Deepseek R1 Distill Qwen 7B Japanese

🚀 lightblue/DeepSeek-R1-Distill-Qwen-7B-Japanese

Deepseek's R1 models are excellent for reasoning but may output inconsistent languages. This Japanese version reliably outputs Japanese in response to prompts.

🚀 Quick Start

When using these models, we recommend using a sampling temperature of between 0.5 - 0.7, as per the original distilled R1 models.

Additionally, we have observed that the model sometimes tends to repeat itself more than the original R1 model, so we also recommend setting repetition_penalty to 1.1, or higher if the model repeats itself when processing your prompts.

✨ Features

• Reliable Japanese Output: This model is a Japanese fine - tuned version of deepseek-ai/DeepSeek-R1-Distill-Qwen-7B on the lightblue/distilabel-reasoning-R1-Llama-70B dataset, which can reliably and accurately output Japanese in response to prompts.

📦 Installation

To use this model in vLLM, you need to install vLLM first. You can install it using the following command:

pip install vllm

💻 Usage Examples

Basic Usage

from vllm import LLM, SamplingParams

llm = LLM(
    model="lightblue/DeepSeek-R1-Distill-Qwen-7B-Japanese",
    max_model_len=8_000
)

sampling_params = SamplingParams(
    temperature=0.5, 
    max_tokens=8_000,
    repetition_penalty=1.1
)

prompts = [
    """学校には1クラスにつき20人の生徒がおり、クラスは合計3つあります。
学校全体では男子と女子がそれぞれ50%ずついます。
1つ目のクラスには女子が15人、2つ目のクラスには女子が12人います。
3つ目のクラスには何人の男子がいますか？"""
]

conversations = [
    [{"role": "user", "content": x}] for x in prompts
]

outputs = llm.chat(conversations, sampling_params=sampling_params)

for output in outputs:
    print(output.outputs[0].text)

<think>
# まず、学校の総生徒数を算出します。各クラスに20人の生徒があり、クラスは3つあるため、総生徒数は60人です。

# 次に、学校全体で男子と女子は同じ人数で分布しています。したがって、男子と女子各有30人。
...
# したがって、3つ目のクラスの男子数は20 - 3 = 17人です。
# </think>

# **解答：**

# 学校の総生徒数を算出します。
...
# **最終的な答え：**
# \[
# \boxed{17}
# \]

📚 Documentation

Evaluation

We evaluated this model for output accuracy and the percentage of valid Japanese <think> sections using the first 50 rows of the SakanaAI/gsm8k-ja-test_250-1319 dataset.

We compare this to the original R1 model and test in both regimes where repetition penalty is 1.0 and 1.1:

Code for the SakanaAI/gsm8k-ja-test_250-1319 evaluation can be found here.

We further use the first 50 prompts from DeL-TaiseiOzaki/Tengentoppa-sft-reasoning-ja to evaluate the percentage of valid Japanese <think> sections in model responses. This benchmark contains more varied and complex prompts, meaning this is a more realistic evaluation of how reliably this model can output Japanese.

Code for the DeL-TaiseiOzaki/Tengentoppa-sft-reasoning-ja evaluation can be found here.

How this model was made

We made the data for this model using the following steps:

1. Sample English reasoning - style prompts from argilla/distilabel-reasoning-prompts.
2. Remove similar prompts using text similarity based on BAAI/bge-m3 embeddings.
3. Translate English prompts to Japanese using gpt-4o-mini-2024-07-18.
4. Generate answers to prompts using deepseek-ai/DeepSeek-R1-Distill-Llama-70B.
5. Filter out responses which did not:
   • Finish within 2048 tokens
   • Contain a valid <think> section
   • Have the <think> section written in Japanese

Training details

### model
model_name_or_path: deepseek-ai/DeepSeek-R1-Distill-Qwen-7B

### method
stage: sft
do_train: true
finetuning_type: full
deepspeed: /root/LLaMA-Factory/examples/deepspeed/ds_z2_config.json

### dataset
dataset: distilabel-reasoning-R1-Llama-70B-ja-train
template: qwen
cutoff_len: 4500
overwrite_cache: true
preprocessing_num_workers: 16
packing: true

### output
output_dir: /root/train_outputs/DeepSeek-R1-Distill-Qwen-7B/distilabel-reasoning-R1-Llama-70B-ja-train
logging_steps: 1
save_steps: 0.99999
plot_loss: true
overwrite_output_dir: true

### train
per_device_train_batch_size: 1
gradient_accumulation_steps: 1
learning_rate: 1.0e-5
num_train_epochs: 1.0
lr_scheduler_type: cosine
warmup_ratio: 0.01
bf16: true
ddp_timeout: 180000000

### eval
val_size: 0.01
per_device_eval_batch_size: 1
eval_strategy: steps
eval_steps: 0.1

echo '{
  "distilabel-reasoning-R1-Llama-70B-ja-train": {
    "hf_hub_url": "lightblue/distilabel-reasoning-R1-Llama-70B-ja-train",
    "formatting": "sharegpt"
  }
}' > /root/LLaMA-Factory/data/dataset_info.json

cd /root/LLaMA-Factory && llamafactory-cli train /root/reasoning_train.yaml

rm -r /root/train_outputs/DeepSeek-R1-Distill-Qwen-7B/distilabel-reasoning-R1-Llama-70B-ja-train/checkpoint*
huggingface-cli upload lightblue/DeepSeek-R1-Distill-Qwen-7B-Japanese /root/train_outputs/DeepSeek-R1-Distill-Qwen-7B/distilabel-reasoning-R1-Llama-70B-ja-train

📄 License

We share this model under an Apache 2.0 license.

Developed by

This model was trained by Peter Devine (ptrdvn) for Lightblue