Seed-Coder: Let the Code Model Curate Data for Itself

🌐 Homepage   |   🤗 Hugging Face   |   📄 arXiv

We are thrilled to introduce Seed-Coder (previously known as Doubao-Coder), a family of lightweight yet powerful open-source code LLMs comprising base, instruct and reasoning models of 8B size.

Seed-Coder demonstrates that, with minimal human effort, LLMs can effectively curate code training data by themselves to drastically enhance coding capabilities.

Seed-Coder represents our initial step towards contributing to the open-source LLM ecosystem. We look forward to seeing Seed-Coder drive advances in code intelligence and empower broader applications in the open-source LLM community!

📢 News

[2025/05/08]🔥 We release Seed-Coder!

[2025/05/14] We found an inconsistent setting of our BigCodeBench evaluation results. We already updated all the results with the aligned BigCodeBench-Completion setting.

🌟 Highlights

• Model-centric: Seed-Coder predominantly leverages LLMs instead of hand-crafted rules for code data filtering, minimizing manual effort in pretraining data construction.

• Transparent: We openly share detailed insights into our model-centric data pipeline, including methods for curating GitHub data, commits data, and code-related web data.

• Powerful: Seed-Coder achieves state-of-the-art performance among open-source models of comparable size across a diverse range of coding tasks.

⚡ Quick Start

We are excited to introduce Seed-Coder, featuring three powerful models:

Model Name	Length	Download	Notes
Seed-Coder-8B-Base	32K	🤗 Model	Pretrained on our model-centric code data.
Seed-Coder-8B-Instruct	32K	🤗 Model	Instruction-tuned for alignment with user intent.
Seed-Coder-8B-Reasoning	64K	🤗 Model	RL trained to boost reasoning capabilities.
Seed-Coder-8B-Reasoning-bf16	64K	🤗 Model	RL trained to boost reasoning capabilities.

All the models are publicly available on Hugging Face collection.

👉🏻 Deploying Seed-Coder-8B-Instruct with transformers

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "ByteDance-Seed/Seed-Coder-8B-Instruct"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, device_map="auto", trust_remote_code=True)

messages = [
    {"role": "user", "content": "Write a quick sort algorithm."},
]

input_ids = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    return_tensors="pt",
    add_generation_prompt=True,  
).to(model.device)

outputs = model.generate(input_ids, max_new_tokens=512)
response = tokenizer.decode(outputs[0][input_ids.shape[-1]:], skip_special_tokens=True)
print(response)

For more detailed usage instructions and model-specific configurations, please refer to the Hugging Face model pages linked above.

👉🏻 Deploying Seed-Coder-8B-Instruct with vLLM

As part of the new generation of code models, Seed-Coder-8B-Instruct is fully supported by vLLM. A detailed tutorial can be found in the vLLM documentation.

Here, we provide a simple example for offline batched inference using vLLM.

Offline Batched Inference

from transformers import AutoTokenizer
from vllm import LLM, SamplingParams
# Initialize the tokenizer
tokenizer = AutoTokenizer.from_pretrained("ByteDance-Seed/Seed-Coder-8B-Instruct")

# Pass the default decoding hyperparameters of Seed-Coder-8B-Instruct
# max_tokens is for the maximum length for generation.
sampling_params = SamplingParams(temperature=0.6, top_p=0.8, repetition_penalty=1.05, max_tokens=512)

# Input the model name or path. Can be GPTQ or AWQ models.
llm = LLM(model="ByteDance-Seed/Seed-Coder-8B-Instruct")

# Prepare your prompts
prompt = "#write a quick sort algorithm."

# generate outputs
outputs = llm.generate([prompt], sampling_params)

# Print the outputs.
for output in outputs:
    prompt = output.prompt
    generated_text = output.outputs[0].text
    print(f"Prompt: {prompt!r}\n\nGenerated content: {generated_text!r}")

Multi-GPU Distributed Serving

To further boost the serving throughput, distributed inference is supported by leveraging multiple GPU devices.

When handling long-context inputs (up to 32K tokens), tensor parallelism can help mitigate GPU memory limitations.

Here is how to run Seed-Coder-8B-Instruct with tensor parallelism:

llm = LLM(model="ByteDance-Seed/Seed-Coder-8B-Instruct", tensor_parallel_size=8)

🏆 Performance

We evaluated Seed-Coder across a wide range of coding benchmarks, including code generation, code completion, code editing, code reasoning, and software engineering tasks.

Seed-Coder achieves state-of-the-art performance among open-source models at the 8B scale, and even surpasses some much larger models.

For detailed benchmark results and analysis, please refer to our Technical Report.

📖 Citation

If you find Seed-Coder useful for your research and applications, feel free to give us a star ⭐ or cite us using:

@misc{seed2025seedcoderletcodemodel,
      title={{Seed-Coder}: Let the Code Model Curate Data for Itself}, 
      author={{ByteDance Seed} and Yuyu Zhang and Jing Su and Yifan Sun and Chenguang Xi and Xia Xiao and Shen Zheng and Anxiang Zhang and Kaibo Liu and Daoguang Zan and Tao Sun and Jinhua Zhu and Shulin Xin and Dong Huang and Yetao Bai and Lixin Dong and Chao Li and Jianchong Chen and Hanzhi Zhou and Yifan Huang and Guanghan Ning and Xierui Song and Jiaze Chen and Siyao Liu and Kai Shen and Liang Xiang and Yonghui Wu},
      year={2025},
      eprint={2506.03524},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2506.03524}, 
}

📜 License

This project is licensed under MIT License. See the LICENSE file for details.

🏢 About ByteDance Seed Team

Founded in 2023, ByteDance Seed Team is dedicated to crafting the industry's most advanced AI foundation models. The team aspires to become a world-class research team and make significant contributions to the advancement of science and society.