llama : add support for NVIDIA Nemotron 3 Nano (ggml-org#18058)

* llama : add support for NVIDIA Nemotron Nano 3

This commit adds support for the NVIDIA Nemotron Nano 3 model, enabling
the conversion and running of this model.

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

Author: danbev

convert_hf_to_gguf.py

@@ -8490,8 +8490,18 @@ def set_vocab(self):
 class NemotronHModel(GraniteHybridModel):
     """Hybrid mamba2/attention model from NVIDIA"""
     model_arch = gguf.MODEL_ARCH.NEMOTRON_H
+    is_moe: bool = False
 
     def __init__(self, *args, **kwargs):
+        # We have to determine the correct model architecture (MoE vs non-MoE) before
+        # calling the parent __init__. This is because the parent constructor
+        # uses self.model_arch to build the tensor name map, and all MoE-specific
+        # mappings would be missed if it were called with the default non-MoE arch.
+        hparams = ModelBase.load_hparams(args[0], self.is_mistral_format)
+        if "num_experts_per_tok" in hparams:
+            self.model_arch = gguf.MODEL_ARCH.NEMOTRON_H_MOE
+            self.is_moe = True
+
         super().__init__(*args, **kwargs)
 
         # Save the top-level head_dim for later
@@ -8503,9 +8513,11 @@ def __init__(self, *args, **kwargs):
 
         # Update the ssm / attn / mlp layers
         # M: Mamba2, *: Attention, -: MLP
+        # MoE:
+        # M: Mamba2, *: Attention, E: Expert
         hybrid_override_pattern = self.hparams["hybrid_override_pattern"]
         self._ssm_layers = [i for i, val in enumerate(hybrid_override_pattern) if val == "M"]
-        self._mlp_layers = [i for i, val in enumerate(hybrid_override_pattern) if val == "-"]
+        self._mlp_layers = [i for i, val in enumerate(hybrid_override_pattern) if val == ("E" if self.is_moe else "-")]
 
     def get_attn_layers(self):
         hybrid_override_pattern = self.hparams["hybrid_override_pattern"]
@@ -8521,18 +8533,110 @@ def set_gguf_parameters(self):
         # Set feed_forward_length
         # NOTE: This will trigger an override warning. This is preferrable to
         #   duplicating all the parent logic
-        n_ff = self.find_hparam(["intermediate_size", "n_inner", "hidden_dim"])
-        self.gguf_writer.add_feed_forward_length([
-            n_ff if i in self._mlp_layers else 0 for i in range(self.block_count)
-        ])
+        if not self.is_moe:
+            n_ff = self.find_hparam(["intermediate_size", "n_inner", "hidden_dim"])
+            self.gguf_writer.add_feed_forward_length([
+                n_ff if i in self._mlp_layers else 0 for i in range(self.block_count)
+            ])
+        else:
+            moe_intermediate_size = self.hparams["moe_intermediate_size"]
+            self.gguf_writer.add_feed_forward_length([
+                moe_intermediate_size if i in self._mlp_layers else 0 for i in range(self.block_count)
+            ])
+            self.gguf_writer.add_expert_used_count(self.hparams["num_experts_per_tok"])
+            self.gguf_writer.add_expert_feed_forward_length(self.hparams["moe_intermediate_size"])
+            self.gguf_writer.add_expert_shared_feed_forward_length(self.hparams["moe_shared_expert_intermediate_size"])
+            self.gguf_writer.add_expert_count(self.hparams["n_routed_experts"])
+            self.gguf_writer.add_expert_shared_count(self.hparams["n_shared_experts"])
+            self.gguf_writer.add_expert_weights_norm(self.hparams["norm_topk_prob"])
+            self.gguf_writer.add_expert_weights_scale(self.hparams["routed_scaling_factor"])
+            self.gguf_writer.add_expert_group_count(self.hparams["n_group"])
+
+            # number of experts used per token (top-k)
+            if (n_experts_used := self.hparams.get("num_experts_per_tok")) is not None:
+                self.gguf_writer.add_expert_used_count(n_experts_used)
 
     def set_vocab(self):
         super().set_vocab()
 
