Downloads · 30 days
0
just1nseo/llama31-8b-if-rlvr-constraint-only
llama31-8b-if-rlvr-constraint-only is a text generation model from just1nseo. Use it when you need the model to write or continue text. It is set up for transformers.
GRPO / IF-RLVR checkpoints for meta-llama/Llama-3.1-8B-Instruct, trained with the bidirectional anchor reward (p(y|x) anchor coefficient 0.0 (constraint-only baseline)).
Downloads · 30 days
0
Access
Public
Updated Aug 7, 2026
Repo size
64.3 GB
Likes
0
Public
Click a slice to open those files.
.safetensors64.2 GB · 100%
From the Hugging Face model README
GRPO / IF-RLVR checkpoints for meta-llama/Llama-3.1-8B-Instruct, trained with the
bidirectional anchor reward (p(y|x) anchor coefficient
0.0 (constraint-only baseline)).
Each checkpoint is a complete HF model directory in its own subfolder:
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("just1nseo/llama31-8b-if-rlvr-constraint-only", subfolder="global_step_91")
tok = AutoTokenizer.from_pretrained("just1nseo/llama31-8b-if-rlvr-constraint-only", subfolder="global_step_91")
| subfolder | epoch | uploaded (UTC) |
|---|---|---|
global_step_91 | 1 | 2026-08-06 06:53:49 |
global_step_182 | 2 | 2026-08-06 13:51:35 |
global_step_273 | 3 | 2026-08-06 20:33:44 |
global_step_364 | 4 | 2026-08-07 03:02:53 |
| experiment | llama31_8b_grpo_nonthink_constraint_only_b1024_c1 |
| anchor cache | (row filter only) 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 |