当前位置： 首页 > article >正文

llama.cpp LLM_ARCH_DEEPSEEK and LLM_ARCH_DEEPSEEK2

article 2026/5/11 17:23:38

llama.cpp LLM_ARCH_DEEPSEEK and LLM_ARCH_DEEPSEEK2

  • 1. `LLM_ARCH_DEEPSEEK` and `LLM_ARCH_DEEPSEEK2`
  • 2. `LLM_ARCH_DEEPSEEK` and `LLM_ARCH_DEEPSEEK2`
  • 3. `struct ggml_cgraph * build_deepseek()` and `struct ggml_cgraph * build_deepseek2()`
  • References

不宜吹捧中国大语言模型的同时，又去贬低美国大语言模型。

水是人体的主要化学成分，约占体重的 50% 至 70%。大语言模型的含水量也不会太少。

llama.cpp
https://github.com/ggerganov/llama.cpp

1. LLM_ARCH_DEEPSEEK and LLM_ARCH_DEEPSEEK2

/home/yongqiang/llm_work/llama_cpp_25_01_05/llama.cpp/src/llama-arch.h
/home/yongqiang/llm_work/llama_cpp_25_01_05/llama.cpp/src/llama-arch.cpp

  • LLM_ARCH_DEEPSEEK and LLM_ARCH_DEEPSEEK2
//
// gguf constants (sync with gguf.py)
//enum llm_arch {LLM_ARCH_LLAMA,LLM_ARCH_DECI,LLM_ARCH_FALCON,LLM_ARCH_BAICHUAN,LLM_ARCH_GROK,LLM_ARCH_GPT2,LLM_ARCH_GPTJ,LLM_ARCH_GPTNEOX,LLM_ARCH_MPT,LLM_ARCH_STARCODER,LLM_ARCH_REFACT,LLM_ARCH_BERT,LLM_ARCH_NOMIC_BERT,LLM_ARCH_JINA_BERT_V2,LLM_ARCH_BLOOM,LLM_ARCH_STABLELM,LLM_ARCH_QWEN,LLM_ARCH_QWEN2,LLM_ARCH_QWEN2MOE,LLM_ARCH_QWEN2VL,LLM_ARCH_PHI2,LLM_ARCH_PHI3,LLM_ARCH_PHIMOE,LLM_ARCH_PLAMO,LLM_ARCH_CODESHELL,LLM_ARCH_ORION,LLM_ARCH_INTERNLM2,LLM_ARCH_MINICPM,LLM_ARCH_MINICPM3,LLM_ARCH_GEMMA,LLM_ARCH_GEMMA2,LLM_ARCH_STARCODER2,LLM_ARCH_MAMBA,LLM_ARCH_XVERSE,LLM_ARCH_COMMAND_R,LLM_ARCH_COHERE2,LLM_ARCH_DBRX,LLM_ARCH_OLMO,LLM_ARCH_OLMO2,LLM_ARCH_OLMOE,LLM_ARCH_OPENELM,LLM_ARCH_ARCTIC,LLM_ARCH_DEEPSEEK,LLM_ARCH_DEEPSEEK2,LLM_ARCH_CHATGLM,LLM_ARCH_BITNET,LLM_ARCH_T5,LLM_ARCH_T5ENCODER,LLM_ARCH_JAIS,LLM_ARCH_NEMOTRON,LLM_ARCH_EXAONE,LLM_ARCH_RWKV6,LLM_ARCH_RWKV6QWEN2,LLM_ARCH_GRANITE,LLM_ARCH_GRANITE_MOE,LLM_ARCH_CHAMELEON,LLM_ARCH_WAVTOKENIZER_DEC,LLM_ARCH_UNKNOWN,
};
  • { LLM_ARCH_DEEPSEEK, "deepseek" } and { LLM_ARCH_DEEPSEEK2, "deepseek2" }
static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {{ LLM_ARCH_LLAMA,            "llama"            },{ LLM_ARCH_DECI,             "deci"             },{ LLM_ARCH_FALCON,           "falcon"           },{ LLM_ARCH_GROK,             "grok"             },{ LLM_ARCH_GPT2,             "gpt2"             },{ LLM_ARCH_GPTJ,             "gptj"             },{ LLM_ARCH_GPTNEOX,          "gptneox"          },{ LLM_ARCH_MPT,              "mpt"              },{ LLM_ARCH_BAICHUAN,         "baichuan"         },{ LLM_ARCH_STARCODER,        "starcoder"        },{ LLM_ARCH_REFACT,           "refact"           },{ LLM_ARCH_BERT,             "bert"             },{ LLM_ARCH_NOMIC_BERT,       "nomic-bert"       },{ LLM_ARCH_JINA_BERT_V2,     "jina-bert-v2"     },{ LLM_ARCH_BLOOM,            "bloom"            },{ LLM_ARCH_STABLELM,         "stablelm"         },{ LLM_ARCH_QWEN,             "qwen"             },{ LLM_ARCH_QWEN2,            "qwen2"            },{ LLM_ARCH_QWEN2MOE,         "qwen2moe"         },{ LLM_ARCH_QWEN2VL,          "qwen2vl"          },{ LLM_ARCH_PHI2,             "phi2"             },{ LLM_ARCH_PHI3,             "phi3"             },{ LLM_ARCH_PHIMOE,           "phimoe"           },{ LLM_ARCH_PLAMO,            "plamo"            },{ LLM_ARCH_CODESHELL,        "codeshell"        },{ LLM_ARCH_ORION,            "orion"            },{ LLM_ARCH_INTERNLM2,        "internlm2"        },{ LLM_ARCH_MINICPM,          "minicpm"          },{ LLM_ARCH_MINICPM3,         "minicpm3"         },{ LLM_ARCH_GEMMA,            "gemma"            },{ LLM_ARCH_GEMMA2,           "gemma2"           },{ LLM_ARCH_STARCODER2,       "starcoder2"       },{ LLM_ARCH_MAMBA,            "mamba"            },{ LLM_ARCH_XVERSE,           "xverse"           },{ LLM_ARCH_COMMAND_R,        "command-r"        },{ LLM_ARCH_COHERE2,          "cohere2"          },{ LLM_ARCH_DBRX,             "dbrx"             },{ LLM_ARCH_OLMO,             "olmo"             },{ LLM_ARCH_OLMO2,            "olmo2"            },{ LLM_ARCH_OLMOE,            "olmoe"            },{ LLM_ARCH_OPENELM,          "openelm"          },{ LLM_ARCH_ARCTIC,           "arctic"           },{ LLM_ARCH_DEEPSEEK,         "deepseek"         },{ LLM_ARCH_DEEPSEEK2,        "deepseek2"        },{ LLM_ARCH_CHATGLM,          "chatglm"          },{ LLM_ARCH_BITNET,           "bitnet"           },{ LLM_ARCH_T5,               "t5"               },{ LLM_ARCH_T5ENCODER,        "t5encoder"        },{ LLM_ARCH_JAIS,             "jais"             },{ LLM_ARCH_NEMOTRON,         "nemotron"         },{ LLM_ARCH_EXAONE,           "exaone"           },{ LLM_ARCH_RWKV6,            "rwkv6"            },{ LLM_ARCH_RWKV6QWEN2,       "rwkv6qwen2"       },{ LLM_ARCH_GRANITE,          "granite"          },{ LLM_ARCH_GRANITE_MOE,      "granitemoe"       },{ LLM_ARCH_CHAMELEON,        "chameleon"        },{ LLM_ARCH_WAVTOKENIZER_DEC, "wavtokenizer-dec" },{ LLM_ARCH_UNKNOWN,          "(unknown)"        },
};

