ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-25 17:09:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

WIP: Cohere2

Browse Source
This commit is contained in:
Iwan Kawrakow
2025-12-22 09:57:38 +00:00
parent d50ef0165e
commit aa3f14b963
3 changed files with 69 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
ggml/src/ggml-backend.cpp
View File

@@ -1414,7 +1414,6 @@ static void ggml_backend_sched_split_graph(ggml_backend_sched_t sched, struct gg
// do not overwrite user assignments
if (*leaf_backend_id == -1) {
*leaf_backend_id = ggml_backend_sched_backend_id_from_cur(sched, leaf);
//printf("Pass 1: assigned backend %d to leaf %d, %s\n", *leaf_backend_id, i, graph->leafs[i]->name);
}
}
@@ -1443,6 +1442,31 @@ static void ggml_backend_sched_split_graph(ggml_backend_sched_t sched, struct gg
*node_backend_id = *this_node_backend_id;
}
}
else if (node->op == GGML_OP_MUL && node->src[0]->op == GGML_OP_NORM) {
// This is a hack for Cohere2. Without this hack the scheduler creates
// totally nonsensical splits for that arch
int * src1_id = &tensor_backend_id(node->src[1]);
if (*src1_id >= 0) {
int * src0_id = &tensor_backend_id(node->src[0]);
int * dst_id = &tensor_backend_id(node);
*src0_id = *src1_id;
*dst_id = *src1_id;
// For some reason that I don't understand, we can have norm backend already assigned
// at this point. How? That's why this more logical approach of first checking is commented out
//if (*src0_id < 0) {
// *src0_id = *src1_id;
//} else {
// printf("Oops: backend_id_src0(%s) = %d, backend_id_src1(%s) = %d\n", node->src[0]->name, *src0_id, node->src[1]->name, *src1_id);
// //GGML_ASSERT(*src0_id == *src1_id);
//}
//if (*dst_id < 0) {
// *dst_id = *src1_id;
//} else {
// printf("Oops: backend_id_dst(%s) = %d, backend_id_src1(%s) = %d\n", node->name, *dst_id, node->src[1]->name, *src1_id);
// //GGML_ASSERT(*dst_id == *src1_id);
//}
}
}
// do not overwrite user assignments
if (*node_backend_id == -1) {
*node_backend_id = ggml_backend_sched_backend_id_from_cur(sched, node);