Jbyang d92f5835b8 * [Metal:Speed] LLM prefill/decode kernel + shape/RoPE infrastructure + Metal Op Profile. (#4650) ...

Combined Metal LLM decode/prefill optimization work plus supporting
core-graph and RoPE infrastructure.

GitOrigin-RevId: b71528f0de1cf17f68628031e6e59c373227f07d

2026-07-21 19:32:55 +08:00

..

audio

[LLM:Bugfix] Fix Qwen2.5-Omni audio preprocessing edge cases (#4459)

2026-06-11 19:24:36 +08:00

converter

* [Metal:Speed] LLM prefill/decode kernel + shape/RoPE infrastructure + Metal Op Profile. (#4650)

2026-07-21 19:32:55 +08:00

cpp

[MNN:Bugfix] fix QNN windows msvc compile error.

2026-05-19 16:55:21 +08:00

cv

MNN:Sync: Sync Internal 3.2.4

2025-09-22 23:05:26 +08:00

evaluation

Github release 1.1.0

2020-11-05 16:49:17 +08:00

mnncompress

[MNN:Sync] Sync Internal 2.4.1

2023-03-20 11:32:29 +08:00

quantization

[LLM:Feature] support fp16 storage for weight-quant scale/zero-point

2026-05-18 14:21:37 +08:00

script

[MNN:Refact] Delete 'MNN_CPU_DEQUANT_GEMM' code. (#4506)

2026-06-03 14:10:43 +08:00

train

[MNN:Bugfix] Fix regressions introduced by recent PRs

2026-03-11 09:35:49 +08:00