model : add support for JinaBertModel with non-gated ffn (ggml-org#18475)

* WIP: Initial commit for fixing JinaBert original FF type support

* convert: add jina-v2-de tokenizer variant for German_Semantic_V3

* convert: fix token collision in BERT phantom vocab conversion

* convert: add feed_forward_type metadata

* model: add feed_forward_type metadata for jina-bert-v2

* model: jina-bert-v2 support standard GELU FFN variant

* model: remove ffn_type, detect FFN variant from tensor dimensions

* Update src/llama-model.cpp

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

* Update src/llama-model.cpp

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

* Update src/models/bert.cpp

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

* Update src/models/bert.cpp

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

* revert collision fix to be handled in separate PR

---------

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

Author: o7si

convert_hf_to_gguf.py

@@ -1062,6 +1062,9 @@ def get_vocab_base_pre(self, tokenizer) -> str:
         if chkhsh == "66b8d4e19ab16c3bfd89bce5d785fb7e0155e8648708a1f42077cb9fe002c273":
             # ref: https://huggingface.co/alvarobartt/grok-2-tokenizer
             res = "grok-2"
+        if chkhsh == "b3d1dd861f1d4c5c0d2569ce36baf3f90fe8a102db3de50dd71ff860d91be3df":
+            # ref: https://huggingface.co/aari1995/German_Semantic_V3
+            res = "jina-v2-de"
         if chkhsh == "0ef9807a4087ebef797fc749390439009c3b9eda9ad1a097abbe738f486c01e5":
             # ref: https://huggingface.co/meta-llama/Meta-Llama-3-8B
             res = "llama-bpe"

convert_hf_to_gguf_update.py

@@ -166,6 +166,8 @@ class TOKENIZER_TYPE(IntEnum):
     {"name": "kimi-k2",   "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/moonshotai/Kimi-K2-Base",   "chkhsh": "81212dc7cdb7e0c1074ca62c5aeab0d43c9f52b8a737be7b12a777c953027890"},
     {"name": "qwen2",     "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen3-Embedding-0.6B", "chkhsh": "d4540891389ea895b53b399da6ac824becc30f2fba0e9ddbb98f92e55ca0e97c"},
     {"name": "grok-2",    "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/alvarobartt/grok-2-tokenizer", "chkhsh": "66b8d4e19ab16c3bfd89bce5d785fb7e0155e8648708a1f42077cb9fe002c273"},
+    # jina-v2-de variants
+    {"name": "jina-v2-de", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/aari1995/German_Semantic_V3", "chkhsh": "b3d1dd861f1d4c5c0d2569ce36baf3f90fe8a102db3de50dd71ff860d91be3df"},
 ]
 
 

src/llama-model.cpp

@@ -3322,7 +3322,14 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
                         layer.attn_norm_2_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM_2, "bias",   i), {n_embd}, TENSOR_NOT_REQUIRED);
 
                         layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, TENSOR_NOT_REQUIRED);
-                        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd, layer.ffn_gate ? n_ff : n_ff * 2}, 0);
+
+                        const auto tn_ffn_up_weight = tn(LLM_TENSOR_FFN_UP, "weight", i);
+                        ggml_tensor * t_ffn_up = ml.get_tensor_meta(tn_ffn_up_weight.str().c_str());
+                        const int64_t n_ffn_up = t_ffn_up ? t_ffn_up->ne[1] : n_ff;
+
+                        GGML_ASSERT(n_ffn_up == n_ff || n_ffn_up == n_ff * 2);
+                        layer.ffn_up   = create_tensor(tn_ffn_up_weight, {n_embd, n_ffn_up}, 0);
+                        layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ffn_up}, TENSOR_NOT_REQUIRED);
 
                         layer.ffn_down   = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);
                         layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias",   i), {n_embd}, 0);

src/models/bert.cpp

@@ -142,11 +142,13 @@ llm_build_bert::llm_build_bert(const llama_model & model, const llm_graph_params
                     LLM_FFN_GELU, LLM_FFN_SEQ, il);
             cb(cur, "ffn_out", il);
         } else if (model.arch == LLM_ARCH_JINA_BERT_V2) {
+            const bool up_contains_gate = !model.layers[il].ffn_gate && model.layers[il].ffn_up->ne[1] != hparams.n_ff();
+            auto type_op = up_contains_gate ? LLM_FFN_GEGLU : LLM_FFN_GELU;
             cur = build_ffn(cur,
-                    model.layers[il].ffn_up, NULL, NULL,
+                    model.layers[il].ffn_up, model.layers[il].ffn_up_b, NULL,
                     model.layers[il].ffn_gate, NULL, NULL,
                     model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL, NULL,
-                    model.layers[il].ffn_gate ? LLM_FFN_GELU : LLM_FFN_GEGLU, LLM_FFN_PAR, il);
+                    type_op, LLM_FFN_PAR, il);
             cb(cur, "ffn_out", il);
         } else {
             cur = build_ffn(cur,