tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-24 03:31:31 +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

16
gguf-py/README.md
View File

@ -1,9 +1,9 @@
## gguf
This is a Python package for writing binary files in the [GGUF](https://github.com/ggerganov/ggml/pull/302)
This is a Python package for writing binary files in the [GGUF](https://github.com/ggml-org/ggml/pull/302)
(GGML Universal File) format.
See [convert_hf_to_gguf.py](https://github.com/ggerganov/llama.cpp/blob/master/convert_hf_to_gguf.py)
See [convert_hf_to_gguf.py](https://github.com/ggml-org/llama.cpp/blob/master/convert_hf_to_gguf.py)
as an example for its usage.
## Installation
@ -13,15 +13,17 @@ pip install gguf
## API Examples/Simple Tools
[examples/writer.py](https://github.com/ggerganov/llama.cpp/blob/master/gguf-py/examples/writer.py) — Generates `example.gguf` in the current directory to demonstrate generating a GGUF file. Note that this file cannot be used as a model.
[examples/writer.py](https://github.com/ggml-org/llama.cpp/blob/master/gguf-py/examples/writer.py) — Generates `example.gguf` in the current directory to demonstrate generating a GGUF file. Note that this file cannot be used as a model.
[scripts/gguf_dump.py](https://github.com/ggerganov/llama.cpp/blob/master/gguf-py/scripts/gguf_dump.py) — Dumps a GGUF file's metadata to the console.
[examples/reader.py](https://github.com/ggml-org/llama.cpp/blob/master/gguf-py/examples/reader.py) — Extracts and displays key-value pairs and tensor details from a GGUF file in a readable format.
[scripts/gguf_set_metadata.py](https://github.com/ggerganov/llama.cpp/blob/master/gguf-py/scripts/gguf_set_metadata.py) — Allows changing simple metadata values in a GGUF file by key.
[gguf/scripts/gguf_dump.py](https://github.com/ggml-org/llama.cpp/blob/master/gguf-py/gguf/scripts/gguf_dump.py) — Dumps a GGUF file's metadata to the console.
[scripts/gguf_convert_endian.py](https://github.com/ggerganov/llama.cpp/blob/master/gguf-py/scripts/gguf_convert_endian.py) — Allows converting the endianness of GGUF files.
[gguf/scripts/gguf_set_metadata.py](https://github.com/ggml-org/llama.cpp/blob/master/gguf-py/gguf/scripts/gguf_set_metadata.py) — Allows changing simple metadata values in a GGUF file by key.
[scripts/gguf_new_metadata.py](https://github.com/ggerganov/llama.cpp/blob/master/gguf-py/scripts/gguf_new_metadata.py) — Copies a GGUF file with added/modified/removed metadata values.
[gguf/scripts/gguf_convert_endian.py](https://github.com/ggml-org/llama.cpp/blob/master/gguf-py/gguf/scripts/gguf_convert_endian.py) — Allows converting the endianness of GGUF files.
[gguf/scripts/gguf_new_metadata.py](https://github.com/ggml-org/llama.cpp/blob/master/gguf-py/gguf/scripts/gguf_new_metadata.py) — Copies a GGUF file with added/modified/removed metadata values.
## Development
Maintainers who participate in development of this package are advised to install it in editable mode:

4
gguf-py/examples/reader.py
View File

@ -2,12 +2,14 @@
import logging
import sys
from pathlib import Path
from gguf.gguf_reader import GGUFReader
logger = logging.getLogger("reader")
# Necessary to load the local gguf package
sys.path.insert(0, str(Path(__file__).parent.parent))
from gguf.gguf_reader import GGUFReader
def read_gguf_file(gguf_file_path):
"""

