CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-07 09:57:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into compilade/mamba2

Browse Source
This commit is contained in:
Francis Couture-Harpin
2025-02-26 15:32:20 -05:00
parent 1ee6c482d0 a800ae46da
commit c9ecf620d6
702 changed files with 98035 additions and 62199 deletions
Download Patch File Download Diff File Expand all files Collapse all files

451
gguf-py/gguf/constants.py
View File

@@ -90,6 +90,7 @@ class Keys:
VOCAB_SIZE = "{arch}.vocab_size"
CONTEXT_LENGTH = "{arch}.context_length"
EMBEDDING_LENGTH = "{arch}.embedding_length"
FEATURES_LENGTH = "{arch}.features_length"
BLOCK_COUNT = "{arch}.block_count"
LEADING_DENSE_BLOCK_COUNT = "{arch}.leading_dense_block_count"
FEED_FORWARD_LENGTH = "{arch}.feed_forward_length"
@@ -101,6 +102,8 @@ class Keys:
EXPERT_USED_COUNT = "{arch}.expert_used_count"
EXPERT_SHARED_COUNT = "{arch}.expert_shared_count"
EXPERT_WEIGHTS_SCALE = "{arch}.expert_weights_scale"
EXPERT_WEIGHTS_NORM = "{arch}.expert_weights_norm"
EXPERT_GATING_FUNC = "{arch}.expert_gating_func"
POOLING_TYPE = "{arch}.pooling_type"
LOGIT_SCALE = "{arch}.logit_scale"
DECODER_START_TOKEN_ID = "{arch}.decoder_start_token_id"
@@ -112,6 +115,7 @@ class Keys:
TIME_DECAY_EXTRA_DIM = "{arch}.time_decay_extra_dim"
RESIDUAL_SCALE = "{arch}.residual_scale"
EMBEDDING_SCALE = "{arch}.embedding_scale"
TOKEN_SHIFT_COUNT = "{arch}.token_shift_count"
class Attention:
HEAD_COUNT = "{arch}.attention.head_count"
@@ -122,6 +126,8 @@ class Keys:
VALUE_LENGTH = "{arch}.attention.value_length"
LAYERNORM_EPS = "{arch}.attention.layer_norm_epsilon"
LAYERNORM_RMS_EPS = "{arch}.attention.layer_norm_rms_epsilon"
GROUPNORM_EPS = "{arch}.attention.group_norm_epsilon"
GROUPNORM_GROUPS = "{arch}.attention.group_norm_groups"
CAUSAL = "{arch}.attention.causal"
Q_LORA_RANK = "{arch}.attention.q_lora_rank"
KV_LORA_RANK = "{arch}.attention.kv_lora_rank"
@@ -131,6 +137,7 @@ class Keys:
class Rope:
DIMENSION_COUNT = "{arch}.rope.dimension_count"
DIMENSION_SECTIONS = "{arch}.rope.dimension_sections"
FREQ_BASE = "{arch}.rope.freq_base"
SCALING_TYPE = "{arch}.rope.scaling.type"
SCALING_FACTOR = "{arch}.rope.scaling.factor"
@@ -155,6 +162,14 @@ class Keys:
class WKV:
HEAD_SIZE = "{arch}.wkv.head_size"
class PosNet:
EMBEDDING_LENGTH = "{arch}.posnet.embedding_length"
BLOCK_COUNT = "{arch}.posnet.block_count"
class ConvNext:
EMBEDDING_LENGTH = "{arch}.convnext.embedding_length"
BLOCK_COUNT = "{arch}.convnext.block_count"
class Tokenizer:
MODEL = "tokenizer.ggml.model"
