qwen3_4b_grpo_think_constraint_only_b1024_c1_kl0p02

GRPO checkpoints exported by verl.

Every global_step_<N>/ folder is a complete bfloat16 Hugging Face model, uploaded as soon as the trainer finished writing that step.

from transformers import AutoModelForCausalLM, AutoTokenizer

step = "global_step_91"
model = AutoModelForCausalLM.from_pretrained("just1nseo/qwen3-4b-think-if-rlvr-constraint-only-kl0p02", subfolder=step)
tokenizer = AutoTokenizer.from_pretrained("just1nseo/qwen3-4b-think-if-rlvr-constraint-only-kl0p02", subfolder=step)
  • Base model: Qwen/Qwen3-4B
  • Run name: qwen3_4b_grpo_think_constraint_only_b1024_c1_kl0p02
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for just1nseo/qwen3-4b-think-if-rlvr-constraint-only-kl0p02

Finetuned
Qwen/Qwen3-4B
Finetuned
(1,290)
this model