Modern Bert Support

convert_hf_to_gguf.py

@@ -1204,6 +1204,9 @@ def get_vocab_base_pre(self, tokenizer) -> str:
         if chkhsh == "a1e163ecab2e718a4c829d1148b6e86824ec36163bb71941c3dca9cd5ac25756":
             # ref: https://huggingface.co/JetBrains/Mellum-4b-base
             res = "mellum"
+        if chkhsh == "a0b64b4385f123663873756336c085744376d015ff328bb1d901598f63c44152":
+            # ref: https://huggingface.co/answerdotai/ModernBERT-base
+            res = "modern-bert"
         if chkhsh == "49fc0303c9e0d2c2c565c510f64b2d9b271276acdcdadff733249eda9f7d59df":
             # ref: https://huggingface.co/arcee-ai/Trinity-Tokenizer
             res = "afmoe"
@@ -9991,6 +9994,45 @@ def prepare_tensors(self):
                 raise ValueError(f"Unprocessed experts: {experts}")
 
 
+@ModelBase.register("ModernBertModel", "ModernBertForMaskedLM", "ModernBertForSequenceClassification")
+class ModernBertModel(BertModel):
+    model_arch = gguf.MODEL_ARCH.MODERN_BERT
+
+    def set_vocab(self):
+        self.gguf_writer.add_add_bos_token(True)
+        self.gguf_writer.add_add_eos_token(True)
+        self.gguf_writer.add_add_sep_token(True)
+        self._set_vocab_gpt2()
+
+    def set_gguf_parameters(self):
+        super().set_gguf_parameters()
+        self.gguf_writer.add_dense_every_n_layers(self.hparams["global_attn_every_n_layers"])
+        self.gguf_writer.add_sliding_window(self.hparams["local_attention"])
+        self.gguf_writer.add_rope_freq_base_swa(self.rope_parameters.get("sliding_attention", {"rope_theta": self.hparams.get("local_rope_theta")})["rope_theta"])
+        self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.NONE)
+        self.gguf_writer.add_vocab_size(self.hparams["vocab_size"])
+
+    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
+        # rename custom "head" layers to standard bert "cls.predictions" names for compatibility
+        if name == "head.norm.weight":
+            name = "cls.predictions.transform.LayerNorm.weight"
+        elif name == "head.norm.bias":
+            name = "cls.predictions.transform.LayerNorm.bias"
+        elif name == "head.dense.weight":
+            name = "cls.predictions.transform.dense.weight"
+        elif name == "head.dense.bias":
+            name = "cls.predictions.transform.dense.bias"
+
+        # These layers act as MLM head, so we don't need them
+        if name.startswith("decoder."):
+            return []
+
+        if name.startswith("model."):
+            name = name[6:]
+
+        return super().modify_tensors(data_torch, name, bid)
+
+
 @ModelBase.register("ApertusForCausalLM")
 class ApertusModel(LlamaModel):
     model_arch = gguf.MODEL_ARCH.APERTUS

convert_hf_to_gguf_update.py

@@ -139,6 +139,7 @@ class TOKENIZER_TYPE(IntEnum):
     {"name": "lfm2",             "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LiquidAI/LFM2-Tokenizer"},
     {"name": "exaone4",          "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B", },
     {"name": "mellum",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/JetBrains/Mellum-4b-base", },
+    {"name": "modern-bert",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/answerdotai/ModernBERT-base", },
     {"name": "afmoe",            "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/arcee-ai/Trinity-Tokenizer", },
     {"name": "bailingmoe2",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/inclusionAI/Ling-mini-base-2.0", },
     {"name": "granite-docling",  "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/ibm-granite/granite-docling-258M", },

gguf-py/gguf/constants.py

@@ -176,11 +176,13 @@ class Attention:
         SHARED_KV_LAYERS             = "{arch}.attention.shared_kv_layers"
         SLIDING_WINDOW_PATTERN       = "{arch}.attention.sliding_window_pattern"
         TEMPERATURE_SCALE            = "{arch}.attention.temperature_scale"
+        DENSE_EVERY_N_LAYERS         = "{arch}.attention.dense_every_n_layers"
 
