Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-24 13:28:50 +01:00
Code Issues Packages Projects Releases Wiki Activity

llama : remove redundant GQA check (#4796)

Browse Source
This commit is contained in:
Georgi Gerganov 2024-01-07 11:21:53 +02:00 committed by GitHub
parent 72d8407b36
commit 3c36213df8
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 0 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
llama.cpp
View File

@ -4776,7 +4776,6 @@ struct llm_build_context {
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
GGML_ASSERT(n_embd_gqa == n_embd);
struct ggml_tensor * cur; struct ggml_tensor * cur;
struct ggml_tensor * inpL; struct ggml_tensor * inpL;
@ -4900,7 +4899,6 @@ struct llm_build_context {
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
GGML_ASSERT(n_embd_gqa == n_embd);
struct ggml_tensor * cur; struct ggml_tensor * cur;
struct ggml_tensor * pos; struct ggml_tensor * pos;
@ -5001,7 +4999,6 @@ struct llm_build_context {
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
GGML_ASSERT(n_embd_gqa == n_embd);
const int64_t n_rot = n_embd_head_k / 2; const int64_t n_rot = n_embd_head_k / 2;
@ -5215,7 +5212,6 @@ struct llm_build_context {
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
GGML_ASSERT(n_embd_gqa == n_embd);
struct ggml_tensor * cur; struct ggml_tensor * cur;
struct ggml_tensor * inpL; struct ggml_tensor * inpL;
@ -5308,7 +5304,6 @@ struct llm_build_context {
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
GGML_ASSERT(n_embd_gqa == n_embd);
struct ggml_tensor * cur; struct ggml_tensor * cur;
struct ggml_tensor * inpL; struct ggml_tensor * inpL;
@ -5404,7 +5399,6 @@ struct llm_build_context {
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
GGML_ASSERT(n_embd_gqa == n_embd);
struct ggml_tensor * cur; struct ggml_tensor * cur;
struct ggml_tensor * inpL; struct ggml_tensor * inpL;
@ -5731,7 +5725,6 @@ struct llm_build_context {
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
GGML_ASSERT(n_embd_gqa == n_embd);
struct ggml_tensor * cur; struct ggml_tensor * cur;
struct ggml_tensor * attn_norm_output; struct ggml_tensor * attn_norm_output;
@ -5955,7 +5948,6 @@ struct llm_build_context {
const int64_t n_embd_head = hparams.n_embd_head_v; const int64_t n_embd_head = hparams.n_embd_head_v;
const int64_t n_embd_gqa = hparams.n_embd_v_gqa(); const int64_t n_embd_gqa = hparams.n_embd_v_gqa();
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k); GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
GGML_ASSERT(n_embd_gqa == n_embd);
struct ggml_tensor * cur; struct ggml_tensor * cur;
struct ggml_tensor * pos; struct ggml_tensor * pos;