Version | Commit message | Commit author | Last scanned | ||
|---|---|---|---|---|---|
| Mark this repo as the canonical shipped build; text sibling is a pointer | avlp12 | Aug 28, 2026 | |||
| docs: link parent collection | avlp12 | Jul 15, 2026 | |||
| LLM quality: frame the 4-bit reference as the original (INT4 master) | avlp12 | Never scanned | |||
| quality.svg source | avlp12 | Never scanned | |||
| quality figure: measure vs the original (INT4 master) — KL 0.199 / 10.2% flip | avlp12 | Jul 6, 2026 | |||
| Reposition as +vision add-on of the text build + add LLM KL quality (0.199 nats / 10.2% flip) | avlp12 | Never scanned | |||
| Add total_parameters to index (fix HF showing 133B instead of ~1T) | avlp12 | Never scanned | |||
| vision 3D: block-diagonal varlen attention (no multi-chunk OOM) | avlp12 | Never scanned | |||
| video_infer: native fast generate + wired_limit + multi-chunk | avlp12 | Never scanned | |||
| Video now fast (wired_limit) + multi-chunk motion + block-diagonal attention | avlp12 | Never scanned | |||
| Kimi-K2.7-Code 3.6bpw image+video MLX VLM (MoonViT kept) + card + helper | avlp12 | Never scanned | |||
| initial commit | avlp12 | Jan 27, 2026 | |||
12
123