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

64
gguf-py/gguf/gguf_reader.py
View File

@ -6,6 +6,7 @@ from __future__ import annotations
import logging
import os
import sys
from collections import OrderedDict
from typing import Any, Literal, NamedTuple, TypeVar, Union
@ -15,7 +16,6 @@ import numpy.typing as npt
from .quants import quant_shape_to_byte_shape
if __name__ == "__main__":
import sys
from pathlib import Path
# Allow running file in package as a script.
@ -28,6 +28,7 @@ from gguf.constants import (
GGUF_VERSION,
GGMLQuantizationType,
GGUFValueType,
GGUFEndian,
)
logger = logging.getLogger(__name__)
@ -53,6 +54,48 @@ class ReaderField(NamedTuple):
types: list[GGUFValueType] = []
def contents(self, index_or_slice: int | slice = slice(None)) -> Any:
if self.types:
to_string = lambda x: str(x.tobytes(), encoding='utf-8') # noqa: E731
main_type = self.types[0]
if main_type == GGUFValueType.ARRAY:
sub_type = self.types[-1]
if sub_type == GGUFValueType.STRING:
indices = self.data[index_or_slice]
if isinstance(index_or_slice, int):
return to_string(self.parts[indices]) # type: ignore
else:
return [to_string(self.parts[idx]) for idx in indices] # type: ignore
else:
# FIXME: When/if _get_field_parts() support multi-dimensional arrays, this must do so too
# Check if it's unsafe to perform slice optimization on data
# if any(True for idx in self.data if len(self.parts[idx]) != 1):
# optim_slice = slice(None)
# else:
# optim_slice = index_or_slice
# index_or_slice = slice(None)
# if isinstance(optim_slice, int):
# return self.parts[self.data[optim_slice]].tolist()[0]
# else:
# return [pv for idx in self.data[optim_slice] for pv in self.parts[idx].tolist()][index_or_slice]
if isinstance(index_or_slice, int):
return self.parts[self.data[index_or_slice]].tolist()[0]
else:
return [pv for idx in self.data[index_or_slice] for pv in self.parts[idx].tolist()]
if main_type == GGUFValueType.STRING:
return to_string(self.parts[-1])
else:
return self.parts[-1].tolist()[0]
return None
class ReaderTensor(NamedTuple):
name: str
@ -101,10 +144,19 @@ class GGUFReader:
# If we get 0 here that means it's (probably) a GGUF file created for
# the opposite byte order of the machine this script is running on.
self.byte_order = 'S'
temp_version = temp_version.newbyteorder(self.byte_order)
temp_version = temp_version.view(temp_version.dtype.newbyteorder(self.byte_order))
version = temp_version[0]
if version not in READER_SUPPORTED_VERSIONS:
raise ValueError(f'Sorry, file appears to be version {version} which we cannot handle')
if sys.byteorder == "little":
# Host is little endian
host_endian = GGUFEndian.LITTLE
swapped_endian = GGUFEndian.BIG
else:
# Sorry PDP or other weird systems that don't use BE or LE.
host_endian = GGUFEndian.BIG
swapped_endian = GGUFEndian.LITTLE
self.endianess = swapped_endian if self.byte_order == "S" else host_endian
self.fields: OrderedDict[str, ReaderField] = OrderedDict()
self.tensors: list[ReaderTensor] = []
offs += self._push_field(ReaderField(offs, 'GGUF.version', [temp_version], [0], [GGUFValueType.UINT32]))
@ -145,11 +197,8 @@ class GGUFReader:
count = int(count)
itemsize = int(np.empty([], dtype = dtype).itemsize)
end_offs = offset + itemsize * count
return (
self.data[offset:end_offs]
.view(dtype = dtype)[:count]
.newbyteorder(override_order or self.byte_order)
)
arr = self.data[offset:end_offs].view(dtype=dtype)[:count]
return arr.view(arr.dtype.newbyteorder(self.byte_order if override_order is None else override_order))
def _push_field(self, field: ReaderField, skip_sum: bool = False) -> int:
if field.name in self.fields:
@ -191,6 +240,7 @@ class GGUFReader:
offs += int(alen.nbytes)
aparts: list[npt.NDArray[Any]] = [raw_itype, alen]
data_idxs: list[int] = []
# FIXME: Handle multi-dimensional arrays properly instead of flattening
for idx in range(alen[0]):
curr_size, curr_parts, curr_idxs, curr_types = self._get_field_parts(offs, raw_itype[0])
if idx == 0:

