Roadmap

[zetavg]

• Tokenizer playground.
• Flagging inference results.
• Load configs form config.json and deploy on HF space.
• Support saving flagging logs on HF.
• Model demo mode.
• Chat UI.
• Support dialogue fine-tuning, where user input can be masked out in loss.
• "Generate more like this" for fine-tuning datasets with OpenAI API.
• Save fine-tuning parameters and info to model dir.
• Support loading fine-tune parameters from previously trained models.
• Support "continue from model" and "resume from checkpoint".
• CLI fine-tuning support, and show “CLI equivalent” on UI
• Convert models to different formats and download.
• Upload models to HF with one click.
• “Continue” button in inference.
• Inference history log.

[zetavg]

Currently working on research on Transfer Learning of models while using a new language the model has not been heavily trained on.

Planning on doing a re-write on the Inference and Fine-tuning UI, with "model presets", model output comparing, loading datasets from Hugging Face, etc.