happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

continuing

This commit is contained in:
ryan-mangeno 2025-08-26 12:38:38 -04:00
parent 41b6864333
commit cc3d7abab4
2 changed files with 7 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml.c
View File

@ -3015,6 +3015,10 @@ struct ggml_tensor * ggml_mul_mat(
struct ggml_context * ctx, struct ggml_context * ctx,
struct ggml_tensor * a, struct ggml_tensor * a,
struct ggml_tensor * b) { struct ggml_tensor * b) {
printf("Up: {%lld, %lld}\n", a->ne[0], a->ne[1]);
printf("Cur: {%lld, %lld}\n", b->ne[0], b->ne[1]);
GGML_ASSERT(ggml_can_mul_mat(a, b)); GGML_ASSERT(ggml_can_mul_mat(a, b));
GGML_ASSERT(!ggml_is_transposed(a)); GGML_ASSERT(!ggml_is_transposed(a));

4
src/llama-graph.cpp
View File

@ -644,6 +644,8 @@ ggml_tensor * llm_graph_context::build_ffn(
llm_ffn_op_type type_op, llm_ffn_op_type type_op,
llm_ffn_gate_type type_gate, llm_ffn_gate_type type_gate,
int il) const { int il) const {
ggml_tensor * tmp = up ? build_lora_mm(up, cur) : cur; ggml_tensor * tmp = up ? build_lora_mm(up, cur) : cur;
cb(tmp, "ffn_up", il); cb(tmp, "ffn_up", il);
@ -1377,7 +1379,7 @@ ggml_tensor * llm_graph_context::build_attn(
// TODO: if ubatch.equal_seqs() == true, we can split the three tensors below into ubatch.n_seqs_unq streams // TODO: if ubatch.equal_seqs() == true, we can split the three tensors below into ubatch.n_seqs_unq streams
LLAMA_LOG_INFO("ubatch.equal_seqs() = %d, n_seqs = %d\n", ubatch.equal_seqs(), ubatch.n_seqs); LLAMA_LOG_INFO("ubatch.equal_seqs() = %d, n_seqs = %d\n", ubatch.equal_seqs(), ubatch.n_seqs);
assert(!ubatch.equal_seqs()); // sassert(!ubatch.equal_seqs());
ggml_tensor * q = q_cur; ggml_tensor * q = q_cur;
ggml_tensor * k = k_cur; ggml_tensor * k = k_cur;