2. LLM_ARCH_DEEPSEEK and LLM_ARCH_DEEPSEEK2

/home/yongqiang/llm_work/llama_cpp_25_01_05/llama.cpp/src/llama-arch.cpp

  • LLM_ARCH_DEEPSEEK and LLM_ARCH_DEEPSEEK2
static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_NAMES = {{LLM_ARCH_LLAMA,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },{ LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_GATE_EXP,    "blk.%d.ffn_gate.%d" },{ LLM_TENSOR_FFN_DOWN_EXP,    "blk.%d.ffn_down.%d" },{ LLM_TENSOR_FFN_UP_EXP,      "blk.%d.ffn_up.%d" },{ LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },{ LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },{ LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },},},{LLM_ARCH_DECI,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },{ LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_GATE_EXP,    "blk.%d.ffn_gate.%d" },{ LLM_TENSOR_FFN_DOWN_EXP,    "blk.%d.ffn_down.%d" },{ LLM_TENSOR_FFN_UP_EXP,      "blk.%d.ffn_up.%d" },{ LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },{ LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },{ LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },},},{LLM_ARCH_BAICHUAN,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_FALCON,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_NORM_2,     "blk.%d.attn_norm_2" },{ LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_GROK,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },{ LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE_EXP,    "blk.%d.ffn_gate.%d" },{ LLM_TENSOR_FFN_DOWN_EXP,    "blk.%d.ffn_down.%d" },{ LLM_TENSOR_FFN_UP_EXP,      "blk.%d.ffn_up.%d" },{ LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },{ LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },{ LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },{ LLM_TENSOR_LAYER_OUT_NORM,  "blk.%d.layer_output_norm" },{ LLM_TENSOR_ATTN_OUT_NORM,   "blk.%d.attn_output_norm" },},},{LLM_ARCH_GPT2,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_POS_EMBD,        "position_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },},},{LLM_ARCH_GPTJ,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },},},{LLM_ARCH_GPTNEOX,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_MPT,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output"},{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_ACT,         "blk.%d.ffn.act" },{ LLM_TENSOR_POS_EMBD,        "position_embd" },{ LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm"},{ LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm"},},},{LLM_ARCH_STARCODER,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_POS_EMBD,        "position_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },},},{LLM_ARCH_REFACT,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_BERT,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" },{ LLM_TENSOR_TOKEN_TYPES,     "token_types" },{ LLM_TENSOR_POS_EMBD,        "position_embd" },{ LLM_TENSOR_ATTN_OUT_NORM,   "blk.%d.attn_output_norm" },{ 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" },{ LLM_TENSOR_LAYER_OUT_NORM,  "blk.%d.layer_output_norm" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_CLS,             "cls" },{ LLM_TENSOR_CLS_OUT,         "cls.output" },},},{LLM_ARCH_NOMIC_BERT,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" },{ LLM_TENSOR_TOKEN_TYPES,     "token_types" },{ LLM_TENSOR_ATTN_OUT_NORM,   "blk.%d.attn_output_norm" },{ LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_LAYER_OUT_NORM,  "blk.%d.layer_output_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_JINA_BERT_V2,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" },{ LLM_TENSOR_TOKEN_TYPES,     "token_types" },{ LLM_TENSOR_ATTN_NORM_2,     "blk.%d.attn_norm_2" },{ LLM_TENSOR_ATTN_OUT_NORM,   "blk.%d.attn_output_norm" },{ LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },{ LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },{ LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },{ LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },{ LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_LAYER_OUT_NORM,  "blk.%d.layer_output_norm" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_CLS,             "cls" },},},{LLM_ARCH_BLOOM,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },},},{LLM_ARCH_STABLELM,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },{ LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },},},{LLM_ARCH_QWEN,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_QWEN2,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_QWEN2VL,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_QWEN2MOE,{{ LLM_TENSOR_TOKEN_EMBD,         "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,        "output_norm" },{ LLM_TENSOR_OUTPUT,             "output" },{ LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },{ LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },{ LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },{ LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },{ LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" },{ LLM_TENSOR_FFN_GATE_SHEXP,     "blk.%d.ffn_gate_shexp" },{ LLM_TENSOR_FFN_DOWN_SHEXP,     "blk.%d.ffn_down_shexp" },{ LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },},},{LLM_ARCH_PHI2,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },{ 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" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_PHI3,{{ LLM_TENSOR_TOKEN_EMBD,         "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,        "output_norm" },{ LLM_TENSOR_OUTPUT,             "output" },{ LLM_TENSOR_ROPE_FACTORS_LONG,  "rope_factors_long" },{ LLM_TENSOR_ROPE_FACTORS_SHORT, "rope_factors_short" },{ LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_QKV,           "blk.%d.attn_qkv" },{ 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" },{ LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },},},{LLM_ARCH_PHIMOE,{{ LLM_TENSOR_TOKEN_EMBD,         "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,        "output_norm" },{ LLM_TENSOR_OUTPUT,             "output" },{ LLM_TENSOR_ROPE_FACTORS_LONG,  "rope_factors_long" },{ LLM_TENSOR_ROPE_FACTORS_SHORT, "rope_factors_short" },{ LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_QKV,           "blk.%d.attn_qkv" },{ 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" },{ LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },{ LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },{ LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },{ LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },},},{LLM_ARCH_PLAMO,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_CODESHELL,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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_QKV,        "blk.%d.attn_qkv" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_ORION,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_INTERNLM2,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_MINICPM,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ROPE_FACTORS_LONG,  "rope_factors_long" },{ LLM_TENSOR_ROPE_FACTORS_SHORT, "rope_factors_short" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },{ LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_GATE_EXP,    "blk.%d.ffn_gate.%d" },{ LLM_TENSOR_FFN_DOWN_EXP,    "blk.%d.ffn_down.