mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 25 Packages Projects Releases Wiki Activity

model : full modern bert support (#18330)

Browse Source 
* full modern bert support

* added gelu op in rank pooling for modern bert

* still working on stuff, added mean calculation before classifier head

* Update convert_hf_to_gguf.py

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* first layer is dense, as per modern bert research paper

* Update src/llama-graph.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* fixed set input for mean pooling to check if pooling type is ranking since modern bert does mean & rank

* Update src/llama-graph.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* Update convert_hf_to_gguf.py

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
This commit is contained in:
Ryan Mangeno
2026-02-19 02:52:21 -05:00
committed by GitHub
parent 3bb2fcc856
commit c0d0430340
12 changed files with 54 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-model.cpp
+6 -5
View File
@@ -908,7 +908,7 @@ void llama_model::load_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_ROPE_FREQ_BASE_SWA, hparams.rope_freq_base_train_swa);
ml.get_key_or_arr(LLM_KV_ATTENTION_SLIDING_WINDOW_PATTERN, swa_period, false);
hparams.set_swa_pattern(swa_period);
hparams.set_swa_pattern(swa_period, true);
} else {
hparams.swa_type = LLAMA_SWA_TYPE_NONE;
}
@@ -3513,9 +3513,10 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
}
cls = create_tensor(tn(LLM_TENSOR_CLS, "weight"), {n_embd, n_embd}, TENSOR_NOT_REQUIRED);
cls_out = create_tensor(tn(LLM_TENSOR_CLS_OUT, "weight"), {n_embd, hparams.n_cls_out}, TENSOR_NOT_REQUIRED);
cls_out_b = create_tensor(tn(LLM_TENSOR_CLS_OUT, "bias"), {hparams.n_cls_out}, TENSOR_NOT_REQUIRED);
cls_out = create_tensor(tn(LLM_TENSOR_CLS_OUT, "weight"), {n_embd, hparams.n_cls_out}, TENSOR_NOT_REQUIRED);
cls_out_b = create_tensor(tn(LLM_TENSOR_CLS_OUT, "bias"), {hparams.n_cls_out}, TENSOR_NOT_REQUIRED);
cls = create_tensor(tn(LLM_TENSOR_CLS, "weight"), {n_embd, n_embd}, TENSOR_NOT_REQUIRED);
cls_norm = create_tensor(tn(LLM_TENSOR_CLS_NORM, "weight"), {n_embd}, TENSOR_NOT_REQUIRED);
} break;
case LLM_ARCH_NEO_BERT:
@@ -8734,7 +8735,7 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
}
// add on pooling layer
llm->build_pooling(cls, cls_b, cls_out, cls_out_b);
llm->build_pooling(cls, cls_b, cls_out, cls_out_b, cls_norm);
// add backend sampling layers (if any)
llm->build_sampling();