vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

bert : add some missing graph callbacks

Browse source
This commit is contained in:
Jared Van Bortel 2024-02-08 21:22:23 -05:00
parent e78388d39a
commit b14c457fb4
1 changed files with 3 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

3
llama.cpp
View file

@ -5752,6 +5752,7 @@ struct llm_build_context {
// embed layer norm
inpL = llm_build_norm(ctx0, inpL, hparams, model.tok_norm, model.tok_norm_b, LLM_NORM, cb, -1);
cb(inpL, "inp_norm", -1);
// KQ_mask (mask for 1 head, it will be broadcasted to all heads)
struct ggml_tensor * KQ_mask = ggml_view_2d(ctx0, lctx.inp_KQ_mask, n_kv, n_tokens, n_kv*ggml_type_size(lctx.inp_KQ_mask->type), 0);
@ -5788,6 +5789,7 @@ struct llm_build_context {
cur = llm_build_norm(ctx0, cur, hparams, model.layers[il].attn_norm, model.layers[il].attn_norm_b, LLM_NORM, cb, il);
struct ggml_tensor * ffn_inp = cur;
cb(ffn_inp, "ffn_inp", il);
// feed-forward network
cur = llm_build_ffn(ctx0, cur,
@ -5796,6 +5798,7 @@ struct llm_build_context {
model.layers[il].ffn_down, model.layers[il].ffn_down_b,
NULL,
LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
cb(cur, "ffn_out", il);
// attentions bypass the intermediate layer
cur = ggml_add(ctx0, cur, ffn_inp);