migrate to llm-jp-3-3.7b + LoRA (tabitay/meiji-llmjp-3.7b); CPU inference

feat: lower defaults (temp 0.5/top_p 0.85), add ASCII-block checkbox, align author-tag format with local generate.py

chore: upgrade Gradio to 6.13.0 and drop workarounds (jinja2/hf_hub pins no longer needed)

fix: pin jinja2<3.1.5 to avoid unhashable-dict cache bug in Gradio 4.44.0

fix: use T5TokenizerFast (repo has tokenizer.json, no spiece.model)

fix: pin huggingface_hub<0.21.0 and drop gradio (managed by sdk_version) to resolve HfFolder ImportError

Upload README.md with huggingface_hub

Upload folder using huggingface_hub

initial commit