Ettore Di Giacinto 500d653bfa feat(paged): regenerate patch series 0053-0055 (P1 bf16-stream) ...

Additive regen mirroring fork mudler/llama.cpp:localai-paged HEAD 653bb2f3d
(base 1edddc8fe + 3 P1 commits). Patches 0001-0052 are untouched.

- 0053 residual-segment executor + norm-bf16.{cu,cuh} + LLAMA_BF16_CUBLAS_F32_OUT
- 0054 bf16 residual-add + rope op-variants
- 0055 BF16_STREAM_SEGMENT test-backend-ops sentinel

Kill-gate: a fresh detached worktree at pin 0ed235ea2c17a19fc8238668653946721ed136fd
applied all 46 on-disk patches in numeric order (strict git apply) and staged
tree 6cf1523047e0e38679baff20844bdc9e6829eb22, byte-for-byte == fork HEAD tree.
All default-off (LLAMA_BF16_STREAM); default md5 canonical both models
(MoE 8cb0ce23777bf55f92f63d0292c756b0, dense 5951a5b4d624ce891e22ab5fca9bc439).

Assisted-by: Claude:opus-4.8 [Claude Code]
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

2026-07-02 14:34:14 +00:00

..

paged

feat(paged): regenerate patch series 0053-0055 (P1 bf16-stream)

2026-07-02 14:34:14 +00:00