Instructions to use bigai/GS1-0.8B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use bigai/GS1-0.8B with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("feature-extraction", model="bigai/GS1-0.8B", trust_remote_code=True)# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("bigai/GS1-0.8B", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
|
Download assets/architecture.svg from bigai/GS1-0.8B: direct link, hf CLI and curl.
- Browser
- Download file 3.13 kB
-
https://huggingface.co/bigai/GS1-0.8B/resolve/main/assets/architecture.svg
- Command line
-
hf download hf://bigai/GS1-0.8B/assets/architecture.svg
-
curl -L -o architecture.svg https://huggingface.co/bigai/GS1-0.8B/resolve/main/assets/architecture.svg
3.13 kB
| <svg xmlns="http://www.w3.org/2000/svg" width="1280" height="510" viewBox="0 0 1280 510" role="img" aria-labelledby="title desc"> | |
| <title id="title">GS1: from state and typed questions to native decisions</title> | |
| <desc id="desc">A compiler builds isolated branches from shared state, questions and runtime candidates. Hybrid sequence mixing and attention residuals feed a cross-attention decision head, producing typed probabilities without generating text.</desc> | |
| <defs><marker id="arrow" viewBox="0 0 10 10" refX="9" refY="5" markerWidth="7" markerHeight="7" orient="auto-start-reverse"><path d="M0 0L10 5L0 10Z" fill="#C00000"/></marker></defs> | |
| <rect width="1280" height="510" rx="18" fill="#FFFFFF"/> | |
| <g font-family="Inter,Arial,sans-serif"><text x="42" y="51" fill="#29252A" font-size="25" font-weight="700">State in. Decisions out.</text><text x="42" y="81" fill="#82757B" font-size="16">Native model architecture · identical weights in both inference paths</text> | |
| <g fill="#FFF7F7" stroke="#E4BABA"><rect x="42" y="136" width="230" height="76" rx="10"/><rect x="42" y="230" width="230" height="116" rx="10"/><rect x="327" y="136" width="400" height="210" rx="12"/><rect x="782" y="136" width="220" height="210" rx="12"/><rect x="1053" y="136" width="185" height="210" rx="12"/></g> | |
| <g fill="#29252A" font-size="18" font-weight="600"><text x="61" y="167">Shared state</text><text x="61" y="261">Typed questions</text><text x="347" y="169">Hybrid text / vision backbone</text><text x="802" y="169">Native decision head</text><text x="1073" y="169">Typed answers</text></g> | |
| <g fill="#6B5F65" font-size="15"><text x="61" y="192">Text · structured records</text><text x="61" y="290">Instructions + candidates</text><text x="61" y="317">Defined at request time</text><text x="347" y="207">Causal GDN + segmented full attention</text><text x="347" y="237">Carry-preserving Attention Residuals</text><text x="347" y="267">Isolated question branches</text><text x="347" y="314" fill="#C00000">State · question · options · decision token</text><text x="802" y="208">Cross-attention to state</text><text x="802" y="237">Candidate routing</text><text x="802" y="267">Joint FP32 probabilities</text><text x="802" y="314" fill="#C00000">Zero generated tokens</text><text x="1073" y="213">Choice → p(options)</text><text x="1073" y="250">Noul → p(true)</text><text x="1073" y="287">Score → E[level]</text></g> | |
| <g fill="none" stroke="#C00000" stroke-width="2" marker-end="url(#arrow)"><path d="M272 174H299V241H325"/><path d="M272 288H299V241"/><path d="M727 241H780"/><path d="M1002 241H1051"/></g> | |
| <rect x="327" y="381" width="911" height="78" rx="10" fill="#FFF0F0" stroke="#DDAAAA"/><text x="347" y="411" fill="#C00000" font-size="17" font-weight="600">Optimized gs1 runtime</text><text x="347" y="439" fill="#6B4F57" font-size="15">Qualified CUDA Graph replay · compatible batching · immutable shared-state reuse · native eager fallback</text> | |
| <text x="42" y="411" fill="#6B5F65" font-size="16">Hugging Face AutoModel</text><text x="42" y="439" fill="#82757B" font-size="14">Standalone eager inference</text></g></svg> | |