Guilherme34/Firefly-v6-DONTDOWNLOAD-STILLONPROGRESS

Firefly-v6-DONTDOWNLOAD-STILLONPROGRESS

DO NOT DOWNLOAD YET — STILL IN PROGRESS.

Training completed all 3,329 optimizer steps. This is an experimental upload awaiting the owner's evaluation, not a validated release.

Personal-use experimental agentic and roleplay fine-tune of Firefly-v5. Text-only LoRA was trained with Axolotl and merged into the full Gemma 4 model. The original vision and audio components were kept frozen. Their downstream behavior has not been evaluated. No model evaluations were run at the owner's request; no benchmark superiority or improvement claim is made.

This public repository contains the merged model, processor, tokenizer and chat template. Optimizer checkpoints and detailed training records remain in the owner's separate private repository. Use the inference guide for tool messages and see source attribution.

Data sources retain their respective licenses. No Robots is CC BY-NC 4.0; this mixed-source experiment is intended for the owner's personal evaluation. See the source attribution files before redistribution or other use.