Styx 100M

Styx 100M is a 96.5M-parameter causal language model developed by Promethean Studios as part of the Talos model family.

This release contains the model trained through 100,000 training steps.

Screenshot 2026-09-28 12.26.45 AM

Model Details

Property Value
Parameters 96,482,304
Vocabulary 1,024
Hidden size 1,024
Layers 6
Attention heads 64
KV heads 32
Feed-forward network SwiGLU
Maximum sequence length 512
Training steps 100,000
Architecture Causal Transformer

Styx uses grouped-query attention with 32 KV heads and a SwiGLU feed-forward network.

Files

  • model.safetensors - model weights in SafeTensors format
  • config.json - model configuration
  • tokenizer/tokenizer.json - Styx tokenizer
  • checkpoints/step-100000.pt - original training checkpoint

The original .pt checkpoint preserves the training state and metadata, while model.safetensors contains the model weights only.

Training

The released checkpoint corresponds to:

100,000 training steps

The checkpoint was produced using the Talos training infrastructure developed by Promethean Studios.

Training metadata, including the training and validation loss recorded at the checkpoint, is preserved inside the original training checkpoint.

Intended Use

Styx 100M is primarily a research and development model.

It is intended for:

  • experimentation with small language models
  • local inference
  • model architecture research
  • benchmarking
  • studying data and parameter efficiency
  • development within the Talos ecosystem

This model should not be assumed to have the capabilities, reliability, or safety characteristics of larger production language models.

Limitations

Styx 100M is a relatively small language model and may produce:

  • incorrect or fabricated information
  • repetitive text
  • incomplete responses
  • poor reasoning
  • inconsistent instruction following
  • unsafe or undesirable generations

The model has not been presented as a general-purpose production assistant.

About Talos

Talos is the model development project of Promethean Studios, focused on building efficient language models at progressively larger scales.

Styx 100M represents the next stage of that development following the smaller Talos models.

License

Apache License 2.0.

Organization

Promethean Studios

Styx 100M is released as part of the Promethean Studios model family.

Downloads last month
-
Safetensors
Model size
96.5M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Collection including PrometheanStudio/styx-100m