README.md

metadata

license: apache-2.0
base_model:
  - ByteDance-Seed/Seed-Coder-8B-Base

Seed-Coder-8B-Instruct

Introduction

Seed-Coder-8B-Instruct is an 8-billion-parameter model instruction-tuned specifically for code generation, code reasoning, and code understanding. It is built to empower developers with high-quality, efficient code assistance. It features:

• Trained on a massively curated corpus, where an LLM-based filter is applied to select high-quality real-world code, text-code alignment data, and synthetic datasets — ensuring cleaner and more useful data compared to traditional heuristic-based curation.
• Achieves superior performance across code generation, bug fixing, and reasoning tasks, rivaling or surpassing larger open-source code models.
• Instruction-tuned to reliably follow user intents across a diverse range of coding and reasoning prompts.
• Supports long-context handling up to 32K tokens, enabling processing of complex multi-file projects and detailed coding tasks.

Model Downloads

Model Name	Type	Length	Download
Seed-Coder-8B-Base	base	32k	🤗 Hugging Face
👉Seed-Coder-8B-Instruct	instruct	32k	🤗 Hugging Face
Seed-Coder-8B-Reasoning	reasoning	32k	🤗 Hugging Face

Requirements

You will need to install the latest versions of transformers and accelerate:

pip install -U transformers accelerate

Quickstart

Here is a simple example demonstrating how to load the model and generate code using the Hugging Face pipeline API:

import transformers
import torch

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

pipeline = transformers.pipeline(
    "text-generation",
    model=model_id,
    model_kwargs={"torch_dtype": torch.bfloat16},
    device_map="auto",
)

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

outputs = pipeline(
    messages,
    max_new_tokens=512,
)
print(outputs[0]["generated_text"][-1]["content"])

Evaluation

Seed-Coder-8B-Instruct demonstrates strong performance across a variety of coding benchmarks, showing:

• Competitive or superior results compared to similarly sized open-source code models.
• Robustness across different programming languages and domains.
• Ability to understand, reason, and repair complex code snippets.

Model	HumanEval	MBPP	MHPP	BigCodeBench (Full)	BigCodeBench (Hard)	LiveCodeBench (2410-2502)
CodeLlama-7B-Instruct	40.9	54.0	6.7	21.9	3.4	3.6
DeepSeek-Coder-6.7B-Instruct	74.4	74.9	20.0	35.5	10.1	9.6
CodeQwen1.5-7B-Chat	83.5	77.7	17.6	39.6	18.9	3.0
Yi-Coder-9B-Chat	82.3	82.0	26.7	38.1	11.5	17.5
Llama-3.1-8B-Instruct	68.3	70.1	17.1	36.6	13.5	11.5
OpenCoder-8B-Instruct	83.5	79.1	30.5	40.3	16.9	17.1
Qwen2.5-Coder-7B-Instruct	88.4	82.0	26.7	41.0	18.2	17.3
Seed-Coder-8B-Instruct	84.8	85.2	36.2	53.3	20.5	24.7

For detailed results, please check our 📑 paper.

Citation

If you find our work helpful, feel free to give us a cite.

@article{zhang2025seedcoder,
    title={Seed-Coder: Let the Code Model Curate Data for Itself},
    author={Xxx},
    year={2025},
    eprint={2504.xxxxx},
    archivePrefix={arXiv},
    primaryClass={cs.CL},
    url={https://arxiv.org/abs/xxxx.xxxxx}, 
}