37
gguf-py/gguf/gguf_writer.py
View File

@ -26,6 +26,7 @@ from .constants import (
RopeScalingType,
PoolingType,
TokenType,
ExpertGatingFuncType,
)
from .quants import quant_shape_from_byte_shape
@ -631,6 +632,21 @@ class GGUFWriter:
def add_embedding_length(self, length: int) -> None:
self.add_uint32(Keys.LLM.EMBEDDING_LENGTH.format(arch=self.arch), length)
def add_features_length(self, length: int) -> None:
self.add_uint32(Keys.LLM.FEATURES_LENGTH.format(arch=self.arch), length)
def add_posnet_embedding_length(self, length: int) -> None:
self.add_uint32(Keys.PosNet.EMBEDDING_LENGTH.format(arch=self.arch), length)
def add_posnet_block_count(self, length: int) -> None:
self.add_uint32(Keys.PosNet.BLOCK_COUNT.format(arch=self.arch), length)
def add_convnext_embedding_length(self, length: int) -> None:
self.add_uint32(Keys.ConvNext.EMBEDDING_LENGTH.format(arch=self.arch), length)
def add_convnext_block_count(self, length: int) -> None:
self.add_uint32(Keys.ConvNext.BLOCK_COUNT.format(arch=self.arch), length)
def add_block_count(self, length: int) -> None:
self.add_uint32(Keys.LLM.BLOCK_COUNT.format(arch=self.arch), length)
@ -700,6 +716,12 @@ class GGUFWriter:
def add_expert_weights_scale(self, value: float) -> None:
self.add_float32(Keys.LLM.EXPERT_WEIGHTS_SCALE.format(arch=self.arch), value)
def add_expert_weights_norm(self, value: bool) -> None:
self.add_bool(Keys.LLM.EXPERT_WEIGHTS_NORM.format(arch=self.arch), value)
def add_expert_gating_func(self, value: ExpertGatingFuncType) -> None:
self.add_uint32(Keys.LLM.EXPERT_GATING_FUNC.format(arch=self.arch), value.value)
def add_swin_norm(self, value: bool) -> None:
self.add_bool(Keys.LLM.SWIN_NORM.format(arch=self.arch), value)
@ -721,12 +743,21 @@ class GGUFWriter:
def add_wkv_head_size(self, size: int) -> None:
self.add_uint32(Keys.WKV.HEAD_SIZE.format(arch=self.arch), size)
def add_token_shift_count(self, count: int) -> None:
self.add_uint32(Keys.LLM.TOKEN_SHIFT_COUNT.format(arch=self.arch), count)
def add_layer_norm_eps(self, value: float) -> None:
self.add_float32(Keys.Attention.LAYERNORM_EPS.format(arch=self.arch), value)
def add_layer_norm_rms_eps(self, value: float) -> None:
self.add_float32(Keys.Attention.LAYERNORM_RMS_EPS.format(arch=self.arch), value)
def add_group_norm_eps(self, value: float) -> None:
self.add_float32(Keys.Attention.GROUPNORM_EPS.format(arch=self.arch), value)
def add_group_norm_groups(self, value: int) -> None:
self.add_uint32(Keys.Attention.GROUPNORM_GROUPS.format(arch=self.arch), value)
def add_causal_attention(self, value: bool) -> None:
self.add_bool(Keys.Attention.CAUSAL.format(arch=self.arch), value)
@ -751,6 +782,9 @@ class GGUFWriter:
def add_rope_dimension_count(self, count: int) -> None:
self.add_uint32(Keys.Rope.DIMENSION_COUNT.format(arch=self.arch), count)
def add_rope_dimension_sections(self, dims: Sequence[int]) -> None:
self.add_array(Keys.Rope.DIMENSION_SECTIONS.format(arch=self.arch), dims)
def add_rope_freq_base(self, value: float) -> None:
self.add_float32(Keys.Rope.FREQ_BASE.format(arch=self.arch), value)
@ -826,9 +860,6 @@ class GGUFWriter:
def add_pad_token_id(self, id: int) -> None:
self.add_uint32(Keys.Tokenizer.PAD_ID, id)
def add_cls_token_id(self, id: int) -> None:
self.add_uint32(Keys.Tokenizer.CLS_ID, id)
def add_mask_token_id(self, id: int) -> None:
self.add_uint32(Keys.Tokenizer.MASK_ID, id)

