Model Publisher

Use this skill to prepare and publish trained models for public or private distribution.

Checkpoint Packaging

Convert training checkpoints to SafeTensors before publishing. Validation requirements:

• •verify tensor/key integrity after conversion
• •preserve model config/tokenizer compatibility
• •document source checkpoint and conversion command

Model Card Generation

Auto-generate a model card (README.md) with at least:

• •model overview and intended use
• •architecture + tokenizer details
• •training data summary and known limitations
• •training/eval setup and headline metrics
• •safety notes and out-of-scope usage

Hugging Face Hub Publishing

Use huggingface-cli for authentication and push workflows. Support:

• •initial repository creation
• •model + tokenizer artifact upload
• •versioned updates for new checkpoints

Optional GGUF Export

When requested, generate GGUF quantized variants for local inference. Document quantization type, expected quality impact, and target runtime profile.

Environment

Required:

• •HF_TOKEN

Release Outputs

1. •publish_manifest.json (what was published + versions)
2. •README.md (generated model card)
3. •release_notes.md (changes since last version)
4. •optional gguf_manifest.json (quantized artifacts)