     class Rope:
         DIMENSION_COUNT          = "{arch}.rope.dimension_count"
         DIMENSION_SECTIONS       = "{arch}.rope.dimension_sections"
         FREQ_BASE                = "{arch}.rope.freq_base"
+        FREQ_BASE_SWA            = "{arch}.rope.freq_base_swa"
         SCALING_TYPE             = "{arch}.rope.scaling.type"
         SCALING_FACTOR           = "{arch}.rope.scaling.factor"
         SCALING_ATTN_FACTOR      = "{arch}.rope.scaling.attn_factor"
@@ -354,6 +356,7 @@ class MODEL_ARCH(IntEnum):
     STARCODER        = auto()
     REFACT           = auto()
     BERT             = auto()
+    MODERN_BERT      = auto()
     NOMIC_BERT       = auto()
     NOMIC_BERT_MOE   = auto()
     NEO_BERT         = auto()
@@ -747,6 +750,7 @@ class MODEL_TENSOR(IntEnum):
     MODEL_ARCH.STARCODER:        "starcoder",
     MODEL_ARCH.REFACT:           "refact",
     MODEL_ARCH.BERT:             "bert",
+    MODEL_ARCH.MODERN_BERT:      "modern-bert",
     MODEL_ARCH.NOMIC_BERT:       "nomic-bert",
     MODEL_ARCH.NOMIC_BERT_MOE:   "nomic-bert-moe",
     MODEL_ARCH.NEO_BERT:         "neo-bert",
@@ -1367,6 +1371,20 @@ class MODEL_TENSOR(IntEnum):
         MODEL_TENSOR.CLS,
         MODEL_TENSOR.CLS_OUT,
     ],
+    MODEL_ARCH.MODERN_BERT: [
+        MODEL_TENSOR.TOKEN_EMBD,
+        MODEL_TENSOR.TOKEN_EMBD_NORM,
+        MODEL_TENSOR.OUTPUT_NORM,
+        MODEL_TENSOR.ATTN_NORM,
+        MODEL_TENSOR.ATTN_OUT,
+        MODEL_TENSOR.ATTN_QKV,
+        MODEL_TENSOR.POS_EMBD,
+        MODEL_TENSOR.FFN_UP,
+        MODEL_TENSOR.FFN_DOWN,
+        MODEL_TENSOR.FFN_NORM,
+        MODEL_TENSOR.CLS,
+        MODEL_TENSOR.CLS_OUT,
+    ],
     MODEL_ARCH.NOMIC_BERT: [
         MODEL_TENSOR.TOKEN_EMBD,
         MODEL_TENSOR.TOKEN_EMBD_NORM,

gguf-py/gguf/gguf_writer.py

@@ -777,6 +777,9 @@ def add_shared_kv_layers(self, value: int) -> None:
     def add_sliding_window_pattern(self, value: Sequence[bool]) -> None:
         self.add_array(Keys.Attention.SLIDING_WINDOW_PATTERN.format(arch=self.arch), value)
 
+    def add_dense_every_n_layers(self, value: int) -> None:
+        self.add_uint32(Keys.Attention.DENSE_EVERY_N_LAYERS.format(arch=self.arch), value)
+
     def add_dense_features_dims(self, dense:str, in_f:int, out_f:int) -> None:
         self.add_uint32(Keys.LLM.DENSE_FEAT_IN_SIZE.format(arch=self.arch, dense=dense), in_f)
         self.add_uint32(Keys.LLM.DENSE_FEAT_OUT_SIZE.format(arch=self.arch, dense=dense), out_f)
@@ -886,6 +889,9 @@ def add_iclr_lora_rank(self, length: int) -> None:
     def add_value_residual_mix_lora_rank(self, length: int) -> None:
         self.add_uint32(Keys.Attention.VALUE_RESIDUAL_MIX_LORA_RANK.format(arch=self.arch), length)
 
