ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

pooling : make cls_b and cls_out_b optional (#14165)

Browse source 
Co-authored-by: dinhhuy <huy.dinh@brains-tech.co.jp>
This commit is contained in:
Đinh Trọng Huy 2025-06-13 17:34:08 +09:00 • committed by GitHub
parent ffad043973
commit d714dadb57
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 13 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

19
src/llama-graph.cpp
View file

@ -1556,23 +1556,30 @@ void llm_graph_context::build_pooling(
ggml_tensor * inp_cls = build_inp_cls(); ggml_tensor * inp_cls = build_inp_cls();
inp = ggml_get_rows(ctx0, inp, inp_cls); inp = ggml_get_rows(ctx0, inp, inp_cls);
if (cls != nullptr && cls_b != nullptr) { if (cls) {
// classification head // classification head
// https://github.com/huggingface/transformers/blob/5af7d41e49bbfc8319f462eb45253dcb3863dfb7/src/transformers/models/roberta/modeling_roberta.py#L1566 // https://github.com/huggingface/transformers/blob/5af7d41e49bbfc8319f462eb45253dcb3863dfb7/src/transformers/models/roberta/modeling_roberta.py#L1566
cur = ggml_add(ctx0, ggml_mul_mat(ctx0, cls, inp), cls_b); cur = ggml_mul_mat(ctx0, cls, inp);
if (cls_b) {
cur = ggml_add(ctx0, cur, cls_b);
}
cur = ggml_tanh(ctx0, cur); cur = ggml_tanh(ctx0, cur);
// some models don't have `cls_out`, for example: https://huggingface.co/jinaai/jina-reranker-v1-tiny-en // some models don't have `cls_out`, for example: https://huggingface.co/jinaai/jina-reranker-v1-tiny-en
// https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/blob/cb5347e43979c3084a890e3f99491952603ae1b7/modeling_bert.py#L884-L896 // https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/blob/cb5347e43979c3084a890e3f99491952603ae1b7/modeling_bert.py#L884-L896
if (cls_out) { if (cls_out) {
GGML_ASSERT(cls_out_b != nullptr); cur = ggml_mul_mat(ctx0, cls_out, cur);
cur = ggml_add(ctx0, ggml_mul_mat(ctx0, cls_out, cur), cls_out_b); if (cls_out_b) {
cur = ggml_add(ctx0, cur, cls_out_b);
}
} }
} else if (cls_out) { } else if (cls_out) {
// Single layer classification head (direct projection) // Single layer classification head (direct projection)
// https://github.com/huggingface/transformers/blob/f4fc42216cd56ab6b68270bf80d811614d8d59e4/src/transformers/models/bert/modeling_bert.py#L1476 // https://github.com/huggingface/transformers/blob/f4fc42216cd56ab6b68270bf80d811614d8d59e4/src/transformers/models/bert/modeling_bert.py#L1476
GGML_ASSERT(cls_out_b != nullptr); cur = ggml_mul_mat(ctx0, cls_out, inp);
cur = ggml_add(ctx0, ggml_mul_mat(ctx0, cls_out, inp), cls_out_b); if (cls_out_b) {
cur = ggml_add(ctx0, cur, cls_out_b);
}
} else { } else {
GGML_ABORT("RANK pooling requires either cls+cls_b or cls_out+cls_out_b"); GGML_ABORT("RANK pooling requires either cls+cls_b or cls_out+cls_out_b");
} }