%d" },{ LLM_TENSOR_FFN_UP_EXP,      "blk.%d.ffn_up.%d" },},},{LLM_ARCH_MINICPM3,{{ LLM_TENSOR_TOKEN_EMBD,         "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,        "output_norm" },{ LLM_TENSOR_OUTPUT,             "output" },{ LLM_TENSOR_ROPE_FACTORS_LONG,  "rope_factors_long" },{ LLM_TENSOR_ROPE_FACTORS_SHORT, "rope_factors_short" },{ LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_Q_A_NORM,      "blk.%d.attn_q_a_norm" },{ LLM_TENSOR_ATTN_KV_A_NORM,     "blk.%d.attn_kv_a_norm" },{ LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },{ LLM_TENSOR_ATTN_Q_A,           "blk.%d.attn_q_a" },{ LLM_TENSOR_ATTN_Q_B,           "blk.%d.attn_q_b" },{ LLM_TENSOR_ATTN_KV_A_MQA,      "blk.%d.attn_kv_a_mqa" },{ LLM_TENSOR_ATTN_KV_B,          "blk.%d.attn_kv_b" },{ LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },{ LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },},},{LLM_ARCH_GEMMA,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_GEMMA2,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_POST_NORM,  "blk.%d.post_attention_norm" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_POST_NORM,   "blk.%d.post_ffw_norm" },},},{LLM_ARCH_STARCODER2,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_MAMBA,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ LLM_TENSOR_SSM_IN,          "blk.%d.ssm_in" },{ LLM_TENSOR_SSM_CONV1D,      "blk.%d.ssm_conv1d" },{ LLM_TENSOR_SSM_X,           "blk.%d.ssm_x" },{ 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_OUT,         "blk.%d.ssm_out" },},},{LLM_ARCH_XVERSE,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_COMMAND_R,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },{ LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },},},{LLM_ARCH_COHERE2,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_DBRX,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_ATTN_OUT_NORM,   "blk.%d.attn_output_norm" },{ LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },{ LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },{ LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },{ LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },},},{LLM_ARCH_OLMO,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT,          "output" },{ 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" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_OLMO2,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ 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" },{ LLM_TENSOR_ATTN_POST_NORM,  "blk.%d.post_attention_norm" },{ LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },{ LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },{ LLM_TENSOR_FFN_POST_NORM,   "blk.%d.post_ffw_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_OLMOE,{{ LLM_TENSOR_TOKEN_EMBD,         "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,        "output_norm" },{ LLM_TENSOR_OUTPUT,             "output" },{ LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_Q_NORM,        "blk.%d.attn_q_norm" },{ LLM_TENSOR_ATTN_K_NORM,        "blk.%d.attn_k_norm" },{ LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },{ LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },{ LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },{ LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },},},{LLM_ARCH_OPENELM,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },{ LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },{ LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_ARCTIC,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_NORM_EXPS,   "blk.%d.ffn_norm_exps" },{ LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },{ LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },{ LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },},},{LLM_ARCH_DEEPSEEK,{{ LLM_TENSOR_TOKEN_EMBD,         "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,        "output_norm" },{ LLM_TENSOR_OUTPUT,             "output" },{ LLM_TENSOR_ROPE_FREQS,         "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_ROT_EMBD,      "blk.%d.attn_rot_embd" },{ LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },{ LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },{ LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },{ LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },{ LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" },{ LLM_TENSOR_FFN_GATE_SHEXP,     "blk.%d.ffn_gate_shexp" },{ LLM_TENSOR_FFN_DOWN_SHEXP,     "blk.%d.ffn_down_shexp" },{ LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },},},{LLM_ARCH_DEEPSEEK2,{{ LLM_TENSOR_TOKEN_EMBD,         "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,        "output_norm" },{ LLM_TENSOR_OUTPUT,             "output" },{ LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_Q_A_NORM,      "blk.%d.attn_q_a_norm" },{ LLM_TENSOR_ATTN_KV_A_NORM,     "blk.%d.attn_kv_a_norm" },{ LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },{ LLM_TENSOR_ATTN_Q_A,           "blk.%d.attn_q_a" },{ LLM_TENSOR_ATTN_Q_B,           "blk.%d.attn_q_b" },{ LLM_TENSOR_ATTN_KV_A_MQA,      "blk.%d.attn_kv_a_mqa" },{ LLM_TENSOR_ATTN_KV_B,          "blk.%d.attn_kv_b" },{ LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },{ LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },{ LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },{ LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },{ LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },{ LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" },{ LLM_TENSOR_FFN_GATE_SHEXP,     "blk.%d.ffn_gate_shexp" },{ LLM_TENSOR_FFN_DOWN_SHEXP,     "blk.%d.ffn_down_shexp" },{ LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },{ LLM_TENSOR_FFN_EXP_PROBS_B,    "blk.%d.exp_probs_b" },},},{LLM_ARCH_CHATGLM,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },},},{LLM_ARCH_BITNET,{{ LLM_TENSOR_TOKEN_EMBD,         "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,        "output_norm" },{ 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" },{ LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_SUB_NORM,      "blk.%d.attn_sub_norm" },{ LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_SUB_NORM,       "blk.%d.ffn_sub_norm" },},},{LLM_ARCH_T5,{{ LLM_TENSOR_TOKEN_EMBD,           "token_embd" },{ LLM_TENSOR_OUTPUT,               "output" },{ LLM_TENSOR_DEC_OUTPUT_NORM,      "dec.output_norm" },{ LLM_TENSOR_DEC_ATTN_NORM,        "dec.blk.%d.attn_norm" },{ LLM_TENSOR_DEC_ATTN_Q,           "dec.blk.%d.attn_q" },{ LLM_TENSOR_DEC_ATTN_K,           "dec.blk.%d.attn_k" },{ LLM_TENSOR_DEC_ATTN_V,           "dec.blk.%d.attn_v" },{ LLM_TENSOR_DEC_ATTN_OUT,         "dec.blk.%d.attn_o" },{ LLM_TENSOR_DEC_ATTN_REL_B,       "dec.blk.%d.attn_rel_b" },{ LLM_TENSOR_DEC_CROSS_ATTN_NORM,  "dec.blk.%d.cross_attn_norm" },{ LLM_TENSOR_DEC_CROSS_ATTN_Q,     "dec.blk.%d.cross_attn_q" },{ LLM_TENSOR_DEC_CROSS_ATTN_K,     "dec.blk.%d.cross_attn_k" },{ LLM_TENSOR_DEC_CROSS_ATTN_V,     "dec.blk.%d.cross_attn_v" },{ LLM_TENSOR_DEC_CROSS_ATTN_OUT,   "dec.blk.%d.cross_attn_o" },{ LLM_TENSOR_DEC_CROSS_ATTN_REL_B, "dec.blk.%d.cross_attn_rel_b" },{ LLM_TENSOR_DEC_FFN_NORM,         "dec.blk.%d.ffn_norm" },{ LLM_TENSOR_DEC_FFN_GATE,         "dec.blk.%d.ffn_gate" },{ LLM_TENSOR_DEC_FFN_DOWN,         "dec.blk.%d.ffn_down" },{ LLM_TENSOR_DEC_FFN_UP,           "dec.blk.