PR #4657 Add n_key_dim and n_value_dim

Add n_key_dim and n_value_dim

postmasters committed 2 years ago

Fix `llm_build_kqv` to use `n_value_gqa`

postmasters committed 2 years ago

Rebase

postmasters committed 2 years ago

Rename variables

postmasters committed 2 years ago

Fix llm_build_kqv to be more generic wrt n_embd_head_k

postmasters committed 2 years ago

Update default values for n_embd_head_k and n_embd_head_v

postmasters committed 2 years ago

Fix llm_load_tensors: the asserts were not backcompat

postmasters committed 2 years ago

llama.cpp Add n_key_dim and n_value_dim #4657 Merged