PRE = "tokenizer.ggml.pre"
@@ -170,7 +185,6 @@ class Keys:
UNK_ID = "tokenizer.ggml.unknown_token_id"
SEP_ID = "tokenizer.ggml.seperator_token_id"
PAD_ID = "tokenizer.ggml.padding_token_id"
CLS_ID = "tokenizer.ggml.cls_token_id"
MASK_ID = "tokenizer.ggml.mask_token_id"
ADD_BOS = "tokenizer.ggml.add_bos_token"
ADD_EOS = "tokenizer.ggml.add_eos_token"
@@ -209,57 +223,64 @@ class GGUFType:
class MODEL_ARCH(IntEnum):
LLAMA = auto()
FALCON = auto()
BAICHUAN = auto()
GROK = auto()
GPT2 = auto()
GPTJ = auto()
GPTNEOX = auto()
MPT = auto()
STARCODER = auto()
REFACT = auto()
BERT = auto()
NOMIC_BERT = auto()
JINA_BERT_V2 = auto()
BLOOM = auto()
STABLELM = auto()
QWEN = auto()
QWEN2 = auto()
QWEN2MOE = auto()
PHI2 = auto()
PHI3 = auto()
PLAMO = auto()
CODESHELL = auto()
ORION = auto()
INTERNLM2 = auto()
MINICPM = auto()
MINICPM3 = auto()
GEMMA = auto()
GEMMA2 = auto()
STARCODER2 = auto()
RWKV6 = auto()
MAMBA = auto()
MAMBA2 = auto()
XVERSE = auto()
COMMAND_R = auto()
DBRX = auto()
OLMO = auto()
OLMO_1124 = auto()
OLMOE = auto()
OPENELM = auto()
ARCTIC = auto()
DEEPSEEK2 = auto()
CHATGLM = auto()
BITNET = auto()
T5 = auto()
T5ENCODER = auto()
JAIS = auto()
NEMOTRON = auto()
EXAONE = auto()
GRANITE = auto()
GRANITE_MOE = auto()
CHAMELEON = auto()
LLAMA = auto()
DECI = auto()
FALCON = auto()
BAICHUAN = auto()
GROK = auto()
GPT2 = auto()
GPTJ = auto()
GPTNEOX = auto()
MPT = auto()
STARCODER = auto()
REFACT = auto()
BERT = auto()
NOMIC_BERT = auto()
JINA_BERT_V2 = auto()
BLOOM = auto()
STABLELM = auto()
QWEN = auto()
QWEN2 = auto()
QWEN2MOE = auto()
QWEN2VL = auto()
PHI2 = auto()
PHI3 = auto()
PHIMOE = auto()
PLAMO = auto()
CODESHELL = auto()
ORION = auto()
INTERNLM2 = auto()
MINICPM = auto()
MINICPM3 = auto()
GEMMA = auto()
GEMMA2 = auto()
STARCODER2 = auto()
RWKV6 = auto()
RWKV6QWEN2 = auto()
MAMBA = auto()
MAMBA2 = auto()
XVERSE = auto()
COMMAND_R = auto()
COHERE2 = auto()
DBRX = auto()
OLMO = auto()
OLMO2 = auto()
OLMOE = auto()
OPENELM = auto()
ARCTIC = auto()
DEEPSEEK = auto()
DEEPSEEK2 = auto()
CHATGLM = auto()
BITNET = auto()
T5 = auto()
T5ENCODER = auto()
JAIS = auto()
NEMOTRON = auto()
EXAONE = auto()
GRANITE = auto()
GRANITE_MOE = auto()
CHAMELEON = auto()
WAVTOKENIZER_DEC = auto()
class MODEL_TENSOR(IntEnum):
@@ -298,6 +319,7 @@ class MODEL_TENSOR(IntEnum):
FFN_GATE_SHEXP = auto()
FFN_DOWN_SHEXP = auto()
FFN_UP_SHEXP = auto()
FFN_EXP_PROBS_B = auto()
ATTN_Q_NORM = auto()
ATTN_K_NORM = auto()
LAYER_OUT_NORM = auto()
@@ -316,6 +338,7 @@ class MODEL_TENSOR(IntEnum):
TIME_MIX_LERP_V = auto()
TIME_MIX_LERP_R = auto()
