Split launchers into Mitsuba_bridge_NVIDIA/ and Mitsuba_bridge_AMD/

Add AMD Radeon (HIP) beta launchers and diagnostic bat

Add Mitsuba-Bridge step card (English)

Add Mitsuba-Bridge step card (image)

Reorganize: Mitsuba-ComfyUI-27B/Model-v1.18/ (model, mmproj), LoRA, Mitsuba_bridge/ (one-click start)

Move documents and charts into docs/ (top level keeps only files to download)

Add one-click Windows start (HiMitsuba_start/ComfyUI/stop.bat + Mitsuba-Bridge)

README: how HiMitsuba differs from a judge-model approach

Add two-forms diagram: README.md

Add two-forms diagram: two_forms.png

Add HiMitsuba uncensored LoRA: comparison_lora_mobile.png

Add HiMitsuba uncensored LoRA: comparison_lora.png

Add HiMitsuba uncensored LoRA: EVALUATION.md

Add HiMitsuba uncensored LoRA: README.md

Add HiMitsuba uncensored LoRA: HiMitsuba-Uncensored-LoRA.gguf

Fix required VRAM in comparison.png (measured from llama-server's own allocation, CTX 131K)

Open the model to everyone (no approval needed)

Add the original Qwen3.8-27B (BF16) to the comparison, and the non-inferiority chart against the plain Bonsai

Evaluation wording: conditions were in the request text

README: wording of the Japanese summary line

README: add PTQ1_0 thinking-ON columns

README: add thinking-ON measurement (PQ2_0)

Add mmproj-Q8_0.gguf

Add Mitsuba-ComfyUI-27B-v1.18-PTQ1_0.gguf

Add Mitsuba-ComfyUI-27B-v1.18-PQ2_0.gguf

README: thinking must be off

Larger comparison chart; remove how-it-was-made section

Larger comparison chart; remove how-it-was-made section

Add comparison chart; self-control shown after penalty (same as diagnosis)

Add comparison chart; self-control shown after penalty (same as diagnosis)

Add comparison chart; self-control shown after penalty (same as diagnosis)

README: show all 10 axes incl. Uncensored

Add EVALUATION.md

Add NOTICE

Add LICENSE

Add README.md

initial commit