Qwen3-1.7B RLVR — GSM8K Math Reasoning
Fine-tuned from Qwen/Qwen3-1.7B using Reinforcement Learning from Verifiable Rewards (RLVR) on the GSM8K grade school math dataset.
Training Details
| Base model | Qwen/Qwen3-1.7B |
| Technique | RLVR with LoRA (rank 32, alpha 64) |
| Dataset | GSM8K train split, 200 samples |
| Epochs | 2 |
| Learning rate | 1e-5 |
| Batch size | 128 |
| Training platform | Amazon SageMaker Serverless Model Customization |
Intended Use
This model is trained to solve grade school math word problems with step-by-step reasoning. It is intended for educational and research purposes, particularly as a demonstration of RLVR fine-tuning on a small dataset.
Primary users: Workshop participants and developers exploring RLVR on SageMaker.
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("tocsa/qwen3-1.7b-rlvr-gsm8k")
tokenizer = AutoTokenizer.from_pretrained("tocsa/qwen3-1.7b-rlvr-gsm8k")
prompt = "Janet's ducks lay 16 eggs per day. She eats 3 for breakfast and bakes 4 into muffins. She sells the rest for $2 each. How much does she make per day?"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))