You can deploy this model and start earning money today!

just1nseo/
llama31-tulu3-8b-dpo-if-rlvr-constraint-only

Name: llama31-tulu3-8b-dpo-if-rlvr-constraint-only
Availability: InStock

instruction-following

if-rlvr

text-generation

meta-llama/Llama-3.1-8B-Instruct

finetune

endpoints_compatible

Views

No views yet

just1nseo/llama31-tulu3-8b-dpo-if-rlvr-constraint-only

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)).

Each checkpoint is a complete HF model directory in its own subfolder:

python

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("just1nseo/llama31-tulu3-8b-dpo-if-rlvr-constraint-only", subfolder="global_step_91")
tok = AutoTokenizer.from_pretrained("just1nseo/llama31-tulu3-8b-dpo-if-rlvr-constraint-only", subfolder="global_step_91")

Checkpoints

subfolder	epoch	uploaded (UTC)
`global_step_91`	1	2026-08-08 14:11:14
`global_step_182`	2	2026-08-08 18:02:36
`global_step_273`	3	2026-08-08 21:50:59
`global_step_364`	4	2026-08-09 01:37:21

Training configuration


experiment	`llama31_tulu3_8b_dpo_grpo_nonthink_constraint_only_b1024_c1_t1_2k`
anchor cache	`none (constraint-only baseline)`
steps / epoch	91
train batch size	1024
rollout n	8
max prompt / response	2048 / 2048
wandb	`ifif/verl_if_rlvr`