+    def add_rope_freq_base_swa(self, value: float) -> None:
+        self.add_float32(Keys.Rope.FREQ_BASE_SWA.format(arch=self.arch), value)
+
     def add_gate_lora_rank(self, length: int) -> None:
         self.add_uint32(Keys.Attention.GATE_LORA_RANK.format(arch=self.arch), length)
 

gguf-py/gguf/tensor_mapping.py

@@ -17,6 +17,7 @@ class TensorNameMap:
             "embed_tokens",                              # embeddinggemma
             "tok_embeddings",                            # llama-pth
             "embeddings.word_embeddings",                # bert nomic-bert
+            "embeddings.tok_embeddings",                 # modern-bert
             "language_model.embedding.word_embeddings",  # persimmon
             "wte",                                       # gpt2
             "transformer.embd.wte",                      # phi2
@@ -46,6 +47,7 @@ class TensorNameMap:
         MODEL_TENSOR.TOKEN_EMBD_NORM: (
             "word_embeddings_layernorm",  # bloom
             "embeddings.LayerNorm",       # bert
+            "embeddings.norm",            # modern-bert
             "emb_ln",                     # nomic-bert
             "transformer.norm",           # openelm
             "rwkv.blocks.0.pre_ln",       # rwkv
@@ -104,6 +106,7 @@ class TensorNameMap:
             "backbone.final_layer_norm",               # wavtokenizer
             "model.norm",                              # llama4
             "model.transformer.ln_f",                  # llada
+            "final_norm",                              # modern-bert
             "model.norm",                              # cogvlm
         ),
 
@@ -151,6 +154,7 @@ class TensorNameMap:
             "model.layers.{bid}.input_layernorm",                   # llama4
             "layers.{bid}.input_layernorm",                         # embeddinggemma
             "transformer_encoder.{bid}.attention_norm",             # neobert
+            "layers.{bid}.attn_norm",                               # modern-bert
             "model.layers.{bid}.operator_norm",                     # lfm2
             "model.transformer.blocks.{bid}.attn_norm",             # llada
             "layers.{bid}.input_layernorm",                         # qwen3-embedding
@@ -187,6 +191,7 @@ class TensorNameMap:
             "encoder.layers.{bid}.self_attention.query_key_value",                 # chatglm
             "transformer.layers.{bid}.attn.qkv_proj",                              # openelm
             "transformer_encoder.{bid}.qkv",                                       # neobert
+            "layers.{bid}.attn.Wqkv",                                              # modern-bert
             "model.layers.{bid}.self_attn.language_expert_query_key_value",        # cogvlm
         ),
 
@@ -261,6 +266,7 @@ class TensorNameMap:
             "model.layers.{bid}.self_attn.linear_attn",                     # deci
             "layers.{bid}.attention.wo",                                    # llama-pth
             "encoder.layer.{bid}.attention.output.dense",                   # bert
+            "layers.{bid}.attn.Wo",                                         # modern-bert
             "transformer.layer.{bid}.attention.out_lin",                    # distillbert
             "transformer.h.{bid}.attn.out_proj",                            # gpt-j
             "language_model.encoder.layers.{bid}.self_attention.dense",     # persimmon
@@ -344,6 +350,7 @@ class TensorNameMap:
             "layers.{bid}.post_attention_layernorm",                         # qwen3-embedding
             "model.layers.{bid}.feedforward_layernorm",                      # apertus
             "model.layers.{bid}.pre_mlp_layernorm",                          # kormo
+            "layers.{bid}.mlp_norm"                                          # modern-bert
         ),
 