%d.ffn_up" },{ LLM_TENSOR_ENC_OUTPUT_NORM,      "enc.output_norm" },{ LLM_TENSOR_ENC_ATTN_NORM,        "enc.blk.%d.attn_norm" },{ LLM_TENSOR_ENC_ATTN_Q,           "enc.blk.%d.attn_q" },{ LLM_TENSOR_ENC_ATTN_K,           "enc.blk.%d.attn_k" },{ LLM_TENSOR_ENC_ATTN_V,           "enc.blk.%d.attn_v" },{ LLM_TENSOR_ENC_ATTN_OUT,         "enc.blk.%d.attn_o" },{ LLM_TENSOR_ENC_ATTN_REL_B,       "enc.blk.%d.attn_rel_b" },{ LLM_TENSOR_ENC_FFN_NORM,         "enc.blk.%d.ffn_norm" },{ LLM_TENSOR_ENC_FFN_GATE,         "enc.blk.%d.ffn_gate" },{ LLM_TENSOR_ENC_FFN_DOWN,         "enc.blk.%d.ffn_down" },{ LLM_TENSOR_ENC_FFN_UP,           "enc.blk.%d.ffn_up" },},},{LLM_ARCH_T5ENCODER,{{ LLM_TENSOR_TOKEN_EMBD,           "token_embd" },{ LLM_TENSOR_OUTPUT,               "output" },{ LLM_TENSOR_ENC_OUTPUT_NORM,      "enc.output_norm" },{ LLM_TENSOR_ENC_ATTN_NORM,        "enc.blk.%d.attn_norm" },{ LLM_TENSOR_ENC_ATTN_Q,           "enc.blk.%d.attn_q" },{ LLM_TENSOR_ENC_ATTN_K,           "enc.blk.%d.attn_k" },{ LLM_TENSOR_ENC_ATTN_V,           "enc.blk.%d.attn_v" },{ LLM_TENSOR_ENC_ATTN_OUT,         "enc.blk.%d.attn_o" },{ LLM_TENSOR_ENC_ATTN_REL_B,       "enc.blk.%d.attn_rel_b" },{ LLM_TENSOR_ENC_FFN_NORM,         "enc.blk.%d.ffn_norm" },{ LLM_TENSOR_ENC_FFN_GATE,         "enc.blk.%d.ffn_gate" },{ LLM_TENSOR_ENC_FFN_DOWN,         "enc.blk.%d.ffn_down" },{ LLM_TENSOR_ENC_FFN_UP,           "enc.blk.%d.ffn_up" },},},{LLM_ARCH_JAIS,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },{ LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },},},{LLM_ARCH_NEMOTRON,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_EXAONE,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_RWKV6,{{ LLM_TENSOR_TOKEN_EMBD,                "token_embd" },{ LLM_TENSOR_TOKEN_EMBD_NORM,           "token_embd_norm" },{ LLM_TENSOR_OUTPUT_NORM,               "output_norm" },{ LLM_TENSOR_OUTPUT,                    "output" },{ LLM_TENSOR_ATTN_NORM,                 "blk.%d.attn_norm" },{ LLM_TENSOR_ATTN_NORM_2,               "blk.%d.attn_norm_2" },{ LLM_TENSOR_TIME_MIX_W1,               "blk.%d.time_mix_w1" },{ LLM_TENSOR_TIME_MIX_W2,               "blk.%d.time_mix_w2" },{ LLM_TENSOR_TIME_MIX_LERP_X,           "blk.%d.time_mix_lerp_x" },{ LLM_TENSOR_TIME_MIX_LERP_W,           "blk.%d.time_mix_lerp_w" },{ LLM_TENSOR_TIME_MIX_LERP_K,           "blk.%d.time_mix_lerp_k" },{ LLM_TENSOR_TIME_MIX_LERP_V,           "blk.%d.time_mix_lerp_v" },{ LLM_TENSOR_TIME_MIX_LERP_R,           "blk.%d.time_mix_lerp_r" },{ LLM_TENSOR_TIME_MIX_LERP_G,           "blk.%d.time_mix_lerp_g" },{ LLM_TENSOR_TIME_MIX_LERP_FUSED,       "blk.%d.time_mix_lerp_fused" },{ LLM_TENSOR_TIME_MIX_FIRST,            "blk.%d.time_mix_first" },{ LLM_TENSOR_TIME_MIX_DECAY,            "blk.%d.time_mix_decay" },{ LLM_TENSOR_TIME_MIX_DECAY_W1,         "blk.%d.time_mix_decay_w1" },{ LLM_TENSOR_TIME_MIX_DECAY_W2,         "blk.%d.time_mix_decay_w2" },{ LLM_TENSOR_TIME_MIX_KEY,              "blk.%d.time_mix_key" },{ LLM_TENSOR_TIME_MIX_VALUE,            "blk.%d.time_mix_value" },{ LLM_TENSOR_TIME_MIX_RECEPTANCE,       "blk.%d.time_mix_receptance" },{ LLM_TENSOR_TIME_MIX_GATE,             "blk.%d.time_mix_gate" },{ LLM_TENSOR_TIME_MIX_LN,               "blk.%d.time_mix_ln" },{ LLM_TENSOR_TIME_MIX_OUTPUT,           "blk.%d.time_mix_output" },{ LLM_TENSOR_CHANNEL_MIX_LERP_K,        "blk.%d.channel_mix_lerp_k" },{ LLM_TENSOR_CHANNEL_MIX_LERP_R,        "blk.%d.channel_mix_lerp_r" },{ LLM_TENSOR_CHANNEL_MIX_KEY,           "blk.%d.channel_mix_key" },{ LLM_TENSOR_CHANNEL_MIX_VALUE,         "blk.%d.channel_mix_value" },{ LLM_TENSOR_CHANNEL_MIX_RECEPTANCE,    "blk.%d.channel_mix_receptance" },},},{LLM_ARCH_RWKV6QWEN2,{{ LLM_TENSOR_TOKEN_EMBD,                "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,               "output_norm" },{ LLM_TENSOR_OUTPUT,                    "output" },{ LLM_TENSOR_ATTN_NORM,                 "blk.%d.attn_norm" },{ LLM_TENSOR_TIME_MIX_W1,               "blk.%d.time_mix_w1" },{ LLM_TENSOR_TIME_MIX_W2,               "blk.%d.time_mix_w2" },{ LLM_TENSOR_TIME_MIX_LERP_X,           "blk.%d.time_mix_lerp_x" },{ LLM_TENSOR_TIME_MIX_LERP_FUSED,       "blk.%d.time_mix_lerp_fused" },{ LLM_TENSOR_TIME_MIX_FIRST,            "blk.%d.time_mix_first" },{ LLM_TENSOR_TIME_MIX_DECAY,            "blk.%d.time_mix_decay" },{ LLM_TENSOR_TIME_MIX_DECAY_W1,         "blk.%d.time_mix_decay_w1" },{ LLM_TENSOR_TIME_MIX_DECAY_W2,         "blk.%d.time_mix_decay_w2" },{ LLM_TENSOR_TIME_MIX_KEY,              "blk.%d.time_mix_key" },{ LLM_TENSOR_TIME_MIX_VALUE,            "blk.%d.time_mix_value" },{ LLM_TENSOR_TIME_MIX_RECEPTANCE,       "blk.%d.time_mix_receptance" },{ LLM_TENSOR_TIME_MIX_GATE,             "blk.%d.time_mix_gate" },{ LLM_TENSOR_TIME_MIX_OUTPUT,           "blk.%d.time_mix_output" },{ LLM_TENSOR_FFN_NORM,                  "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,                  "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,                  "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,                    "blk.%d.ffn_up" },},},{LLM_ARCH_GRANITE,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },},},{LLM_ARCH_GRANITE_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" },{ 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" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE_INP,    "blk.%d.ffn_gate_inp" },{ LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },{ LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },{ LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },},},{LLM_ARCH_CHAMELEON,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },{ LLM_TENSOR_OUTPUT_NORM,     "output_norm" },{ LLM_TENSOR_OUTPUT,          "output" },{ LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },{ 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" },{ LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },{ LLM_TENSOR_FFN_GATE,        "blk.%d.ffn_gate" },{ LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },{ LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },{ LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm" },{ LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm" },},},{LLM_ARCH_WAVTOKENIZER_DEC,{{ LLM_TENSOR_TOKEN_EMBD,        "token_embd" },{ LLM_TENSOR_TOKEN_EMBD_NORM,   "token_embd_norm" },{ LLM_TENSOR_CONV1D,            "conv1d" },{ LLM_TENSOR_CONVNEXT_DW,       "convnext.%d.dw" },{ LLM_TENSOR_CONVNEXT_NORM,     "convnext.%d.norm" },{ LLM_TENSOR_CONVNEXT_PW1,      "convnext.%d.pw1" },{ LLM_TENSOR_CONVNEXT_PW2,      "convnext.%d.pw2" },{ LLM_TENSOR_CONVNEXT_GAMMA,    "convnext.%d.gamma" },{ LLM_TENSOR_OUTPUT_NORM,       "output_norm" },{ LLM_TENSOR_OUTPUT,            "output" },{ LLM_TENSOR_POS_NET_CONV1,     "posnet.%d.conv1" },{ LLM_TENSOR_POS_NET_CONV2,     "posnet.%d.conv2" },{ LLM_TENSOR_POS_NET_NORM,      "posnet.%d.norm" },{ LLM_TENSOR_POS_NET_NORM1,     "posnet.%d.norm1" },{ LLM_TENSOR_POS_NET_NORM2,     "posnet.%d.norm2" },{ LLM_TENSOR_POS_NET_ATTN_NORM, "posnet.%d.attn_norm" },{ LLM_TENSOR_POS_NET_ATTN_Q,    "posnet.%d.attn_q" },{ LLM_TENSOR_POS_NET_ATTN_K,    "posnet.%d.attn_k" },{ LLM_TENSOR_POS_NET_ATTN_V,    "posnet.%d.attn_v" },{ LLM_TENSOR_POS_NET_ATTN_OUT,  "posnet.%d.attn_output" },},},{LLM_ARCH_UNKNOWN,{{ LLM_TENSOR_TOKEN_EMBD,      "token_embd" },},},
};

