Styx-100M

Styx is a 96.48-million-parameter language model developed by Promethean Studios as part of the Talos model family.

Training

• Training tokens consumed: 600,001,892
• Final training step: 293,543
• Training loss: 1.3478
• Validation loss: 1.3707
• Vocabulary size: 1,024
• Maximum sequence length: 512
• Tokenizer: Included as tokenizer.json

Files

• step-293543.pt: Full PyTorch training checkpoint, including model weights and training state.
• tokenizer.json: Matching tokenizer.
• config.json: Architecture and training metadata.

Usage and limitations

This is a custom Talos checkpoint, not a drop-in Transformers AutoModelForCausalLM model. Use the Talos repository's compatible model and generation code to load it.

Initial generation tests confirm that the checkpoint loads and generates text. The sample outputs are not yet consistently coherent, so this release should be considered an experimental research checkpoint, not a production-ready assistant.

Project

Developed by Promethean Studios.