TIME_MIX_LERP_G = auto()
TIME_MIX_LERP_FUSED = auto()
TIME_MIX_LERP_W = auto()
TIME_MIX_FIRST = auto()
TIME_MIX_DECAY = auto()
@@ -370,60 +393,83 @@ class MODEL_TENSOR(IntEnum):
ENC_OUTPUT_NORM = auto()
CLS = auto() # classifier
CLS_OUT = auto() # classifier output projection
CONV1D = auto()
CONVNEXT_DW = auto()
CONVNEXT_NORM = auto()
CONVNEXT_PW1 = auto()
CONVNEXT_PW2 = auto()
CONVNEXT_GAMMA = auto()
POSNET_CONV1 = auto()
POSNET_CONV2 = auto()
POSNET_NORM = auto()
POSNET_NORM1 = auto()
POSNET_NORM2 = auto()
POSNET_ATTN_NORM = auto()
POSNET_ATTN_Q = auto()
POSNET_ATTN_K = auto()
POSNET_ATTN_V = auto()
POSNET_ATTN_OUT = auto()
MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.LLAMA: "llama",
MODEL_ARCH.FALCON: "falcon",
MODEL_ARCH.BAICHUAN: "baichuan",
MODEL_ARCH.GROK: "grok",
MODEL_ARCH.GPT2: "gpt2",
MODEL_ARCH.GPTJ: "gptj",
MODEL_ARCH.GPTNEOX: "gptneox",
MODEL_ARCH.MPT: "mpt",
MODEL_ARCH.STARCODER: "starcoder",
MODEL_ARCH.REFACT: "refact",
MODEL_ARCH.BERT: "bert",
MODEL_ARCH.NOMIC_BERT: "nomic-bert",
MODEL_ARCH.JINA_BERT_V2: "jina-bert-v2",
MODEL_ARCH.BLOOM: "bloom",
MODEL_ARCH.STABLELM: "stablelm",
MODEL_ARCH.QWEN: "qwen",
MODEL_ARCH.QWEN2: "qwen2",
MODEL_ARCH.QWEN2MOE: "qwen2moe",
MODEL_ARCH.PHI2: "phi2",
MODEL_ARCH.PHI3: "phi3",
MODEL_ARCH.PLAMO: "plamo",
MODEL_ARCH.CODESHELL: "codeshell",
MODEL_ARCH.ORION: "orion",
MODEL_ARCH.INTERNLM2: "internlm2",
MODEL_ARCH.MINICPM: "minicpm",
MODEL_ARCH.MINICPM3: "minicpm3",
MODEL_ARCH.GEMMA: "gemma",
MODEL_ARCH.GEMMA2: "gemma2",
MODEL_ARCH.STARCODER2: "starcoder2",
MODEL_ARCH.RWKV6: "rwkv6",
MODEL_ARCH.MAMBA: "mamba",
MODEL_ARCH.MAMBA2: "mamba2",
MODEL_ARCH.XVERSE: "xverse",
MODEL_ARCH.COMMAND_R: "command-r",
MODEL_ARCH.DBRX: "dbrx",
MODEL_ARCH.OLMO: "olmo",
MODEL_ARCH.OLMO_1124: "olmo_1124",
MODEL_ARCH.OLMOE: "olmoe",
MODEL_ARCH.OPENELM: "openelm",
MODEL_ARCH.ARCTIC: "arctic",
MODEL_ARCH.DEEPSEEK2: "deepseek2",
MODEL_ARCH.CHATGLM: "chatglm",
MODEL_ARCH.BITNET: "bitnet",
MODEL_ARCH.T5: "t5",
MODEL_ARCH.T5ENCODER: "t5encoder",
MODEL_ARCH.JAIS: "jais",
MODEL_ARCH.NEMOTRON: "nemotron",
MODEL_ARCH.EXAONE: "exaone",
MODEL_ARCH.GRANITE: "granite",
MODEL_ARCH.GRANITE_MOE: "granitemoe",
MODEL_ARCH.CHAMELEON: "chameleon",
MODEL_ARCH.LLAMA: "llama",
MODEL_ARCH.DECI: "deci",
MODEL_ARCH.FALCON: "falcon",
MODEL_ARCH.BAICHUAN: "baichuan",
MODEL_ARCH.GROK: "grok",
MODEL_ARCH.GPT2: "gpt2",
MODEL_ARCH.GPTJ: "gptj",
MODEL_ARCH.GPTNEOX: "gptneox",
MODEL_ARCH.MPT: "mpt",
MODEL_ARCH.STARCODER: "starcoder",
MODEL_ARCH.REFACT: "refact",
MODEL_ARCH.BERT: "bert",
MODEL_ARCH.NOMIC_BERT: "nomic-bert",