3. struct ggml_cgraph * build_deepseek() and struct ggml_cgraph * build_deepseek2()

/home/yongqiang/llm_work/llama_cpp_25_01_05/llama.cpp/src/llama.cpp

  • struct ggml_cgraph * build_deepseek()
    struct ggml_cgraph * build_deepseek() {struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, model.max_nodes(), false);// mutable variable, needed during the last layer of the computation to skip unused tokensint32_t n_tokens = this->n_tokens;const int64_t n_embd_head = hparams.n_embd_head_v;GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);GGML_ASSERT(n_embd_head == hparams.n_rot);struct ggml_tensor * cur;struct ggml_tensor * inpL;inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);// inp_pos - contains the positionsstruct ggml_tensor * inp_pos = build_inp_pos();// KQ_mask (mask for 1 head, it will be broadcasted to all heads)struct ggml_tensor * KQ_mask = build_inp_KQ_mask();const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;for (int il = 0; il < n_layer; ++il) {struct ggml_tensor * inpSA = inpL;// normcur = llm_build_norm(ctx0, inpL, hparams,model.layers[il].attn_norm, NULL,LLM_NORM_RMS, cb, il);cb(cur, "attn_norm", il);// self-attention{// rope freq factors for llama3; may return nullptr for llama2 and other modelsstruct ggml_tensor * rope_factors = build_rope_factors(il);// compute Q and K and RoPE themstruct ggml_tensor * Qcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);cb(Qcur, "Qcur", il);if (model.layers[il].bq) {Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);cb(Qcur, "Qcur", il);}struct ggml_tensor * Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);cb(Kcur, "Kcur", il);if (model.layers[il].bk) {Kcur = ggml_add(ctx0, Kcur, model.layers[il].bk);cb(Kcur, "Kcur", il);}struct ggml_tensor * Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);cb(Vcur, "Vcur", il);if (model.layers[il].bv) {Vcur = ggml_add(ctx0, Vcur, model.layers[il].bv);cb(Vcur, "Vcur", il);}Qcur = ggml_rope_ext(ctx0, ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens), inp_pos, rope_factors,n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,ext_factor, attn_factor, beta_fast, beta_slow);cb(Qcur, "Qcur", il);Kcur = ggml_rope_ext(ctx0, ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens), inp_pos, rope_factors,n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,ext_factor, attn_factor, beta_fast, beta_slow);cb(Kcur, "Kcur", il);cur = llm_build_kv(ctx0, lctx, kv_self, gf,model.layers[il].wo, model.layers[il].bo,Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, kq_scale, cb, il);}if (il == n_layer - 1) {// skip computing output for unused tokensstruct ggml_tensor * inp_out_ids = build_inp_out_ids();n_tokens = n_outputs;cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);}struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);cb(ffn_inp, "ffn_inp", il);cur = llm_build_norm(ctx0, ffn_inp, hparams,model.layers[il].ffn_norm, NULL,LLM_NORM_RMS, cb, il);cb(cur, "ffn_norm", il);if ((uint32_t) il < hparams.n_layer_dense_lead) {cur = llm_build_ffn(ctx0, lctx, cur,model.layers[il].ffn_up,   NULL, NULL,model.layers[il].ffn_gate, NULL, NULL,model.layers[il].ffn_down, NULL, NULL,NULL,LLM_FFN_SILU, LLM_FFN_PAR, cb, il);cb(cur, "ffn_out", il);} else {// MoE branchggml_tensor * moe_out =llm_build_moe_ffn(ctx0, lctx, cur,model.layers[il].ffn_gate_inp,model.layers[il].ffn_up_exps,model.layers[il].ffn_gate_exps,model.layers[il].ffn_down_exps,nullptr,n_expert, n_expert_used,LLM_FFN_SILU, false,false, hparams.expert_weights_scale,LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,cb, il);cb(moe_out, "ffn_moe_out", il);// FFN shared expert{ggml_tensor * ffn_shexp = llm_build_ffn(ctx0, lctx, cur,model.layers[il].ffn_up_shexp,   NULL, NULL,model.layers[il].ffn_gate_shexp, NULL, NULL,model.layers[il].ffn_down_shexp, NULL, NULL,NULL,LLM_FFN_SILU, LLM_FFN_PAR, cb, il);cb(ffn_shexp, "ffn_shexp", il);cur = ggml_add(ctx0, moe_out, ffn_shexp);cb(cur, "ffn_out", il);}}cur = ggml_add(ctx0, cur, ffn_inp);cur = lctx.cvec.apply_to(ctx0, cur, il);cb(cur, "l_out", il);// input for next layerinpL = cur;}cur = inpL;cur = llm_build_norm(ctx0, cur, hparams,model.output_norm, NULL,LLM_NORM_RMS, cb, -1);cb(cur, "result_norm", -1);// lm_headcur = llm_build_lora_mm(lctx, ctx0, model.output, cur);cb(cur, "result_output", -1);ggml_build_forward_expand(gf, cur);return gf;}
  • struct ggml_cgraph * build_deepseek2()
    struct ggml_cgraph * build_deepseek2() {struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, model.max_nodes(), false);// mutable variable, needed during the last layer of the computation to skip unused tokensint32_t n_tokens = this->n_tokens;bool is_lite = (hparams.n_layer == 27);// We have to pre-scale kq_scale and attn_factor to make the YaRN RoPE work correctly.// See https://github.com/ggerganov/llama.cpp/discussions/7416 for detailed explanation.const float mscale = attn_factor * (1.0f + hparams.rope_yarn_log_mul * logf(1.0f / freq_scale));const float kq_scale = 1.0f*mscale*mscale/sqrtf(float(hparams.n_embd_head_k));const float attn_factor_scaled = 1.0f / (1.0f + 0.1f * logf(1.0f / freq_scale));const uint32_t n_embd_head_qk_rope = hparams.n_rot;const uint32_t n_embd_head_qk_nope = hparams.n_embd_head_k - hparams.n_rot;const uint32_t kv_lora_rank = hparams.n_lora_kv;struct ggml_tensor * cur;struct ggml_tensor * inpL;// {n_embd, n_tokens}inpL = llm_build_inp_embd(ctx0, lctx, hparams, ubatch, model.tok_embd, cb);// inp_pos - contains the positionsstruct ggml_tensor * inp_pos = build_inp_pos();// KQ_mask (mask for 1 head, it will be broadcasted to all heads)struct ggml_tensor * KQ_mask = build_inp_KQ_mask();for (int il = 0; il < n_layer; ++il) {struct ggml_tensor * inpSA = inpL;// normcur = llm_build_norm(ctx0, inpL, hparams,model.layers[il].attn_norm, NULL,LLM_NORM_RMS, cb, il);cb(cur, "attn_norm", il);// self_attention{struct ggml_tensor * q = NULL;if (!is_lite) {// {n_embd, q_lora_rank} * {n_embd, n_tokens} -> {q_lora_rank, n_tokens}q = ggml_mul_mat(ctx0, model.layers[il].wq_a, cur);cb(q, "q", il);q = llm_build_norm(ctx0, q, hparams,model.layers[il].attn_q_a_norm, NULL,LLM_NORM_RMS, cb, il);cb(q, "q", il);// {q_lora_rank, n_head * hparams.n_embd_head_k} * {q_lora_rank, n_tokens} -> {n_head * hparams.n_embd_head_k, n_tokens}q = ggml_mul_mat(ctx0, model.layers[il].wq_b, q);cb(q, "q", il);} else {q = ggml_mul_mat(ctx0, model.layers[il].wq, cur);cb(q, "q", il);}// split into {n_head * n_embd_head_qk_nope, n_tokens}struct ggml_tensor * q_nope = ggml_view_3d(ctx0, q, n_embd_head_qk_nope, n_head, n_tokens,ggml_row_size(q->type, hparams.n_embd_head_k),ggml_row_size(q->type, hparams.n_embd_head_k * n_head),0);cb(q_nope, "q_nope", il);// and {n_head * n_embd_head_qk_rope, n_tokens}struct ggml_tensor * q_pe = ggml_view_3d(ctx0, q, n_embd_head_qk_rope, n_head, n_tokens,ggml_row_size(q->type, hparams.n_embd_head_k),ggml_row_size(q->type, hparams.n_embd_head_k * n_head),ggml_row_size(q->type, n_embd_head_qk_nope));cb(q_pe, "q_pe", il);// {n_embd, kv_lora_rank + n_embd_head_qk_rope} * {n_embd, n_tokens} -> {kv_lora_rank + n_embd_head_qk_rope, n_tokens}struct ggml_tensor * kv_pe_compresseed = ggml_mul_mat(ctx0, model.layers[il].wkv_a_mqa, cur);cb(kv_pe_compresseed, "kv_pe_compresseed", il);// split into {kv_lora_rank, n_tokens}struct ggml_tensor * kv_compressed = ggml_view_2d(ctx0, kv_pe_compresseed, kv_lora_rank, n_tokens,kv_pe_compresseed->nb[1],0);cb(kv_compressed, "kv_compressed", il);// and {n_embd_head_qk_rope, n_tokens}struct ggml_tensor * k_pe = ggml_view_3d(ctx0, kv_pe_compresseed, n_embd_head_qk_rope, 