Rename nemotronh to nemotron_h for consistency

- Update architecture name from NEMOTRONH to NEMOTRON_H in constants.py
- Change architecture string from 'nemotronh' to 'nemotron_h' in all files
- Update enum LLM_ARCH_NEMOTRONH to LLM_ARCH_NEMOTRON_H
- Update class name llm_build_nemotronh to llm_build_nemotron_h
- Consistent naming with underscore convention (nemotron_h vs nemotronh)

Author: jwjohns

convert_hf_to_gguf.py

@@ -7648,7 +7648,7 @@ def set_vocab(self):
 @ModelBase.register("NemotronHForCausalLM")
 class NemotronHModel(GraniteHybridModel):
     """Hybrid mamba2/attention model from NVIDIA"""
-    model_arch = gguf.MODEL_ARCH.NEMOTRONH
+    model_arch = gguf.MODEL_ARCH.NEMOTRON_H
 
     def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)

gguf-py/gguf/constants.py

@@ -367,7 +367,7 @@ class MODEL_ARCH(IntEnum):
     T5ENCODER        = auto()
     JAIS             = auto()
     NEMOTRON         = auto()
-    NEMOTRONH        = auto()
+    NEMOTRON_H       = auto()
     EXAONE           = auto()
     EXAONE4          = auto()
     GRANITE          = auto()
@@ -701,7 +701,7 @@ class MODEL_TENSOR(IntEnum):
     MODEL_ARCH.T5ENCODER:        "t5encoder",
     MODEL_ARCH.JAIS:             "jais",
     MODEL_ARCH.NEMOTRON:         "nemotron",
-    MODEL_ARCH.NEMOTRONH:        "nemotronh",
+    MODEL_ARCH.NEMOTRON_H:       "nemotron_h",
     MODEL_ARCH.EXAONE:           "exaone",
     MODEL_ARCH.EXAONE4:          "exaone4",
     MODEL_ARCH.GRANITE:          "granite",
@@ -2299,7 +2299,7 @@ class MODEL_TENSOR(IntEnum):
         MODEL_TENSOR.FFN_DOWN,
         MODEL_TENSOR.FFN_UP,
     ],
-    MODEL_ARCH.NEMOTRONH: [
+    MODEL_ARCH.NEMOTRON_H: [
         MODEL_TENSOR.TOKEN_EMBD,
         MODEL_TENSOR.OUTPUT_NORM,
         MODEL_TENSOR.OUTPUT,

src/llama-arch.cpp

@@ -69,7 +69,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_T5ENCODER,        "t5encoder"        },
     { LLM_ARCH_JAIS,             "jais"             },
     { LLM_ARCH_NEMOTRON,         "nemotron"         },
-    { LLM_ARCH_NEMOTRONH,        "nemotronh"        },
+    { LLM_ARCH_NEMOTRON_H,       "nemotron_h"       },
     { LLM_ARCH_EXAONE,           "exaone"           },
     { LLM_ARCH_EXAONE4,          "exaone4"          },
     { LLM_ARCH_RWKV6,            "rwkv6"            },
@@ -1552,7 +1552,7 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
         },
     },
     {
-        LLM_ARCH_NEMOTRONH,
+        LLM_ARCH_NEMOTRON_H,
         {
             { LLM_TENSOR_TOKEN_EMBD,     "token_embd" },
             { LLM_TENSOR_OUTPUT_NORM,    "output_norm" },
@@ -2381,7 +2381,7 @@ bool llm_arch_is_hybrid(const llm_arch & arch) {
         case LLM_ARCH_PLAMO2:
         case LLM_ARCH_GRANITE_HYBRID:
         case LLM_ARCH_LFM2:
-        case LLM_ARCH_NEMOTRONH:
+        case LLM_ARCH_NEMOTRON_H:
             return true;
         default:
             return false;