MODEL_ARCH.JINA_BERT_V2: "jina-bert-v2",
MODEL_ARCH.BLOOM: "bloom",
MODEL_ARCH.STABLELM: "stablelm",
MODEL_ARCH.QWEN: "qwen",
MODEL_ARCH.QWEN2: "qwen2",
MODEL_ARCH.QWEN2MOE: "qwen2moe",
MODEL_ARCH.QWEN2VL: "qwen2vl",
MODEL_ARCH.PHI2: "phi2",
MODEL_ARCH.PHI3: "phi3",
MODEL_ARCH.PHIMOE: "phimoe",
MODEL_ARCH.PLAMO: "plamo",
MODEL_ARCH.CODESHELL: "codeshell",
MODEL_ARCH.ORION: "orion",
MODEL_ARCH.INTERNLM2: "internlm2",
MODEL_ARCH.MINICPM: "minicpm",
MODEL_ARCH.MINICPM3: "minicpm3",
MODEL_ARCH.GEMMA: "gemma",
MODEL_ARCH.GEMMA2: "gemma2",
MODEL_ARCH.STARCODER2: "starcoder2",
MODEL_ARCH.RWKV6: "rwkv6",
MODEL_ARCH.RWKV6QWEN2: "rwkv6qwen2",
MODEL_ARCH.MAMBA: "mamba",
MODEL_ARCH.MAMBA2: "mamba2",
MODEL_ARCH.XVERSE: "xverse",
MODEL_ARCH.COMMAND_R: "command-r",
MODEL_ARCH.COHERE2: "cohere2",
MODEL_ARCH.DBRX: "dbrx",
MODEL_ARCH.OLMO: "olmo",
MODEL_ARCH.OLMO2: "olmo2",
MODEL_ARCH.OLMOE: "olmoe",
MODEL_ARCH.OPENELM: "openelm",
MODEL_ARCH.ARCTIC: "arctic",
MODEL_ARCH.DEEPSEEK: "deepseek",
MODEL_ARCH.DEEPSEEK2: "deepseek2",
MODEL_ARCH.CHATGLM: "chatglm",
MODEL_ARCH.BITNET: "bitnet",
MODEL_ARCH.T5: "t5",
MODEL_ARCH.T5ENCODER: "t5encoder",
MODEL_ARCH.JAIS: "jais",
MODEL_ARCH.NEMOTRON: "nemotron",
MODEL_ARCH.EXAONE: "exaone",
MODEL_ARCH.GRANITE: "granite",
MODEL_ARCH.GRANITE_MOE: "granitemoe",
MODEL_ARCH.CHAMELEON: "chameleon",
MODEL_ARCH.WAVTOKENIZER_DEC: "wavtokenizer-dec",
}
TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
@@ -464,6 +510,7 @@ TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
MODEL_TENSOR.FFN_GATE_EXP: "blk.{bid}.ffn_gate_exps",
MODEL_TENSOR.FFN_DOWN_EXP: "blk.{bid}.ffn_down_exps",
MODEL_TENSOR.FFN_UP_EXP: "blk.{bid}.ffn_up_exps",
MODEL_TENSOR.FFN_EXP_PROBS_B: "blk.{bid}.exp_probs_b",
MODEL_TENSOR.LAYER_OUT_NORM: "blk.{bid}.layer_output_norm",
MODEL_TENSOR.SSM_IN: "blk.{bid}.ssm_in",
MODEL_TENSOR.SSM_CONV1D: "blk.{bid}.ssm_conv1d",
@@ -480,6 +527,7 @@ TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
MODEL_TENSOR.TIME_MIX_LERP_V: "blk.{bid}.time_mix_lerp_v",
MODEL_TENSOR.TIME_MIX_LERP_R: "blk.{bid}.time_mix_lerp_r",
MODEL_TENSOR.TIME_MIX_LERP_G: "blk.{bid}.time_mix_lerp_g",
MODEL_TENSOR.TIME_MIX_LERP_FUSED: "blk.{bid}.time_mix_lerp_fused",
MODEL_TENSOR.TIME_MIX_LERP_W: "blk.{bid}.time_mix_lerp_w",
MODEL_TENSOR.TIME_MIX_FIRST: "blk.{bid}.time_mix_first",
MODEL_TENSOR.TIME_MIX_DECAY: "blk.{bid}.time_mix_decay",
@@ -534,6 +582,22 @@ TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
MODEL_TENSOR.ENC_OUTPUT_NORM: "enc.output_norm",
MODEL_TENSOR.CLS: "cls",
