whisper : fix with_state methods to use the correct state (#1519)

whisper : fix overriding the audio context

cuda : assert ggml_add sources to be contiguous

ios : sync submodule

sync : ggml (ggml-alloc + linker + gguf fixes) (#1501)

quantize : add support for K-quant types

bench : fix memcpy bench size

talk-llama : improve quote and backtick handling (#1364)

talk-llama : enable GPU by default

models : add info about distilled models

release : v1.5.0

bench-all : add distil models

js : latest whisper.js

bench-all : indentations

whisper : make large version explicit + fix data size units (#1493)

java : fix test (#1492)

whisper : add batched decoding (#1486)

java : use tiny.en for tests (#1484)

whisper : add grammar-based sampling (#1229)

talk-llama : add n_gpu_layers parameter (#1475)

examples : add whisper.android.java for compatibility with older Android versions using Java (#1382)

readme : update comment about source code

ggml : fix some compile warnings

readme : update GPU / CUDA

whisper : add full CUDA and Metal offloading (#1472)

whisper : return with error from whisper_encode_internal and whisper_decode_internal when abort callback is true (#1456)

talk-llama : add language auto detect (#1467)

openvino : update convert-whisper-to-openvino.py to support v3 (#1459)

coreml : use the correct `n_mel` value (#1458)

whisper : reset mel time when resetting timings (#1452)

ios : add support for Swift Package Manager (#1370)

release : v1.4.3

ggml : fix MIN / MAX macro re-definition

whisper : add support for large v3 (#1444)

android : decouple example into a library and app module (#1445)

whisper : reset ctx->t_start_us when calling whisper_reset_timings() (#1434)

cmake : fix talk-llama build

metal : fix asserts for setThreadgroupMemoryLength (close #1435)

ci : fix variable names in GitHub actions config (#1440)

talk-llama : fix n_gpu_layers usage again (#1442)

whisper : add missing about callback initializers

examples : fix n_gpu_layers usage in talk-llama (#1441)

whisper : add context param to disable gpu (#1293)

whisper : add support for new distilled Whisper models (#1424)

cuda : fix HIPBLAS build

sync : ggml (backend v2, k-quants, CUDA opts, Metal opts, etc.) (#1422)

models : use absolute paths for the converted model (#1356)

talk-llama : move up-to-date demo to top (#1417)

talk-llama : add an up-to-date demo video

examples : Implement JSON output for Token-Level data in main (#1358)