         # The tokenizer _does_ add a BOS token (via post_processor type
         # TemplateProcessing) but does not set add_bos_token to true in the
         # config, so we need to explicitly override it here.
-        self.gguf_writer.add_add_bos_token(True)
+        if not self.is_moe:
+            self.gguf_writer.add_add_bos_token(True)
+
+    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
+        if self.is_moe and bid is not None:
+            if name.endswith("mixer.gate.e_score_correction_bias"):
+                new_name = name.replace("e_score_correction_bias", "e_score_correction.bias")
+                mapped_name = self.map_tensor_name(new_name)
+                return [(mapped_name, data_torch)]
+
+            if name.endswith("mixer.dt_bias"):
+                new_name = name.replace("dt_bias", "dt.bias")
+                mapped_name = self.map_tensor_name(new_name)
+                return [(mapped_name, data_torch)]
+
+            if name.endswith("mixer.conv1d.weight"):
+                squeezed_data = data_torch.squeeze()
+                mapped_name = self.map_tensor_name(name)
+                return [(mapped_name, squeezed_data)]
+
+            if name.endswith("mixer.A_log"):
+                transformed_data = -torch.exp(data_torch)
+                reshaped_data = transformed_data.squeeze().reshape(-1, 1)
+                mapped_name = self.map_tensor_name(name)
+                return [(mapped_name, reshaped_data)]
+
+            if name.endswith("mixer.D"):
+                reshaped_data = data_torch.squeeze().reshape(-1, 1)
+                mapped_name = self.map_tensor_name(name)
+                return [(mapped_name, reshaped_data)]
+
+            if name.endswith("mixer.norm.weight"):
+                reshaped_data = data_torch.reshape(8, 512)
+                mapped_name = self.map_tensor_name(name)
+                return [(mapped_name, reshaped_data)]
+
+            if name.find("mixer.experts") != -1:
+                n_experts = self.hparams["n_routed_experts"]
+                assert bid is not None
+
+                if self._experts is None:
+                    self._experts = [{} for _ in range(self.block_count)]
+
+                self._experts[bid][name] = data_torch
+
+                if len(self._experts[bid]) >= n_experts * 2:
+                    # merge the experts into a single tensor
+                    tensors: list[tuple[str, Tensor]] = []
+                    for w_name in ["down_proj", "up_proj"]:
+                        datas: list[Tensor] = []
+
+                        for xid in range(n_experts):
+                            ename = f"backbone.layers.{bid}.mixer.experts.{xid}.{w_name}.weight"
+                            datas.append(self._experts[bid][ename])
+                            del self._experts[bid][ename]
+
+                        data_torch = torch.stack(datas, dim=0)
+                        merged_name = f"model.layers.{bid}.mlp.experts.{w_name}.weight"
+                        new_name = self.map_tensor_name(merged_name)
+                        tensors.append((new_name, data_torch))
+
+                    return tensors
+                else:
+                    return []
+
+        return super().modify_tensors(data_torch, name, bid)
+
+    def prepare_tensors(self):
+        super().prepare_tensors()
+
+        if self._experts is not None:
+            # flatten `list[dict[str, Tensor]]` into `list[str]`
+            experts = [k for d in self._experts for k in d.keys()]
+            if len(experts) > 0:
+                raise ValueError(f"Unprocessed experts: {experts}")
 
 
 @ModelBase.register("BailingMoeForCausalLM")

gguf-py/gguf/constants.py

@@ -413,6 +413,7 @@ class MODEL_ARCH(IntEnum):
     JAIS             = auto()
     NEMOTRON         = auto()
     NEMOTRON_H       = auto()
+    NEMOTRON_H_MOE   = auto()
     EXAONE           = auto()
     EXAONE4          = auto()
     GRANITE          = auto()
@@ -786,6 +787,7 @@ class MODEL_TENSOR(IntEnum):
     MODEL_ARCH.JAIS:             "jais",
     MODEL_ARCH.NEMOTRON:         "nemotron",
     MODEL_ARCH.NEMOTRON_H:       "nemotron_h",
+    MODEL_ARCH.NEMOTRON_H_MOE:   "nemotron_h_moe",
     MODEL_ARCH.EXAONE:           "exaone",
     MODEL_ARCH.EXAONE4:          "exaone4",
     MODEL_ARCH.GRANITE:          "granite",
@@ -2529,6 +2531,33 @@ class MODEL_TENSOR(IntEnum):
         MODEL_TENSOR.FFN_DOWN,
         MODEL_TENSOR.FFN_UP,
     ],
+    MODEL_ARCH.NEMOTRON_H_MOE: [
+        MODEL_TENSOR.TOKEN_EMBD,
+        MODEL_TENSOR.OUTPUT_NORM,
+        MODEL_TENSOR.OUTPUT,
+        MODEL_TENSOR.ATTN_NORM,
+        MODEL_TENSOR.SSM_IN,
+        MODEL_TENSOR.SSM_CONV1D,
+        MODEL_TENSOR.SSM_DT,
+        MODEL_TENSOR.SSM_A,
+        MODEL_TENSOR.SSM_D,
+        MODEL_TENSOR.SSM_NORM,
+        MODEL_TENSOR.SSM_OUT,
+        MODEL_TENSOR.ATTN_Q,
+        MODEL_TENSOR.ATTN_K,
+        MODEL_TENSOR.ATTN_V,
+        MODEL_TENSOR.ATTN_OUT,
+        MODEL_TENSOR.FFN_DOWN,
+        MODEL_TENSOR.FFN_UP,
+        # experts
+        MODEL_TENSOR.FFN_GATE_INP,
+        MODEL_TENSOR.FFN_UP_EXP,
+        MODEL_TENSOR.FFN_DOWN_EXP,
+        # shared expert
+        MODEL_TENSOR.FFN_DOWN_SHEXP,
+        MODEL_TENSOR.FFN_UP_SHEXP,
+        MODEL_TENSOR.FFN_EXP_PROBS_B,
+    ],
     MODEL_ARCH.EXAONE: [
         MODEL_TENSOR.TOKEN_EMBD,
         MODEL_TENSOR.OUTPUT_NORM,