MODEL_TENSOR.CLS_OUT: "cls.output",
MODEL_TENSOR.CONV1D: "conv1d",
MODEL_TENSOR.CONVNEXT_DW: "convnext.{bid}.dw",
MODEL_TENSOR.CONVNEXT_NORM: "convnext.{bid}.norm",
MODEL_TENSOR.CONVNEXT_PW1: "convnext.{bid}.pw1",
MODEL_TENSOR.CONVNEXT_PW2: "convnext.{bid}.pw2",
MODEL_TENSOR.CONVNEXT_GAMMA: "convnext.{bid}.gamma",
MODEL_TENSOR.POSNET_CONV1: "posnet.{bid}.conv1",
MODEL_TENSOR.POSNET_CONV2: "posnet.{bid}.conv2",
MODEL_TENSOR.POSNET_NORM: "posnet.{bid}.norm",
MODEL_TENSOR.POSNET_NORM1: "posnet.{bid}.norm1",
MODEL_TENSOR.POSNET_NORM2: "posnet.{bid}.norm2",
MODEL_TENSOR.POSNET_ATTN_NORM: "posnet.{bid}.attn_norm",
MODEL_TENSOR.POSNET_ATTN_Q: "posnet.{bid}.attn_q",
MODEL_TENSOR.POSNET_ATTN_K: "posnet.{bid}.attn_k",
MODEL_TENSOR.POSNET_ATTN_V: "posnet.{bid}.attn_v",
MODEL_TENSOR.POSNET_ATTN_OUT: "posnet.{bid}.attn_output",
}
MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
@@ -557,6 +621,26 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
],
MODEL_ARCH.DECI: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.ATTN_ROT_EMBD,
MODEL_TENSOR.FFN_GATE_INP,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
MODEL_TENSOR.FFN_GATE_EXP,
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
],
MODEL_ARCH.GROK: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
@@ -763,6 +847,21 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.QWEN2: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.QWEN2VL: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
@@ -852,6 +951,24 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.PHIMOE: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ROPE_FACTORS_LONG,
MODEL_TENSOR.ROPE_FACTORS_SHORT,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_QKV,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_GATE_INP,
MODEL_TENSOR.FFN_GATE_EXP,
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
],
MODEL_ARCH.CODESHELL: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.POS_EMBD,
@@ -901,6 +1018,8 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ROPE_FACTORS_LONG,
MODEL_TENSOR.ROPE_FACTORS_SHORT,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
@@ -993,6 +1112,7 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.TIME_MIX_LERP_R,
MODEL_TENSOR.TIME_MIX_LERP_G,
MODEL_TENSOR.TIME_MIX_LERP_W,
MODEL_TENSOR.TIME_MIX_LERP_FUSED,
MODEL_TENSOR.TIME_MIX_FIRST,
MODEL_TENSOR.TIME_MIX_DECAY,
MODEL_TENSOR.TIME_MIX_DECAY_W1,
@@ -1009,6 +1129,35 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.CHANNEL_MIX_RECEPTANCE,
MODEL_TENSOR.CHANNEL_MIX_VALUE,
],