         # Pre feed-forward norm
@@ -407,6 +414,7 @@ class TensorNameMap:
             "layers.{bid}.mlp.up_proj",                               # embeddinggemma
             "layers.{bid}.feed_forward.w3",                           # llama-pth
             "encoder.layer.{bid}.intermediate.dense",                 # bert
+            "layers.{bid}.mlp.Wi",                                    # modern-bert
             "transformer.layer.{bid}.ffn.lin1",                       # distillbert
             "transformer.h.{bid}.mlp.fc_in",                          # gpt-j
             "transformer.h.{bid}.mlp.linear_3",                       # refact
@@ -521,6 +529,7 @@ class TensorNameMap:
             "layers.{bid}.mlp.down_proj",                             # embeddinggemma
             "layers.{bid}.feed_forward.w2",                           # llama-pth
             "encoder.layer.{bid}.output.dense",                       # bert
+            "layers.{bid}.mlp.Wo",                                    # modern-bert
             "transformer.layer.{bid}.ffn.lin2",                       # distillbert
             "transformer.h.{bid}.mlp.fc_out",                         # gpt-j
             "language_model.encoder.layers.{bid}.mlp.dense_4h_to_h",  # persimmon

src/CMakeLists.txt

@@ -90,6 +90,7 @@ add_library(llama
             models/mamba.cpp
             models/minicpm3.cpp
             models/minimax-m2.cpp
+            models/modern-bert.cpp
             models/mpt.cpp
             models/nemotron-h.cpp
             models/nemotron.cpp

src/llama-arch.cpp

@@ -20,6 +20,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_STARCODER,        "starcoder"        },
     { LLM_ARCH_REFACT,           "refact"           },
     { LLM_ARCH_BERT,             "bert"             },
+    { LLM_ARCH_MODERN_BERT,      "modern-bert"      },
     { LLM_ARCH_NOMIC_BERT,       "nomic-bert"       },
     { LLM_ARCH_NOMIC_BERT_MOE,   "nomic-bert-moe"   },
     { LLM_ARCH_NEO_BERT,         "neo-bert"         },
@@ -204,6 +205,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
     { LLM_KV_ATTENTION_GATE_LORA_RANK,               "%s.attention.gate_lora_rank"               },
     { LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT,       "%s.attention.relative_buckets_count"       },
     { LLM_KV_ATTENTION_SLIDING_WINDOW,               "%s.attention.sliding_window"               },
+    { LLM_KV_ATTENTION_DENSE_EVERY_N_LAYERS,         "%s.attention.dense_every_n_layers"         },
     { LLM_KV_ATTENTION_SCALE,                        "%s.attention.scale"                        },
     { LLM_KV_ATTENTION_OUTPUT_SCALE,                 "%s.attention.output_scale"                 },
     { LLM_KV_ATTENTION_TEMPERATURE_LENGTH,           "%s.attention.temperature_length"           },
@@ -214,6 +216,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
     { LLM_KV_ROPE_DIMENSION_COUNT,          "%s.rope.dimension_count"                 },
     { LLM_KV_ROPE_DIMENSION_SECTIONS,       "%s.rope.dimension_sections"              },
     { LLM_KV_ROPE_FREQ_BASE,                "%s.rope.freq_base"                       },
+    { LLM_KV_ROPE_FREQ_BASE_SWA,            "%s.rope.freq_base_swa"                   },
     { LLM_KV_ROPE_SCALE_LINEAR,             "%s.rope.scale_linear"                    },
     { LLM_KV_ROPE_SCALING_TYPE,             "%s.rope.scaling.type"                    },
     { LLM_KV_ROPE_SCALING_FACTOR,           "%s.rope.scaling.factor"                  },
@@ -778,6 +781,21 @@ static std::set<llm_tensor> llm_get_tensor_names(llm_arch arch) {
                 LLM_TENSOR_CLS,
                 LLM_TENSOR_CLS_OUT,
             };
+        case LLM_ARCH_MODERN_BERT:
+            return {
+                LLM_TENSOR_TOKEN_EMBD,
+                LLM_TENSOR_TOKEN_EMBD_NORM,
+                LLM_TENSOR_OUTPUT_NORM,
+                LLM_TENSOR_ATTN_NORM,
+                LLM_TENSOR_ATTN_OUT,
+                LLM_TENSOR_ATTN_QKV,
+                LLM_TENSOR_ATTN_ROT_EMBD,
+                LLM_TENSOR_FFN_DOWN,
+                LLM_TENSOR_FFN_UP,
+                LLM_TENSOR_FFN_NORM,
+                LLM_TENSOR_CLS,
+                LLM_TENSOR_CLS_OUT,
+            };
         case LLM_ARCH_JINA_BERT_V2:
             return {
                 LLM_TENSOR_TOKEN_EMBD,