0
gguf-py/scripts/__init__.py → gguf-py/gguf/scripts/__init__.py
View File

82
gguf-py/scripts/gguf_convert_endian.py → gguf-py/gguf/scripts/gguf_convert_endian.py
View File

@ -11,8 +11,8 @@ from pathlib import Path
import numpy as np
# Necessary to load the local gguf package
if "NO_LOCAL_GGUF" not in os.environ and (Path(__file__).parent.parent.parent / 'gguf-py').exists():
sys.path.insert(0, str(Path(__file__).parent.parent))
if "NO_LOCAL_GGUF" not in os.environ and (Path(__file__).parent.parent.parent.parent / 'gguf-py').exists():
sys.path.insert(0, str(Path(__file__).parent.parent.parent))
import gguf
@ -20,22 +20,15 @@ logger = logging.getLogger("gguf-convert-endian")
def convert_byteorder(reader: gguf.GGUFReader, args: argparse.Namespace) -> None:
if np.uint32(1) == np.uint32(1).newbyteorder("<"):
# Host is little endian
host_endian = "little"
swapped_endian = "big"
file_endian = reader.endianess.name
if reader.byte_order == 'S':
host_endian = 'BIG' if file_endian == 'LITTLE' else 'LITTLE'
else:
# Sorry PDP or other weird systems that don't use BE or LE.
host_endian = "big"
swapped_endian = "little"
if reader.byte_order == "S":
file_endian = swapped_endian
else:
file_endian = host_endian
order = host_endian if args.order == "native" else args.order
logger.info(f"* Host is {host_endian.upper()} endian, GGUF file seems to be {file_endian.upper()} endian")
host_endian = file_endian
order = host_endian if args.order == "native" else args.order.upper()
logger.info(f"* Host is {host_endian} endian, GGUF file seems to be {file_endian} endian")
if file_endian == order:
logger.info(f"* File is already {order.upper()} endian. Nothing to do.")
logger.info(f"* File is already {order} endian. Nothing to do.")
sys.exit(0)
logger.info("* Checking tensors for conversion compatibility")
for tensor in reader.tensors:
@ -43,9 +36,11 @@ def convert_byteorder(reader: gguf.GGUFReader, args: argparse.Namespace) -> None
gguf.GGMLQuantizationType.F32,
gguf.GGMLQuantizationType.F16,
gguf.GGMLQuantizationType.Q8_0,
gguf.GGMLQuantizationType.Q4_K,
gguf.GGMLQuantizationType.Q6_K,
):
raise ValueError(f"Cannot handle type {tensor.tensor_type.name} for tensor {repr(tensor.name)}")
logger.info(f"* Preparing to convert from {file_endian.upper()} to {order.upper()}")
logger.info(f"* Preparing to convert from {file_endian} to {order}")
if args.dry_run:
return
logger.warning("*** Warning *** Warning *** Warning **")
@ -96,6 +91,59 @@ def convert_byteorder(reader: gguf.GGUFReader, args: argparse.Namespace) -> None
if block_num % 100000 == 0:
inner_pbar.set_description(f"Byte-swapping Blocks [{(n_blocks - block_num) // n_blocks}]")
elif tensor.tensor_type == gguf.GGMLQuantizationType.Q4_K:
# Handle Q4_K tensor blocks (block_q4_k)
# Specific handling of block_q4_k is required.
# Each block_q4_k consists of 2 f16 values followed by 140 int8 values.
# first flatten structure
newshape = 1
for i in tensor.data.shape:
newshape *= i
tensor.data.resize(newshape)
block_size = 144
n_blocks = len(tensor.data) // block_size
for block_num in (inner_pbar := tqdm(range(n_blocks), desc="Byte-swapping Blocks", leave=False)):
block_offs = block_num * block_size
# Byte-Swap f16 sized fields
delta = tensor.data[block_offs:block_offs + 2].view(dtype=np.uint16)
delta.byteswap(inplace=True)
delta = tensor.data[block_offs + 2:block_offs + 4].view(dtype=np.uint16)
delta.byteswap(inplace=True)
# Byte-Swap
if block_num % 100000 == 0:
inner_pbar.set_description(f"Byte-swapping Blocks [{(n_blocks - block_num) // n_blocks}]")
elif tensor.tensor_type == gguf.GGMLQuantizationType.Q6_K:
# Handle Q6_K tensor blocks (block_q6_k)
# Specific handling of block_q6_k is required.
# Each block_q6_k consists of 208 int8 values followed by 1 f16 value.
# first flatten structure
newshape = 1
for i in tensor.data.shape:
newshape *= i
tensor.data.resize(newshape)
block_size = 210
n_blocks = len(tensor.data) // block_size
for block_num in (inner_pbar := tqdm(range(n_blocks), desc="Byte-swapping Blocks", leave=False)):
block_offs = block_num * block_size
# Byte-Swap f16 sized field
delta = tensor.data[block_offs + 208:block_offs + 210].view(dtype=np.uint16)
delta.byteswap(inplace=True)
# Byte-Swap
if block_num % 100000 == 0:
inner_pbar.set_description(f"Byte-swapping Blocks [{(n_blocks - block_num) // n_blocks}]")
else:
# Handle other tensor types
tensor.data.byteswap(inplace=True)