MODEL_ARCH.RWKV6QWEN2: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.TIME_MIX_W1,
MODEL_TENSOR.TIME_MIX_W2,
MODEL_TENSOR.TIME_MIX_LERP_X,
MODEL_TENSOR.TIME_MIX_LERP_K,
MODEL_TENSOR.TIME_MIX_LERP_V,
MODEL_TENSOR.TIME_MIX_LERP_R,
MODEL_TENSOR.TIME_MIX_LERP_G,
MODEL_TENSOR.TIME_MIX_LERP_W,
MODEL_TENSOR.TIME_MIX_LERP_FUSED,
MODEL_TENSOR.TIME_MIX_FIRST,
MODEL_TENSOR.TIME_MIX_DECAY,
MODEL_TENSOR.TIME_MIX_DECAY_W1,
MODEL_TENSOR.TIME_MIX_DECAY_W2,
MODEL_TENSOR.TIME_MIX_KEY,
MODEL_TENSOR.TIME_MIX_VALUE,
MODEL_TENSOR.TIME_MIX_RECEPTANCE,
MODEL_TENSOR.TIME_MIX_GATE,
MODEL_TENSOR.TIME_MIX_LN,
MODEL_TENSOR.TIME_MIX_OUTPUT,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.MAMBA: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
@@ -1065,6 +1214,18 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.ATTN_K_NORM,
MODEL_TENSOR.ATTN_Q_NORM,
],
MODEL_ARCH.COHERE2: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.DBRX: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
@@ -1089,7 +1250,7 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.OLMO_1124: [
MODEL_ARCH.OLMO2: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
@@ -1156,6 +1317,29 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
],
MODEL_ARCH.DEEPSEEK: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.ATTN_ROT_EMBD,
MODEL_TENSOR.FFN_GATE_INP,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
MODEL_TENSOR.FFN_GATE_EXP,
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
MODEL_TENSOR.FFN_GATE_SHEXP,
MODEL_TENSOR.FFN_DOWN_SHEXP,
MODEL_TENSOR.FFN_UP_SHEXP,
],
MODEL_ARCH.DEEPSEEK2: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
@@ -1182,6 +1366,7 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_GATE_SHEXP,
MODEL_TENSOR.FFN_DOWN_SHEXP,
MODEL_TENSOR.FFN_UP_SHEXP,
MODEL_TENSOR.FFN_EXP_PROBS_B,
],
MODEL_ARCH.CHATGLM : [
MODEL_TENSOR.TOKEN_EMBD,
@@ -1190,6 +1375,9 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_QKV,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_DOWN,
@@ -1345,6 +1533,28 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.WAVTOKENIZER_DEC: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.TOKEN_EMBD_NORM,
MODEL_TENSOR.CONV1D,
MODEL_TENSOR.CONVNEXT_DW,
MODEL_TENSOR.CONVNEXT_NORM,
MODEL_TENSOR.CONVNEXT_PW1,
MODEL_TENSOR.CONVNEXT_PW2,
MODEL_TENSOR.CONVNEXT_GAMMA,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.POSNET_CONV1,
MODEL_TENSOR.POSNET_CONV2,
MODEL_TENSOR.POSNET_NORM,
MODEL_TENSOR.POSNET_NORM1,
MODEL_TENSOR.POSNET_NORM2,
MODEL_TENSOR.POSNET_ATTN_NORM,
MODEL_TENSOR.POSNET_ATTN_Q,
MODEL_TENSOR.POSNET_ATTN_K,
