README.md

---
library_name: transformers
tags:
- small-lm
- math
- reasoning
- slm
license: apache-2.0
datasets:
- openai/gsm8k
base_model:
- Qwen/Qwen3-4B-Instruct-2507
---

# Qwen3-4B-Instruct-2507-Math

This model is obtained by fine-tuning Qwen/Qwen3-4B-Instruct-2507 on the [gsm8k](https://huggingface.co/datasets/openai/gsm8k) train split. 
The model is used in the experiments described in https://bknyaz.github.io/blog/2026/meta-merge/. 
Single A100 was used for fine-tuning and evaluation.

The following versions were used for train/eval:

- python >= 3.10
- torch               : 2.9.0+cu128
- lm_eval             : 0.4.9.1
- vllm                : 0.11.1
- transformers        : 4.57.6
- datasets            : 3.2.0
- numpy               : 2.2.6

## Training

The [TRL](https://github.com/huggingface/trl) library was used with SFT/full-rank options:

```bash
python trl/scripts/sft.py --model_name_or_path Qwen/Qwen3-4B-Instruct-2507 --dataset_name openai/gsm8k --dataset_config main --learning_rate 2e-5 \
--num_train_epochs 1 --per_device_train_batch_size 2 --gradient_accumulation_steps 8 --gradient_checkpointing --eos_token '<|im_end|>' --eval_strategy steps \
--eval_steps 100 --completion_only_loss True --report_to wandb --output_dir /path/to/the/finetuned/model
```

This is by far not the most compute and performance efficient fine-tuning, but it could be a good baseline.

The dataset was preprocessed to the conversational format:

```python
# trl/scripts/sft.py

dataset = load_dataset(...)

def preprocess_function(example):
  return {
  "prompt": [{"role": "user", "content": example["question"]}],
  "completion": [
      {"role": "assistant", "content": example['answer']}
  ],
  }

dataset = dataset.map(preprocess_function)
```

## Evaluation

Evaluation was done with lm_eval on the test split of gsm8k:

```bash
python -m lm_eval --model vllm --model_args pretrained=${model},tensor_parallel_size=1,dtype=auto,gpu_memory_utilization=0.9,data_parallel_size=1 \
 --tasks gsm8k --batch_size 1 --apply_chat_template=True --confirm_run_unsafe_code --trust_remote_code
```

### Results

| Model                       | gsm8k|
|-----------------------------|------|
| Qwen3-4B-Instruct-2507      | 80.4 |
| Qwen3-4B-Instruct-2507-Math | 76.8 |

## License

Please refer to the license of the original model [Qwen/Qwen3-4B-Instruct-2507](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507) and dataset [gsm8k](https://huggingface.co/datasets/openai/gsm8k).