Sequential-Hidden-Decoding-8B-n8-Instruct

This is the instruction-tuned variant of Sequential Hidden Decoding 8B n=8, designed for conversational and instruction-following use cases.

• Base model: Sequential-Hidden-Decoding-8B-n8
• Underlying architecture: Qwen3-8B-Base
• Scale: 8x
• Context Length: 131072
• Dtype: bfloat16

Key Idea

Sequential Hidden Decoding scales sequence length by preparing multiple Embedding matrices for the same token sequence, interleaving the results, and feeding the expanded sequence into the same Transformer. This model is the instruction-tuned release of the 8B n=8 variant.

Serving (SGLang)

This model requires a patched version of SGLang for inference. See the project page for installation options.

python -m sglang.launch_server \
    --model-path tencent/Sequential-Hidden-Decoding-8B-n8-Instruct \
    --trust-remote-code \
    --tp-size 1 \
    --port 30000 --host 0.0.0.0 \
    --chunked-prefill-size -1 \
    --attention-backend fa3 \
    --mem-fraction-static 0.82 \
    --max-running-requests 32 \
    --context-length 131072 \
    --cuda-graph-max-bs 128 \
    --cuda-graph-bs 1 2 4 8 16 32 64 128

Note: Sequential Hidden Decoding models process n×-length sequences internally, so --chunked-prefill-size -1, --attention-backend fa3, and conservative batch sizing are important for stability and performance.

Chat Usage

This is an instruction-tuned model. Use the /v1/chat/completions endpoint:

from openai import OpenAI

client = OpenAI(base_url="http://localhost:30000/v1", api_key="EMPTY")
response = client.chat.completions.create(
    model="tencent/Sequential-Hidden-Decoding-8B-n8-Instruct",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Explain the idea of hidden decoding in simple terms."},
    ],
    max_tokens=512,
    temperature=0.7,
)
print(response.choices[0].message.content)

Files

This repository includes the custom architecture files required by trust_remote_code:

• configuration_qwen3_scale_seq.py
• modeling_qwen3_scale_seq.py

Related Models

Citation

@article{hidden_decoding_2026,
  title   = {Hidden Decoding: Scaling Sequence Length in Pretraining},
  year    = {2026},
  url     = {https://welm.weixin.qq.com/posts/hidden_decoding/}
}

License

This model is released under the License Terms of Sequential-Hidden-Decoding.