root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-13 20:14:29 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

llama : update offload functions for KQ tensors

Browse Source
This commit is contained in:
Georgi Gerganov 2023-10-31 08:24:07 +02:00
parent 2926ef63b1
commit 6669cd8329
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 5 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
llama.cpp
View File

@ -4856,12 +4856,13 @@ struct llm_offload_trie {
static const std::unordered_map<const char *, llm_offload_func_e> k_offload_map = {
//{ "inp_tokens", OFFLOAD_FUNC_NR }, // TODO: missing K-quants get_rows kernel
//{ "inp_embd", OFFLOAD_FUNC_NR }, // TODO: missing K-quants get_rows kernel
{ "inp_pos", OFFLOAD_FUNC_NR },
{ "pos_embd", OFFLOAD_FUNC_NR },
{ "KQ_mask", OFFLOAD_FUNC_NR },
{ "K_shift", OFFLOAD_FUNC_NR },
{ "K_shifted", OFFLOAD_FUNC_NR },
{ "inp_pos", OFFLOAD_FUNC_KQ }, // this is often used for KQ ops (e.g. rope)
{ "KQ_scale", OFFLOAD_FUNC_KQ },
{ "KQ_mask", OFFLOAD_FUNC_KQ },
{ "K_shift", OFFLOAD_FUNC_KQ },
{ "K_shifted", OFFLOAD_FUNC_KQ },
{ "inp_norm", OFFLOAD_FUNC_NR },
{ "inp_norm_w", OFFLOAD_FUNC_NR },