1, n_tokens,kv_pe_compresseed->nb[1],kv_pe_compresseed->nb[1],ggml_row_size(kv_pe_compresseed->type, kv_lora_rank));cb(k_pe, "k_pe", il);kv_compressed = ggml_cont(ctx0, kv_compressed); // TODO: the CUDA backend does not support non-contiguous normkv_compressed = llm_build_norm(ctx0, kv_compressed, hparams,model.layers[il].attn_kv_a_norm, NULL,LLM_NORM_RMS, cb, il);cb(kv_compressed, "kv_compressed", il);// {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)} * {kv_lora_rank, n_tokens} -> {n_head * (n_embd_head_qk_nope + n_embd_head_v), n_tokens}struct ggml_tensor * kv = ggml_mul_mat(ctx0, model.layers[il].wkv_b, kv_compressed);cb(kv, "kv", il);// split into {n_head * n_embd_head_qk_nope, n_tokens}struct ggml_tensor * k_nope = ggml_view_3d(ctx0, kv, n_embd_head_qk_nope, n_head, n_tokens,ggml_row_size(kv->type, n_embd_head_qk_nope + hparams.n_embd_head_v),ggml_row_size(kv->type, n_head * (n_embd_head_qk_nope + hparams.n_embd_head_v)),0);cb(k_nope, "k_nope", il);// and {n_head * n_embd_head_v, n_tokens}struct ggml_tensor * v_states = ggml_view_3d(ctx0, kv, hparams.n_embd_head_v, n_head, n_tokens,ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)),ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)*n_head),ggml_row_size(kv->type, (n_embd_head_qk_nope)));cb(v_states, "v_states", il);v_states = ggml_cont(ctx0, v_states);cb(v_states, "v_states", il);v_states = ggml_view_2d(ctx0, v_states, hparams.n_embd_head_v * n_head, n_tokens,ggml_row_size(kv->type, hparams.n_embd_head_v * n_head),0);cb(v_states, "v_states", il);q_pe = ggml_cont(ctx0, q_pe); // TODO: the CUDA backend used to not support non-cont. RoPE, investigate removing thisq_pe = ggml_rope_ext(ctx0, q_pe, inp_pos, nullptr,n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,ext_factor, attn_factor_scaled, beta_fast, beta_slow);cb(q_pe, "q_pe", il);// shared RoPE keyk_pe = ggml_cont(ctx0, k_pe); // TODO: the CUDA backend used to not support non-cont. RoPE, investigate removing thisk_pe = ggml_rope_ext(ctx0, k_pe, inp_pos, nullptr,n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,ext_factor, attn_factor_scaled, beta_fast, beta_slow);cb(k_pe, "k_pe", il);struct ggml_tensor * q_states = ggml_concat(ctx0, q_nope, q_pe, 0);cb(q_states, "q_states", il);struct ggml_tensor * k_states = ggml_concat(ctx0, k_nope, ggml_repeat(ctx0, k_pe, q_pe), 0);cb(k_states, "k_states", il);cur = llm_build_kv(ctx0, lctx, kv_self, gf,model.layers[il].wo, NULL,k_states, v_states, q_states, KQ_mask, n_tokens, kv_head, n_kv, kq_scale, cb, il);}if (il == n_layer - 1) {// skip computing output for unused tokensstruct ggml_tensor * inp_out_ids = build_inp_out_ids();n_tokens = n_outputs;cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);}struct ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);cb(ffn_inp, "ffn_inp", il);cur = llm_build_norm(ctx0, ffn_inp, hparams,model.layers[il].ffn_norm, NULL,LLM_NORM_RMS, cb, il);cb(cur, "ffn_norm", il);if ((uint32_t) il < hparams.n_layer_dense_lead) {cur = llm_build_ffn(ctx0, lctx, cur,model.layers[il].ffn_up,   NULL, NULL,model.layers[il].ffn_gate, NULL, NULL,model.layers[il].ffn_down, NULL, NULL,NULL,LLM_FFN_SILU, LLM_FFN_PAR, cb, il);cb(cur, "ffn_out", il);} else {// MoE branchggml_tensor * moe_out =llm_build_moe_ffn(ctx0, lctx, cur,model.layers[il].ffn_gate_inp,model.layers[il].ffn_up_exps,model.layers[il].ffn_gate_exps,model.layers[il].ffn_down_exps,model.layers[il].ffn_exp_probs_b,n_expert, n_expert_used,LLM_FFN_SILU, hparams.expert_weights_norm,true, hparams.expert_weights_scale,(enum llama_expert_gating_func_type) hparams.expert_gating_func,cb, il);cb(moe_out, "ffn_moe_out", il);// FFN shared expert{ggml_tensor * ffn_shexp = llm_build_ffn(ctx0, lctx, cur,model.layers[il].ffn_up_shexp,   NULL, NULL,model.layers[il].ffn_gate_shexp, NULL, NULL,model.layers[il].ffn_down_shexp, NULL, NULL,NULL,LLM_FFN_SILU, LLM_FFN_PAR, cb, il);cb(ffn_shexp, "ffn_shexp", il);cur = ggml_add(ctx0, moe_out, ffn_shexp);cb(cur, "ffn_out", il);}}cur = ggml_add(ctx0, cur, ffn_inp);cur = lctx.cvec.apply_to(ctx0, cur, il);cb(cur, "l_out", il);// input for next layerinpL = cur;}cur = inpL;cur = llm_build_norm(ctx0, cur, hparams,model.output_norm, NULL,LLM_NORM_RMS, cb, -1);cb(cur, "result_norm", -1);// lm_headcur = ggml_mul_mat(ctx0, model.output, cur);cb(cur, "result_output", -1);ggml_build_forward_expand(gf, cur);return gf;}
  • case LLM_ARCH_DEEPSEEK: and case LLM_ARCH_DEEPSEEK2:
    switch (model.arch) {case LLM_ARCH_LLAMA:case LLM_ARCH_MINICPM:case LLM_ARCH_GRANITE:case LLM_ARCH_GRANITE_MOE:{result = llm.build_llama();} break;case LLM_ARCH_DECI:{result = llm.build_deci();} break;case LLM_ARCH_BAICHUAN:{result = llm.build_baichuan();} break;case LLM_ARCH_FALCON:{result = llm.build_falcon();} break;case LLM_ARCH_GROK:{result = llm.build_grok();} break;case LLM_ARCH_STARCODER:{result = llm.build_starcoder();} break;case LLM_ARCH_REFACT:{result = llm.build_refact();} break;case LLM_ARCH_BERT:case LLM_ARCH_JINA_BERT_V2:case LLM_ARCH_NOMIC_BERT:{result = llm.build_bert();} break;case LLM_ARCH_BLOOM:{result = llm.build_bloom();} break;case LLM_ARCH_MPT:{result = llm.build_mpt();} break;case LLM_ARCH_STABLELM:{result = llm.build_stablelm();} break;case LLM_ARCH_QWEN:{result = llm.build_qwen();} break;case LLM_ARCH_QWEN2:{result = llm.build_qwen2();} break;case LLM_ARCH_QWEN2VL:{lctx.n_pos_per_token = 4;result = llm.build_qwen2vl();} break;case LLM_ARCH_QWEN2MOE:{result = llm.build_qwen2moe();} break;case LLM_ARCH_PHI2:{result = llm.build_phi2();} break;case LLM_ARCH_PHI3:case LLM_ARCH_PHIMOE:{result = llm.build_phi3();} break;case LLM_ARCH_PLAMO:{result = llm.build_plamo();} break;case LLM_ARCH_GPT2:{result = llm.build_gpt2();} break;case LLM_ARCH_CODESHELL:{result = llm.build_codeshell();} break;case LLM_ARCH_ORION:{result = llm.build_orion();} break;case LLM_ARCH_INTERNLM2:{result = llm.build_internlm2();} break;case LLM_ARCH_MINICPM3:{result = llm.build_minicpm3();} break;case LLM_ARCH_GEMMA:{result = llm.build_gemma();} break;case LLM_ARCH_GEMMA2:{result = llm.build_gemma2();} break;case LLM_ARCH_STARCODER2:{result = llm.build_starcoder2();} break;case LLM_ARCH_MAMBA:{result = llm.build_mamba();} break;case LLM_ARCH_XVERSE:{result = llm.build_xverse();} break;case LLM_ARCH_COMMAND_R:{result = llm.build_command_r();} break;case LLM_ARCH_COHERE2:{result = llm.build_cohere2();} break;case LLM_ARCH_DBRX:{result = llm.build_dbrx();} break;case LLM_ARCH_OLMO:{result = llm.build_olmo();} break;case LLM_ARCH_OLMO2:{result = llm.build_olmo2();} break;case LLM_ARCH_OLMOE:{result = llm.build_olmoe();} break;case LLM_ARCH_OPENELM:{result = llm.build_openelm();} break;case LLM_ARCH_GPTNEOX:{result = llm.build_gptneox();} break;case LLM_ARCH_ARCTIC:{result = llm.build_arctic();} break;case LLM_ARCH_DEEPSEEK:{result = llm.build_deepseek();} break;case LLM_ARCH_DEEPSEEK2:{result = llm.build_deepseek2();} break;case LLM_ARCH_CHATGLM:{result = llm.build_chatglm();} break;case LLM_ARCH_BITNET:{result = llm.build_bitnet();} break;case LLM_ARCH_T5:{if (lctx.is_encoding) {result = llm.build_t5_enc();} else {result = llm.build_t5_dec();}} break;case LLM_ARCH_T5ENCODER:{result = llm.build_t5_enc();} break;case LLM_ARCH_JAIS:{result = llm.build_jais();} break;case LLM_ARCH_NEMOTRON:{result = llm.build_nemotron();} break;case LLM_ARCH_EXAONE:{result = llm.build_exaone();} break;case LLM_ARCH_RWKV6:{result = llm.build_rwkv6();} break;case LLM_ARCH_RWKV6QWEN2:{result = llm.build_rwkv6qwen2();} break;case LLM_ARCH_CHAMELEON:{result = llm.build_chameleon();} break;case LLM_ARCH_WAVTOKENIZER_DEC:{result = llm.build_wavtokenizer_dec();} break;default:GGML_ABORT("fatal error");}

