ksamiein/qwen-d1-cluster

Qwen 2.5 7B GRPO Training Checkpoint

This is a training checkpoint from GRPO (Generative Reinforcement Policy Optimization) training.

Training Status

• Checkpoint saved at: 2025-09-25T10:12:40.062136
• Training step: 764/2250
• Progress: 34%

Configuration

• Base model: Qwen2.5-7B-Instruct
• Training method: GRPO with LoRA
• LoRA rank: 64
• Max sequence length: 4096
• Training data: 9000 math problems from math_ood
• Categories: Number Theory, Logic, Combinatorics, Arithmetic, Algebra

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained("unsloth/Qwen2.5-7B-Instruct")
tokenizer = AutoTokenizer.from_pretrained("unsloth/Qwen2.5-7B-Instruct")

# Load LoRA weights
model = PeftModel.from_pretrained(base_model, "ksamiein/qwen-d1-cluster")

Note

This is an intermediate checkpoint. Training may not be complete.