src/llama-arch.h

@@ -73,7 +73,7 @@ enum llm_arch {
     LLM_ARCH_T5ENCODER,
     LLM_ARCH_JAIS,
     LLM_ARCH_NEMOTRON,
-    LLM_ARCH_NEMOTRONH,
+    LLM_ARCH_NEMOTRON_H,
     LLM_ARCH_EXAONE,
     LLM_ARCH_EXAONE4,
     LLM_ARCH_RWKV6,

src/llama-model.cpp

@@ -1570,7 +1570,7 @@ void llama_model::load_hparams(llama_model_loader & ml) {
                     default: type = LLM_TYPE_UNKNOWN;
                 }
             } break;
-        case LLM_ARCH_NEMOTRONH:
+        case LLM_ARCH_NEMOTRON_H:
             {
                 ml.get_key(LLM_KV_SSM_CONV_KERNEL,    hparams.ssm_d_conv);
                 ml.get_key(LLM_KV_SSM_INNER_SIZE,     hparams.ssm_d_inner);
@@ -4709,7 +4709,7 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
                         layer.ffn_up_b   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);
                     }
                 } break;
-            case LLM_ARCH_NEMOTRONH:
+            case LLM_ARCH_NEMOTRON_H:
                 {
                     // mamba2 Mixer SSM params
                     // NOTE: int64_t for tensor dimensions
@@ -5953,7 +5953,7 @@ void llama_model::print_info() const {
         arch == LLM_ARCH_FALCON_H1 ||
         arch == LLM_ARCH_PLAMO2 ||
         arch == LLM_ARCH_GRANITE_HYBRID ||
-        arch == LLM_ARCH_NEMOTRONH) {
+        arch == LLM_ARCH_NEMOTRON_H) {
         LLAMA_LOG_INFO("%s: ssm_d_conv       = %u\n",     __func__, hparams.ssm_d_conv);
         LLAMA_LOG_INFO("%s: ssm_d_inner      = %u\n",     __func__, hparams.ssm_d_inner);
         LLAMA_LOG_INFO("%s: ssm_d_state      = %u\n",     __func__, hparams.ssm_d_state);
@@ -14220,8 +14220,8 @@ struct llm_build_nemotron : public llm_graph_context {
     }
 };
 
-struct llm_build_nemotronh : public llm_graph_context_mamba {
-    llm_build_nemotronh(
+struct llm_build_nemotron_h : public llm_graph_context_mamba {
+    llm_build_nemotron_h(
             const llama_model      & model,
             const llm_graph_params & params) :
         llm_graph_context_mamba(params) {
@@ -18508,7 +18508,7 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
                     if (arch == LLM_ARCH_FALCON_H1) {
                         filter_attn = [&](int32_t) { return true; };
                         filter_recr = [&](int32_t) { return true; };
-                    } else if (arch == LLM_ARCH_NEMOTRONH) {
+                    } else if (arch == LLM_ARCH_NEMOTRON_H) {
                         filter_attn = [&](int32_t il) {
                             return !hparams.is_recurrent(il) && hparams.n_ff(il) == 0;
                         };
@@ -18865,9 +18865,9 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
             {
                 llm = std::make_unique<llm_build_nemotron>(*this, params);
             } break;
-        case LLM_ARCH_NEMOTRONH:
+        case LLM_ARCH_NEMOTRON_H:
             {
-                llm = std::make_unique<llm_build_nemotronh>(*this, params);
+                llm = std::make_unique<llm_build_nemotron_h>(*this, params);
             } break;
         case LLM_ARCH_EXAONE:
             {
@@ -19104,7 +19104,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
         case LLM_ARCH_RWKV7:
         case LLM_ARCH_ARWKV7:
         case LLM_ARCH_WAVTOKENIZER_DEC:
-        case LLM_ARCH_NEMOTRONH:
+        case LLM_ARCH_NEMOTRON_H:
             return LLAMA_ROPE_TYPE_NONE;
 
         // use what we call a normal RoPE, operating on pairs of consecutive head values