References

[1] Yongqiang Cheng, https://yongqiang.blog.csdn.net/
[2] huggingface/gguf, https://github.com/huggingface/huggingface.js/tree/main/packages/gguf

相关文章：

llama.cpp LLM_ARCH_DEEPSEEK and LLM_ARCH_DEEPSEEK2

llama.cpp LLM_ARCH_DEEPSEEK and LLM_ARCH_DEEPSEEK2 1. LLM_ARCH_DEEPSEEK and LLM_ARCH_DEEPSEEK22. LLM_ARCH_DEEPSEEK and LLM_ARCH_DEEPSEEK23. struct ggml_cgraph * build_deepseek() and struct ggml_cgraph * build_deepseek2()References 不宜吹捧中国大语言模型的同…...

编程日记 2026/4/30 3:45:14

检测到联想鼠标自动调出运行窗口，鼠标自己作为键盘操作

联想鼠标会自动时不时的调用“运行”窗口 然后鼠标自己作为键盘输入 然后打开这个网页 &#xff08;不是点击了什么鼠标外加按键&#xff0c;这个鼠标除了左右和中间滚轮&#xff0c;没有其他按键了&#xff09;...

编程日记 2026/5/6 20:30:55

-bash: ./uninstall.command: /bin/sh^M: 坏的解释器: 没有那个文件或目录

终端报错&#xff1a; -bash: ./uninstall.command: /bin/sh^M: 坏的解释器: 没有那个文件或目录原因&#xff1a;由于文件行尾符不匹配导致的。当脚本文件在Windows环境中创建或编辑后&#xff0c;行尾符为CRLF&#xff08;即回车和换行&#xff0c;\r\n&#xff09;&#xf…...

编程日记 2026/3/1 21:37:24

15天基础内容总复习

总复习 一.day01内容 1.JVM,JRE,JDK的关系 JVM: java虚拟机,用来运行java程序的,JVM本身是不夸平台的,每个操作系统都需要安装针对本操作系统的JVM所以: java通过jvm的不夸平台实现了java的跨平台JRE:java运行环境,包含jvm和核心类库JDK:java开发工具包,包含开发工具和JRE三…...

编程日记 2025/11/18 7:06:10

星火大模型接入及文本生成HTTP流式、非流式接口（JAVA）

文章目录 一、接入星火大模型二、基于JAVA实现HTTP非流式接口1.配置2.接口实现&#xff08;1&#xff09;分析接口请求&#xff08;2&#xff09;代码实现 3.功能测试&#xff08;1&#xff09;测试对话功能&#xff08;2&#xff09;测试记住上下文功能 三、基于JAVA实现HTTP流…...

编程日记 2026/4/26 14:58:17

如何将电脑桌面默认的C盘设置到D盘？详细操作步骤！

将电脑桌面默认的C盘设置到D盘的详细操作步骤&#xff01; 本博文介绍如何将电脑桌面&#xff08;默认为C盘&#xff09;设置在D盘下。 首先&#xff0c;在D盘建立文件夹Desktop&#xff0c;完整的路径为D:\Desktop。winR&#xff0c;输入Regedit命令。&#xff08;或者单击【…...

编程日记 2026/5/6 20:25:45

toRow和markRow的用法以及使用场景

Vue3 Raw API 完整指南 1. toRaw vs markRaw 1.1 基本概念 toRaw: 返回响应式对象的原始对象&#xff0c;用于临时获取原始数据结构&#xff0c;标记过后将会失去响应式markRaw: 标记一个对象永远不会转换为响应式对象&#xff0c;返回对象本身 1.2 使用对比 // toRaw 示例…...

编程日记 2025/12/28 7:23:06

Java中ExecutorService接口介绍、应用场景和示例代码

概述 ExecutorService 是 Java 中用于管理线程池的接口&#xff0c;它属于 java.util.concurrent 包。它提供了用于管理并发任务的功能&#xff0c;包括任务的提交、执行和线程池的生命周期管理。以下是对 ExecutorService 的详细讲解、应用场景和示例代码。 1. 详细讲解 1.…...

编程日记 2026/5/8 23:26:42

java 判断Date是上午还是下午

我要用Java生成表格统计信息&#xff0c;如下图所示&#xff1a; 所以就诞生了本文的内容。 在 Java 里&#xff0c;判断 Date 对象代表的时间是上午还是下午有多种方式&#xff0c;下面为你详细介绍不同的实现方法。 方式一&#xff1a;使用 java.util.Calendar Calendar 类…...

编程日记 2026/5/1 2:57:52

开源 CSS 框架 Tailwind CSS v4.0

开源 CSS 框架 Tailwind CSS v4.0 于 1 月 22 日正式发布&#xff0c;除了显著提升性能、简化配置体验外&#xff0c;还增强了功能特性&#xff0c;具体如下1&#xff1a; 性能提升 采用全新的高性能引擎 Oxide&#xff0c;带来了构建速度的巨大飞跃&#xff1a; 全量构建速度…...

编程日记 2026/4/3 3:34:14

微信小程序中实现进入页面时数字跳动效果(自定义animate-numbers组件)

微信小程序中实现进入页面时数字跳动效果 1. 组件定义,新建animate-numbers组件1.1 index.js1.2 wxml1.3 wxss 2. 使用组件 1. 组件定义,新建animate-numbers组件 1.1 index.js // components/animate-numbers/index.js Component({properties: {number: {type: Number,value…...

编程日记 2026/3/7 11:23:11

Kafka生产者ACK参数与同步复制

目录 生产者的ACK参数 ack等于0 ack等于1&#xff08;默认&#xff09; ack等于-1或all Kafka的同步复制 使用误区 生产者的ACK参数 Kafka的ack机制可以保证生产者发送的消息被broker接收成功。 Kafka producer有三种ack机制 &#xff0c;分别是 0&#xff0c;1&#xf…...

编程日记 2026/2/16 21:32:53

C语言------数组从入门到精通

1.一维数组 目标:通过思维导图了解学习一维数组的核心知识点: 1.1定义 使用 类型名 数组名[数组长度]; 定义数组。 // 示例&#xff1a; int arr[5]; 1.2一维数组初始化 数组的初始化可以分为静态初始化和动态初始化两种方式。 它们的主要区别在于初始化的时机和内存分配的方…...

