--- base_model: allenai/Llama-3.1-Tulu-3-8B-DPO library_name: transformers pipeline_tag: text-generation tags: - verl - grpo - rlvr - instruction-following --- # llama31_tulu3_8b_dpo_grpo_nonthink_constraint_only_b1024_c1_kl0p05 GRPO checkpoints exported by [verl](https://github.com/volcengine/verl). Every `global_step_/` folder is a complete bfloat16 Hugging Face model, uploaded as soon as the trainer finished writing that step. ```python from transformers import AutoModelForCausalLM, AutoTokenizer step = "global_step_91" model = AutoModelForCausalLM.from_pretrained("just1nseo/llama31-tulu3-8b-dpo-if-rlvr-constraint-only-kl0p05", subfolder=step) tokenizer = AutoTokenizer.from_pretrained("just1nseo/llama31-tulu3-8b-dpo-if-rlvr-constraint-only-kl0p05", subfolder=step) ``` - Base model: `allenai/Llama-3.1-Tulu-3-8B-DPO` - Run name: `llama31_tulu3_8b_dpo_grpo_nonthink_constraint_only_b1024_c1_kl0p05`