src/llama-arch.h

@@ -24,6 +24,7 @@ enum llm_arch {
     LLM_ARCH_STARCODER,
     LLM_ARCH_REFACT,
     LLM_ARCH_BERT,
+    LLM_ARCH_MODERN_BERT,
     LLM_ARCH_NOMIC_BERT,
     LLM_ARCH_NOMIC_BERT_MOE,
     LLM_ARCH_NEO_BERT,
@@ -188,6 +189,7 @@ enum llm_kv {
     LLM_KV_EMBEDDING_SCALE,
     LLM_KV_TOKEN_SHIFT_COUNT,
     LLM_KV_INTERLEAVE_MOE_LAYER_STEP,
+    LLM_KV_DENSE_EVERY_N_LAYERS,
 
     LLM_KV_ATTENTION_HEAD_COUNT,
     LLM_KV_ATTENTION_HEAD_COUNT_KV,
@@ -208,6 +210,7 @@ enum llm_kv {
     LLM_KV_ATTENTION_GATE_LORA_RANK,
     LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT,
     LLM_KV_ATTENTION_SLIDING_WINDOW,
+    LLM_KV_ATTENTION_DENSE_EVERY_N_LAYERS,
     LLM_KV_ATTENTION_SCALE,
     LLM_KV_ATTENTION_OUTPUT_SCALE,
     LLM_KV_ATTENTION_TEMPERATURE_LENGTH,
@@ -218,6 +221,7 @@ enum llm_kv {
     LLM_KV_ROPE_DIMENSION_COUNT,
     LLM_KV_ROPE_DIMENSION_SECTIONS,
     LLM_KV_ROPE_FREQ_BASE,
+    LLM_KV_ROPE_FREQ_BASE_SWA,
     LLM_KV_ROPE_SCALE_LINEAR,
     LLM_KV_ROPE_SCALING_TYPE,
     LLM_KV_ROPE_SCALING_FACTOR,

src/llama-hparams.h

@@ -123,6 +123,7 @@ struct llama_hparams {
     llama_swa_type swa_type = LLAMA_SWA_TYPE_NONE;
     // the size of the sliding window (0 - no SWA)
     uint32_t n_swa = 0;
+    uint32_t swa_period = 1;
     // if swa_layers[il] == true, then layer il is SWA
     // if swa_layers[il] == false, then layer il is dense (i.e. non-SWA)
     // by default, all layers are dense

src/llama-model-saver.cpp

@@ -182,6 +182,7 @@ void llama_model_saver::add_kv_from_model() {
     add_kv(LLM_KV_ATTENTION_KV_LORA_RANK,            hparams.n_lora_kv);
     add_kv(LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT,  hparams.n_rel_attn_bkts);
     add_kv(LLM_KV_ATTENTION_SLIDING_WINDOW,          hparams.n_swa);
+    add_kv(LLM_KV_ATTENTION_DENSE_EVERY_N_LAYERS,    hparams.swa_period);
     add_kv(LLM_KV_ATTENTION_SCALE,                   hparams.f_attention_scale);
 
     const float rope_scaling_factor = hparams.rope_freq_scale_train == 1.0f ? 0.0f : 1.0f/hparams.rope_freq_scale_train;