编程日记 2026/4/30 10:46:26

FLTK - FLTK1.4.1 - 搭建模板，将FLTK自带的实现搬过来做实验

文章目录 FLTK - FLTK1.4.1 - 搭建模板&#xff0c;将FLTK自带的实现搬过来做实验概述笔记my_fltk_test.cppfltk_test.hfltk_test.cxx用adjuster工程试了一下&#xff0c;好使。END FLTK - FLTK1.4.1 - 搭建模板&#xff0c;将FLTK自带的实现搬过来做实验 概述 用fluid搭建UI…...

编程日记 2026/5/2 17:21:30

postgres基准测试工具pgbench如何使用自定义的表结构和自定义sql

使用 pgbench 进行 PostgreSQL 性能测试时&#xff0c;可以自定义表结构和测试脚本来更好地模拟实际使用场景。以下是一个示例&#xff0c;说明如何自定义表结构和测试脚本。 自定义表结构 创建自定义表结构的 SQL 脚本。例如&#xff0c;创建一个名为 custom_schema.sql 的文…...

编程日记 2025/8/11 1:36:31

开发者交流平台项目部署到阿里云服务器教程

本文使用PuTTY软件在本地Windows系统远程控制Linux服务器&#xff1b;其中&#xff0c;Windows系统为Windows 10专业版&#xff0c;Linux系统为CentOS 7.6 64位。 1.工具软件的准备 maven&#xff1a;https://archive.apache.org/dist/maven/maven-3/3.6.1/binaries/apache-m…...

编程日记 2025/8/15 15:30:28

Seed Edge- AGI（人工智能通用智能）长期研究计划

Seed Edge 是字节跳动豆包大模型团队推出的 AGI&#xff08;人工智能通用智能&#xff09;长期研究计划12。以下是对它的具体介绍1&#xff1a; 名称含义 “Seed” 即豆包大模型团队名称&#xff0c;“Edge” 代表最前沿的 AGI 探索&#xff0c;整体意味着该项目将在 AGI 领域…...

编程日记 2026/4/26 5:37:44

DeepSeek学术写作测评第二弹：数据分析、图表解读，效果怎么样？

我是娜姐 迪娜学姐 &#xff0c;一个SCI医学期刊编辑&#xff0c;探索用AI工具提效论文写作和发表。 针对最近全球热议的DeepSeek开源大模型&#xff0c;娜姐昨天分析了关于论文润色、中译英的详细效果测评&#xff1a; DeepSeek学术写作测评第一弹&#xff1a;论文润色&#…...

编程日记 2026/5/5 2:41:14

从单体应用到微服务的迁移过程

目录 1. 理解单体应用与微服务架构2. 微服务架构的优势3. 迁移的步骤步骤 1&#xff1a;评估当前单体应用步骤 2&#xff1a;确定服务边界步骤 3&#xff1a;逐步拆分单体应用步骤 4&#xff1a;微服务的基础设施和工具步骤 5&#xff1a;管理和优化微服务步骤 6&#xff1a;逐…...

编程日记 2026/5/3 6:42:27

Direct2D 极速教程(2) —— 画淳平

极速导航 创建新项目&#xff1a;002-DrawJunpeiWIC 是什么用 WIC 加载图片画淳平 创建新项目&#xff1a;002-DrawJunpei 右键解决方案 -> 添加 -> 新建项目 选择"空项目"&#xff0c;项目名称为 “002-DrawJunpei”&#xff0c;然后按"创建" 将 “…...

编程日记 2026/5/5 8:35:18

Lustre Core 语法 - 比较表达式

概述 Lustre v6 中的 Lustre Core 部分支持的表达式种类中&#xff0c;支持比较表达式。相关的表达式包括 , <>, <, >, <, >。 相应的文法定义为 Expression :: Expression Expression | Expression <> Expression | Expression < Expression |…...

编程日记 2026/3/10 1:43:19

C# 中 [MethodImpl(MethodImplOptions.Synchronized)] 的使用详解

总目录 前言 在C#中&#xff0c;[MethodImpl(MethodImplOptions.Synchronized)] 是一个特性&#xff08;attribute&#xff09;&#xff0c;用于标记方法&#xff0c;使其在执行时自动获得锁。这类似于Java中的 synchronized 关键字&#xff0c;确保同一时刻只有一个线程可以执…...

编程日记 2026/4/16 5:41:45

在win11系统笔记本中使用Ollama部署deepseek制作一个本地AI小助手！原来如此简单！！！

大家新年好啊&#xff0c;明天就是蛇年啦&#xff0c;蛇年快乐&#xff01; 最近DeepSeek真的太火了&#xff0c;我也跟随B站&#xff0c;使用Ollama在一台Win11系统的笔记本电脑部署了DeepSeek。由于我的云服务器性能很差&#xff0c;虽然笔记本的性能也一般&#xff0c;但是…...

编程日记 2026/5/9 13:32:52

03.01、三合一

03.01、[简单] 三合一 1、题目描述 三合一。描述如何只用一个数组来实现三个栈。 你应该实现push(stackNum, value)、pop(stackNum)、isEmpty(stackNum)、peek(stackNum)方法。stackNum表示栈下标&#xff0c;value表示压入的值。 构造函数会传入一个stackSize参数&#xf…...

编程日记 2026/5/11 12:12:38

【Super Tilemap Editor使用详解】（十五）：从 TMX 文件导入地图(Importing from TMX files)

Super Tilemap Editor 支持从 TMX 文件（Tiled Map Editor 的文件格式）导入图块地图。通过导入 TMX 文件，你可以将 Tiled 中设计的地图快速转换为 Unity 中的图块地图，并自动创建图块地图组（Tilemap Group）。以下是详细的导入步骤和准备工作。 一、导入前的准备工作 在导…...

编程日记 2026/5/5 5:48:57

在FreeBSD下安装Ollama并体验DeepSeek r1大模型

在FreeBSD下安装Ollama并体验DeepSeek r1大模型 在FreeBSD下安装Ollama 直接使用pkg安装即可&#xff1a; sudo pkg install ollama 安装完成后&#xff0c;提示&#xff1a; You installed ollama: the AI model runner. To run ollama, plese open 2 terminals. 1. In t…...

编程日记 2025/7/12 19:27:55

低代码系统-产品架构案例介绍、明道云（十一）

明道云HAP-超级应用平台(Hyper Application Platform)&#xff0c;其实就是企业级应用平台&#xff0c;跟微搭类似。 通过自设计底层架构&#xff0c;兼容各种平台&#xff0c;使用低代码做到应用搭建、应用运维。 企业级应用平台最大的特点就是隐藏在冰山下的功能很深&#xf…...

编程日记 2026/5/11 11:55:07

编解码技术：最大秩距离码（Maximum Rank Distance Code)

最大秩距离码&#xff08;Maximum Rank Distance Code&#xff0c;简称MRD码&#xff09;是一类用于处理矩阵或线性空间中错误校正的编码。其主要特点是在矩阵数据结构中具备检测和纠正错误的能力&#xff0c;设计目标是实现给定矩阵尺寸和错误纠正能力下的最大可能码字数。MRD…...

编程日记 2025/8/10 7:33:10

Linux 4.19内核中的内存管理：x86_64架构下的实现与源码解析

在现代操作系统中，内存管理是核心功能之一，它直接影响系统的性能、稳定性和多任务处理能力。Linux 内核在 x86_64 架构下，通过复杂的机制实现了高效的内存管理，涵盖了虚拟内存、分页机制、内存分配、内存映射、内存保护、缓存管理等多个方面。本文将深入探讨这些机制，并结…...

编程日记 2025/12/6 7:03:30

python：taichi 绘制太极图

安装 pip install taichi pip install opencv-python pycairo where ti # -- taichi 高性能可视化 Demo 展览 ti gallery D:\Python39\Lib\site-packages\taichi\examples\algorithm\circle-packing\ 点击图片&#xff0c;执行 circle_packing_image.py 可见 编写 taijitu.py 如…...

编程日记 2026/4/26 18:44:13