ggml : fix build with OpenBLAS (close ggml-org#2066)

embd-input : fix returning ptr to temporary

metal : release buffers when freeing metal context (ggml-org#2062)

Better CUDA synchronization logic (ggml-org#2057)

Use unsigned for random seed (ggml-org#2006)

* Use unsigned for random seed. Keep -1 as the value to use a time based seed.

Co-authored-by: Georgi Gerganov <[email protected]>

llama : replacing auto &kv with const auto &kv (ggml-org#2041)

* Replacing auto &kv with const auto &kv

* Create codacy.yml

* Delete codacy.yml

llama : remove shards weight file support (ggml-org#2000)

* Remove multiple shards

* Remove multiple file loaders

* Remove llama_load_tensor_shard class

* Simplify load logic

* Remove dead code guess_n_parts function

* Remove vocab_only from constructor of llama_model_loader

* Remove alignment_prevents_mmap which is not more needed.

* Remove useless check

cuda : fix missing const qualifier in casts (ggml-org#2027)

CUDA GPU acceleration for LoRAs + f16 models (ggml-org#1970)

cuda : remove nchannels_x argument from mul_mat_vec_nc_f16_f32 (ggml-…

…org#2028)

- Not used