Syn-0.1
Syn-0.1 is an early experimental release of Syn, the option-scoring service behind sifty.dev. This repository contains data, benchmark results, logs, and checkpoints. It is an artifact collection, not a single model loadable from the repository root.
The pointer run adapts Qwen/Qwen3-8B-Base; its weights were not trained from scratch. Results are experiment-specific. Check the run configuration and evaluation before using a checkpoint.
| Path | Contents |
|---|---|
data/ |
Labeled rows for training and evaluation. |
pointer-data/ |
Pointer training rows and data provenance. |
pointers/ |
Adapted backbone and pointer checkpoints. |
jevbench/ |
Benchmark artifacts. |
logs/ |
Training logs. |
The Syn README explains the scoring API and training methods. The RunPod guide explains how to train with this repository as the artifact store. For upload or training jobs that publish artifacts, use an HF_TOKEN with write access to siftylabs/Syn-0.1; public downloads do not need a token.
To serve an available pointer run, set SYN_MODEL=hf://siftylabs/Syn-0.1/pointers/<model>/<run>/backbone, SYN_READOUT=pointer, and SYN_POINTER_PATH=hf://siftylabs/Syn-0.1/pointers/<model>/<run>/pointer.safetensors. Replace <model> and <run> with a directory listed in this repository.
Model tree for siftylabs/Syn-0.1
Base model
Qwen/Qwen3-8B-Base