Views
No views yet
Qwen/Qwen3-4B-Instruct-2507sft_adapter/| Parameter | Value |
|---|---|
| Method | LoRA (Supervised Fine-Tuning) |
| LoRA rank | 32 |
| Epochs | 2 |
| Training samples | 500 |
| Task | Math reasoning (GSM8K + NuminaMath) |
| Size | 270.92 MB |
grpo_adapter/| Parameter | Value |
|---|---|
| Method | GRPO (Group Relative Policy Optimization) |
| Training samples | 400 |
| Group size | 8 |
| Learning rate | 3e-6 |
| Substeps | 1 |
| Curriculum | easy → intermediate → hard |
| Size | 270.92 MB |
Recommended: Usegrpo_adapter/— trained through the full SFT + GRPO pipeline.
1from peft import PeftModel
2from transformers import AutoModelForCausalLM, AutoTokenizer
3
4base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-4B-Instruct-2507")
5tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-4B-Instruct-2507")
6
7# Load GRPO adapter (recommended)
8model = PeftModel.from_pretrained(base, "MohammadRafiML/Qwen3-4B-Instruct-2507-Capstone-MathRL", subfolder="grpo_adapter")
9model = model.merge_and_unload()
10
11# Load SFT adapter only
12# model = PeftModel.from_pretrained(base, "MohammadRafiML/Qwen3-4B-Instruct-2507-Capstone-MathRL", subfolder="sft_adapter")
13# model = model.merge_and_unload()