0.8B/README.md

SymbolicLight V1 0.8B

Inference-only checkpoint:

• File: weights/pytorch/latest-inference.pt
• Bytes: 3,500,562,525
• SHA-256: 5edc3b6a6d3fb9cfaf713775133361b3ff907abf293a22ef2226785f9fbc0540

The matching tokenizer is stored in tokenizer/. It has 55,296 active tokens inside the model's 57,344-entry structural vocabulary. Do not use the 194M tokenizer with this checkpoint.

With the 0.8B package from the public code repository:

python src/inference.py \
  --checkpoint /path/to/latest-inference.pt \
  --prompt "Once upon a time"

This checkpoint contains model tensors and inference configuration only. It does not contain optimizer/scaler state, counters, data-loader state, local paths, or training-result metadata.