Nightly 2026-07-12: SPIR-V decode +30%, LFM2.5 function calling, GGUF import (container + K-quant dequant) - #4

Merged

physics515 merged 6 commits into

mainfrom

mummu-nightly-2026-07-12

Jul 13, 2026

Commits on Jul 12, 2026

chore(deps): sha2 0.10 -> 0.11; pins held (burn 0.21 / wgpu 29 / tokenizers 0.22)

physics515
and
claude
committed
658be5c
feat(p4): LFM2.5 Pythonic tool-calling — template, renderer, bounded parser

physics515
and
claude
committed
c8f03b8
perf(p6): adopt CubeCL SPIR-V compiler on Vulkan — decode +30% (14.1 -> 18.4 tok/s)

physics515
and
claude
committed
e44debf
feat(p3): GGUF container reader — typed metadata + validated tensor table

physics515
and
claude
committed
913ca25
feat(p3): GGUF dequantization — Q8_0 + Q4_K/Q6_K superblocks, proven bit-exact/0.9975 vs true weights

physics515
and
claude
committed
a59b288
docs: research fold — Qwen3.5 small-tier zoo target; LFM2.5 llama.cpp reference caveat

physics515
and
claude
committed
39a0620