23:26:48 |
global_step_455 | 5 | 2026-08-07 12:10:35 |
global_step_546 | 6 | 2026-08-07 20:17:49 |
| experiment | llama31_8b_grpo_nonthink_pyx01_t8banchor_s8b_b1024_c1 |
| anchor cache | if_ref_anchor_teacher_llama31_8b_instruct_nonreason_train_seed1_val512_scored_by_llama31_8b_instruct_topp095_topk20.json |
| steps / epoch | 91 |
| train batch size | 1024 |
| rollout n | 8 |
| max prompt / response | 2048 / 2048 |
| wandb | ifif/verl_if_rlvr |