File size: 1,211 Bytes
8d1e8df
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c25fe69
 
 
8d1e8df
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
---
base_model: Qwen/Qwen3-4B
library_name: peft
pipeline_tag: text-generation
tags:
- base_model:adapter:Qwen/Qwen3-4B
- sft
- grpo
- lora
- transformers
- trl
---

# Combined SFT + GRPO LoRA Adapter for Qwen3-4B

This adapter combines two LoRA training stages into a single adapter:

1. **SFT** (Supervised Fine-Tuning) on Qwen/Qwen3-4B
2. **GRPO** (Group Relative Policy Optimization) on the SFT model

The two rank-32 adapters were merged into a single **rank-64** adapter (lossless).
Apply directly to `Qwen/Qwen3-4B` — no intermediate merged model needed.

## Usage

```python
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-4B")
model = PeftModel.from_pretrained(base_model, "abdul-hannan/qwen3-math-grpo")
tokenizer = AutoTokenizer.from_pretrained("abdul-hannan/qwen3-math-grpo")
```

## Training Details

- **Base model:** Qwen/Qwen3-4B
- **LoRA rank:** 64 (combined from two rank-32 adapters)
- **LoRA alpha:** 128
- **Target modules:** q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
- **PEFT version:** 0.18.1

### Contact
Syed Abdul Hannan

### Framework versions

- PEFT 0.18.1