Skip to content

Commit

Permalink
llm : fix llm_build_kqv taking unused tensor (benign, ggerganov#3837)
Browse files Browse the repository at this point in the history
  • Loading branch information
@ggerganov
ggerganov committed Nov 1, 2023
1 parent 523e49b commit c43c2da
Showing 1 changed file with 9 additions and 10 deletions.
19 changes: 9 additions & 10 deletions llama.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3345,7 +3345,6 @@ static struct ggml_tensor * llm_build_ffn(
// if max_alibi_bias > 0 then apply ALiBi
static struct ggml_tensor * llm_build_kqv(
struct ggml_context * ctx,
struct ggml_tensor * cur,
const llama_hparams & hparams,
const llama_kv_cache & kv,
struct ggml_tensor * wo,
Expand Down Expand Up @@ -3411,7 +3410,7 @@ static struct ggml_tensor * llm_build_kqv(
struct ggml_tensor * kqv_merged = ggml_permute(ctx, kqv, 0, 2, 1, 3);
cb(kqv_merged, "kqv_merged", il);

cur = ggml_cont_2d(ctx, kqv_merged, n_embd, n_tokens);
struct ggml_tensor * cur = ggml_cont_2d(ctx, kqv_merged, n_embd, n_tokens);
cb(cur, "kqv_merged_cont", il);

cur = ggml_mul_mat(ctx, wo, cur);
Expand Down Expand Up @@ -3565,7 +3564,7 @@ struct llm_build_context {

llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);

cur = llm_build_kqv(ctx0, cur, hparams, kv_self,
cur = llm_build_kqv(ctx0, hparams, kv_self,
model.layers[il].wo, NULL,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, cb, il);
cb(cur, "kqv_out", il);
Expand Down Expand Up @@ -3677,7 +3676,7 @@ struct llm_build_context {
// apply ALiBi for 13B model
const float max_alibi_bias = model.type == MODEL_13B ? 8.0f : -1.0f;

cur = llm_build_kqv(ctx0, cur, hparams, kv_self,
cur = llm_build_kqv(ctx0, hparams, kv_self,
model.layers[il].wo, NULL,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, max_alibi_bias, cb, il);
cb(cur, "kqv_out", il);
Expand Down Expand Up @@ -3795,7 +3794,7 @@ struct llm_build_context {

llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);

cur = llm_build_kqv(ctx0, attn_norm, hparams, kv_self,
cur = llm_build_kqv(ctx0, hparams, kv_self,
model.layers[il].wo, NULL,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, cb, il);
cb(cur, "kqv_out", il);
Expand Down Expand Up @@ -3895,7 +3894,7 @@ struct llm_build_context {

llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);

cur = llm_build_kqv(ctx0, cur, hparams, kv_self,
cur = llm_build_kqv(ctx0, hparams, kv_self,
model.layers[il].wo, model.layers[il].bo,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, cb, il);
cb(cur, "kqv_out", il);
Expand Down Expand Up @@ -4100,7 +4099,7 @@ struct llm_build_context {
llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);

// TODO: not tested, could be broken
cur = llm_build_kqv(ctx0, Q, hparams, kv_self,
cur = llm_build_kqv(ctx0, hparams, kv_self,
model.layers[il].wo, model.layers[il].bo,
Q, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, cb, il);
cb(cur, "kqv_out", il);
Expand Down Expand Up @@ -4191,7 +4190,7 @@ struct llm_build_context {

llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);

cur = llm_build_kqv(ctx0, Qcur, hparams, kv_self,
cur = llm_build_kqv(ctx0, hparams, kv_self,
model.layers[il].wo, NULL,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, 8.0f, cb, il);
cb(cur, "kqv_out", il);
Expand Down Expand Up @@ -4288,7 +4287,7 @@ struct llm_build_context {

llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);

cur = llm_build_kqv(ctx0, Qcur, hparams, kv_self,
cur = llm_build_kqv(ctx0, hparams, kv_self,
model.layers[il].wo, model.layers[il].bo,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, 8.0f, cb, il);
cb(cur, "kqv_out", il);
Expand Down Expand Up @@ -4382,7 +4381,7 @@ struct llm_build_context {

llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);

cur = llm_build_kqv(ctx0, Qcur, hparams, kv_self,
cur = llm_build_kqv(ctx0, hparams, kv_self,
model.layers[il].wo, NULL,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, hparams.f_max_alibi_bias, cb, il);
cb(cur, "kqv_out", il);
Expand Down

0 comments on commit c43c2da

Please sign in to comment.