MODEL_TENSOR.POSNET_ATTN_V,
MODEL_TENSOR.POSNET_ATTN_OUT,
],
# TODO
}
@@ -1354,6 +1564,10 @@ MODEL_TENSOR_SKIP: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_ROT_EMBD,
],
MODEL_ARCH.DECI: [
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_ROT_EMBD,
],
MODEL_ARCH.BAICHUAN: [
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_ROT_EMBD,
@@ -1378,6 +1592,10 @@ MODEL_TENSOR_SKIP: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_ROT_EMBD,
],
MODEL_ARCH.DEEPSEEK: [
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_ROT_EMBD,
],
MODEL_ARCH.DEEPSEEK2: [
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_ROT_EMBD,
@@ -1406,9 +1624,10 @@ class TokenType(IntEnum):
class RopeScalingType(Enum):
NONE = 'none'
LINEAR = 'linear'
YARN = 'yarn'
NONE = 'none'
LINEAR = 'linear'
YARN = 'yarn'
LONGROPE = 'longrope'
class PoolingType(IntEnum):
@@ -1447,13 +1666,15 @@ class GGMLQuantizationType(IntEnum):
F64 = 28
IQ1_M = 29
BF16 = 30
Q4_0_4_4 = 31
Q4_0_4_8 = 32
Q4_0_8_8 = 33
TQ1_0 = 34
TQ2_0 = 35
class ExpertGatingFuncType(IntEnum):
SOFTMAX = 1
SIGMOID = 2
# TODO: add GGMLFileType from ggml_ftype in ggml.h
@@ -1493,9 +1714,9 @@ class LlamaFileType(IntEnum):
MOSTLY_IQ4_XS = 30 # except 1d tensors
MOSTLY_IQ1_M = 31 # except 1d tensors
MOSTLY_BF16 = 32 # except 1d tensors
MOSTLY_Q4_0_4_4 = 33 # except 1d tensors
MOSTLY_Q4_0_4_8 = 34 # except 1d tensors
MOSTLY_Q4_0_8_8 = 35 # except 1d tensors
# MOSTLY_Q4_0_4_4 = 33 # removed from gguf files, use Q4_0 and runtime repack
# MOSTLY_Q4_0_4_8 = 34 # removed from gguf files, use Q4_0 and runtime repack
# MOSTLY_Q4_0_8_8 = 35 # removed from gguf files, use Q4_0 and runtime repack
MOSTLY_TQ1_0 = 36 # except 1d tensors
MOSTLY_TQ2_0 = 37 # except 1d tensors
@@ -1571,9 +1792,6 @@ GGML_QUANT_SIZES: dict[GGMLQuantizationType, tuple[int, int]] = {
GGMLQuantizationType.F64: (1, 8),
GGMLQuantizationType.IQ1_M: (256, QK_K // 8 + QK_K // 16 + QK_K // 32),
GGMLQuantizationType.BF16: (1, 2),
GGMLQuantizationType.Q4_0_4_4:(32, 2 + 16),
GGMLQuantizationType.Q4_0_4_8:(32, 2 + 16),
GGMLQuantizationType.Q4_0_8_8:(32, 2 + 16),
GGMLQuantizationType.TQ1_0: (256, 2 + 4 * 13),
GGMLQuantizationType.TQ2_0: (256, 2 + 64),
}
@@ -1640,7 +1858,6 @@ KEY_TOKENIZER_EOM_ID = Keys.Tokenizer.EOM_ID
KEY_TOKENIZER_UNK_ID = Keys.Tokenizer.UNK_ID
KEY_TOKENIZER_SEP_ID = Keys.Tokenizer.SEP_ID
KEY_TOKENIZER_PAD_ID = Keys.Tokenizer.PAD_ID
KEY_TOKENIZER_CLS_ID = Keys.Tokenizer.CLS_ID
KEY_TOKENIZER_MASK_ID = Keys.Tokenizer.MASK_ID
KEY_TOKENIZER_HF_JSON = Keys.Tokenizer.HF_JSON
KEY_TOKENIZER_RWKV = Keys.Tokenizer.RWKV