ggml : fix UB (int << 31)

whisper : add integer quantization support (#540)

ggml : fix WASM build

ggml : fix 32-bit ARM NEON (#836)

ggml : use vzip instead of vuzp for consistency

ggml : fix WASM build

ggml : sync with ggml repo (warning fixes + asserts)

ggml : sync latest ggml + llama.cpp updates (quantization)

ggml, ci : fix build on whisper.android (ARM_NEON) + add CI (#764)

ggml : sync latest ggml

ggml : fix q4_1 dot product types (#759)

ggml : sync latest changes from ggml and llama.cpp

ggml : fix WASM build

ggml : backport llama.cpp updates (close #709)

talk-llama : add new example + sync ggml from llama.cpp (#664)

whisper : reduce memory usage during inference (#431)

whisper : PPC64 big-endian support (#398)

bench : add memcpy and ggml_mul_mat benchmarks

ggml : remove obsolete zeroing + comment fixes (#390)

ggml : correct behaviour of ggml_vec_sum_f32 (#390)

ggml : improve vec_dot_f16 unrolling in flash_attn_f16

ggml : fix bug in new soft max computation

ggml : when using BLAS start only 1 CPU thread

ggml : fix running tasks with variable number of threads

ggml : unroll ggml_vec_dot_f16 in ggml_compute_forward_flash_attn_f16

whisper : revert accidental MB change

ggml : speed-up soft max via Accelerate + unroll

ggml : use vDSP_sve and vDSP_maxv from Accelerate

ggml : make gcc happy (minor)

ggml : add SSE3 and fp16 conversion lookup table (#368)

whisper : document POWER VSX support

ggml : reorganize POWER9 ppc64le SIMD code

ggml : change f16 load and store macro arguments

ggml : add void to argument-less functions

ggml : define MIN / MAX only if not defined (minor)

ggml : improve f16 acceleration for POWER9 ppc64le

ggml : barrier refactor + static functions

ggml : simplify the SIMD code (#324)

ggml : use vaddvq_f32 for slightly more efficient reduce

ggml : add f16 acceleration for POWER9 ppc64le

ggml : make consts static (#317)

minor : small code cleanups (#302)

Check for both __ARM_NEON and __ARM_FEATURE_FMA so that the project can be compiled for armv7a.

Add AVX,AVX2 support for ggml_vec_scale_f32

ggml : implement ggml_compute_forward_dup_f16() special cases

ggml : make more compatible with c99 (#262)

ggml : fix indentation

ggml : make compatible with c99 (#262)

Remove C++20 requirement (#257)

talk : talk with AI in the terminal