gguf-py/gguf/tensor_mapping.py

@@ -379,6 +379,7 @@ class TensorNameMap:
             "model.layers.{bid}.feed_forward.gate",               # lfm2moe
             "model.layers.{bid}.mlp.router.gate",               # afmoe
             "layers.{bid}.gate",                                # mistral-large
+            "backbone.layers.{bid}.mixer.gate",                 # nemotron-h-moe
         ),
 
         MODEL_TENSOR.FFN_GATE_INP_SHEXP: (
@@ -392,6 +393,7 @@ class TensorNameMap:
             "model.layers.{bid}.mlp.expert_bias",                           # afmoe
             "model.layers.{bid}.feed_forward.expert_bias",                  # lfm2moe
             "model.layers.{bid}.block_sparse_moe.e_score_correction",       # minimax-m2
+            "backbone.layers.{bid}.mixer.gate.e_score_correction"           # nemotron-h-moe
         ),
 
         # Feed-forward up
@@ -440,7 +442,7 @@ class TensorNameMap:
             "layers.{bid}.feed_forward.experts.w3",                 # mixtral (merged)
             "transformer.decoder_layer.{bid}.moe.linear_v",         # Grok (merged)
             "transformer.blocks.{bid}.ffn.experts.mlp.v1",          # dbrx
-            "model.layers.{bid}.mlp.experts.up_proj",               # qwen2moe olmoe (merged) ernie4.5-moe
+            "model.layers.{bid}.mlp.experts.up_proj",               # qwen2moe olmoe (merged) ernie4.5-moe, nemotron-h-moe (merged)
             "model.layers.{bid}.block_sparse_moe.experts.w3",       # phimoe (merged)
             "model.layers.{bid}.feed_forward.experts.up_proj",      # llama4
             "encoder.layers.{bid}.mlp.experts.mlp.w1",              # nomic-bert-moe
@@ -454,6 +456,7 @@ class TensorNameMap:
             "model.layers.{bid}.feed_forward.down_proj",
             "model.layers.{bid}.mlp.shared_mlp.up_proj",             # hunyuan
             "layers.{bid}.shared_experts.w3",                        # mistral-large
+            "backbone.layers.{bid}.mixer.shared_experts.up_proj",    # nemotron-h-moe
         ),
 
         MODEL_TENSOR.FFN_UP_CHEXP: (
@@ -548,7 +551,7 @@ class TensorNameMap:
             "layers.{bid}.feed_forward.experts.w2",                 # mixtral (merged)
             "transformer.decoder_layer.{bid}.moe.linear_1",         # Grok (merged)
             "transformer.blocks.{bid}.ffn.experts.mlp.w2",          # dbrx
-            "model.layers.{bid}.mlp.experts.down_proj",             # qwen2moe olmoe (merged) ernie4.5-moe
+            "model.layers.{bid}.mlp.experts.down_proj",             # qwen2moe olmoe (merged) ernie4.5-moe nemotron-h-moe (merged)
             "model.layers.{bid}.block_sparse_moe.output_linear",    # granitemoe
             "model.layers.{bid}.block_sparse_moe.experts.w2",       # phimoe (merged)
             "model.layers.{bid}.feed_forward.experts.down_proj",    # llama4
@@ -563,6 +566,7 @@ class TensorNameMap:
             "model.layers.{bid}.shared_mlp.output_linear",             # granitemoe
             "model.layers.{bid}.mlp.shared_mlp.down_proj",             # hunyuan
             "layers.{bid}.shared_experts.w2",                          # mistral-large
+            "backbone.layers.{bid}.mixer.shared_experts.down_proj",    # nemotron-h-moe
         ),
 
         MODEL_TENSOR.FFN_DOWN_CHEXP: (
@@ -706,6 +710,7 @@ class TensorNameMap:
             "model.layers.{bid}.mamba.dt_proj",         # jamba falcon-h1 granite-hybrid
             "model.layers.layers.{bid}.mixer.dt_proj",  # plamo2
             "model.layers.{bid}.linear_attn.dt_proj",   # qwen3next
+            "backbone.layers.{bid}.mixer.dt",           # nemotron-h-moe
         ),
 
