Hermes Katana v15 MiniLM PyTorch

This is the PyTorch/safetensors runtime artifact for the distilled Hermes Katana v15 MiniLM classifier. It is intended as an optional checkpoint runtime for users who prefer PyTorch or whose systems run this checkpoint faster than the ONNX artifact.

The default small public runtime remains the ONNX artifact. This repository exists so setup can offer a clear choice between ONNX Runtime and PyTorch checkpoint runtimes.

Files

• model.safetensors
• config.json
• tokenizer.json
• tokenizer_config.json
• special_tokens_map.json
• added_tokens.json
• vocab.txt
• artifact_manifest.json

No training data or evaluation corpora are included in this artifact repository.

Hermes Katana usage

pip install "hermes-katana[torch-cpu]"
katana artifacts download minilm_torch

Or through guided setup:

katana setup --small-torch --torch-cpu

Pin a release revision for reproducible deployments.