         MODEL_TENSOR.SSM_DT_NORM: (

src/llama-arch.cpp

@@ -75,6 +75,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_JAIS,             "jais"             },
     { LLM_ARCH_NEMOTRON,         "nemotron"         },
     { LLM_ARCH_NEMOTRON_H,       "nemotron_h"       },
+    { LLM_ARCH_NEMOTRON_H_MOE,   "nemotron_h_moe"   },
     { LLM_ARCH_EXAONE,           "exaone"           },
     { LLM_ARCH_EXAONE4,          "exaone4"          },
     { LLM_ARCH_RWKV6,            "rwkv6"            },
@@ -1763,6 +1764,39 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
             { LLM_TENSOR_FFN_UP,         "blk.%d.ffn_up" },
         },
     },
+    {
+        LLM_ARCH_NEMOTRON_H_MOE,
+        {
+            { LLM_TENSOR_TOKEN_EMBD,     "token_embd" },
+            { LLM_TENSOR_OUTPUT_NORM,    "output_norm" },
+            { LLM_TENSOR_OUTPUT,         "output" },
+            { LLM_TENSOR_ATTN_NORM,      "blk.%d.attn_norm" },
+            // mamba(2) ssm layers
+            { LLM_TENSOR_SSM_IN,         "blk.%d.ssm_in" },
+            { LLM_TENSOR_SSM_CONV1D,     "blk.%d.ssm_conv1d" },
+            { LLM_TENSOR_SSM_DT,         "blk.%d.ssm_dt" },
+            { LLM_TENSOR_SSM_A,          "blk.%d.ssm_a" },
+            { LLM_TENSOR_SSM_D,          "blk.%d.ssm_d" },
+            { LLM_TENSOR_SSM_NORM,       "blk.%d.ssm_norm" },
+            { LLM_TENSOR_SSM_OUT,        "blk.%d.ssm_out" },
+            // attention layers
+            { LLM_TENSOR_ATTN_Q,         "blk.%d.attn_q" },
+            { LLM_TENSOR_ATTN_K,         "blk.%d.attn_k" },
+            { LLM_TENSOR_ATTN_V,         "blk.%d.attn_v" },
+            { LLM_TENSOR_ATTN_OUT,       "blk.%d.attn_output" },
+            // dense FFN
+            { LLM_TENSOR_FFN_DOWN,       "blk.%d.ffn_down" },
+            { LLM_TENSOR_FFN_UP,         "blk.%d.ffn_up" },
+            // MoE FFN (for MoE layers)
+            { LLM_TENSOR_FFN_GATE_INP,   "blk.%d.ffn_gate_inp" },
+            { LLM_TENSOR_FFN_UP_EXPS,    "blk.%d.ffn_up_exps" },
+            { LLM_TENSOR_FFN_DOWN_EXPS,  "blk.%d.ffn_down_exps" },
+            { LLM_TENSOR_FFN_EXP_PROBS_B,"blk.%d.exp_probs_b" },
+            // MoE shared expert layer
+            { LLM_TENSOR_FFN_DOWN_SHEXP,  "blk.%d.ffn_down_shexp" },
+            { LLM_TENSOR_FFN_UP_SHEXP,    "blk.%d.ffn_up_shexp" },
+        },
+    },
     {
         LLM_ARCH_EXAONE,
         {
@@ -2817,6 +2851,7 @@ bool llm_arch_is_hybrid(const llm_arch & arch) {
         case LLM_ARCH_LFM2:
         case LLM_ARCH_LFM2MOE:
         case LLM_ARCH_NEMOTRON_H:
+        case LLM_ARCH_NEMOTRON_H_MOE:
         case LLM_ARCH_QWEN3NEXT:
             return true;
         default:

src/llama-arch.h

@@ -79,6 +79,7 @@ enum llm_arch {
     LLM_ARCH_JAIS,
     LLM_ARCH_NEMOTRON,
     LLM_ARCH_NEMOTRON_H,
+    LLM_ARCH_NEMOTRON_H_MOE,
     LLM_ARCH_EXAONE,
     LLM_ARCH_EXAONE4,
     LLM_ARCH_RWKV6,

src/llama-graph.cpp

@@ -1089,6 +1089,15 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
                 cur = ggml_relu(ctx0, cur);
                 cb(cur, "ffn_moe_relu", il);
             } break;
+        case LLM_FFN_RELU_SQR:
+            if (gate_exps) {
+                // TODO: add support for gated squared relu
+                GGML_ABORT("fatal error: gated squared relu not implemented");
+            } else {
+                cur = ggml_relu(ctx0, cur);
+                cur = ggml_sqr(ctx0, cur);
+                cb(cur, "ffn_moe_relu_sqr", il);
+            } break;
         default:
             GGML_ABORT("fatal error");
     }