40
gguf-py/scripts/gguf_dump.py → gguf-py/gguf/scripts/gguf_dump.py
View File

@ -9,11 +9,9 @@ import sys
from pathlib import Path
from typing import Any
import numpy as np
# Necessary to load the local gguf package
if "NO_LOCAL_GGUF" not in os.environ and (Path(__file__).parent.parent.parent / 'gguf-py').exists():
sys.path.insert(0, str(Path(__file__).parent.parent))
if "NO_LOCAL_GGUF" not in os.environ and (Path(__file__).parent.parent.parent.parent / 'gguf-py').exists():
sys.path.insert(0, str(Path(__file__).parent.parent.parent))
from gguf import GGUFReader, GGUFValueType, ReaderTensor # noqa: E402
@ -21,11 +19,11 @@ logger = logging.getLogger("gguf-dump")
def get_file_host_endian(reader: GGUFReader) -> tuple[str, str]:
host_endian = 'LITTLE' if np.uint32(1) == np.uint32(1).newbyteorder("<") else 'BIG'
file_endian = reader.endianess.name
if reader.byte_order == 'S':
file_endian = 'BIG' if host_endian == 'LITTLE' else 'LITTLE'
host_endian = 'BIG' if file_endian == 'LITTLE' else 'LITTLE'
else:
file_endian = host_endian
host_endian = file_endian
return (host_endian, file_endian)
@ -45,12 +43,20 @@ def dump_metadata(reader: GGUFReader, args: argparse.Namespace) -> None:
pretty_type = str(field.types[-1].name)
log_message = f' {n:5}: {pretty_type:10} | {len(field.data):8} | {field.name}'
if len(field.types) == 1:
if field.types:
curr_type = field.types[0]
if curr_type == GGUFValueType.STRING:
log_message += ' = {0}'.format(repr(str(bytes(field.parts[-1]), encoding='utf-8')[:60]))
elif field.types[0] in reader.gguf_scalar_to_np:
log_message += ' = {0}'.format(field.parts[-1][0])
content = field.contents()
if len(content) > 60:
content = content[:57] + '...'
log_message += ' = {0}'.format(repr(content))
elif curr_type in reader.gguf_scalar_to_np:
log_message += ' = {0}'.format(field.contents())
else:
content = repr(field.contents(slice(6)))
if len(field.data) > 6:
content = content[:-1] + ', ...]'
log_message += ' = {0}'.format(content)
print(log_message) # noqa: NP100
if args.no_tensors:
return
@ -82,15 +88,9 @@ def dump_metadata_json(reader: GGUFReader, args: argparse.Namespace) -> None:
curr["array_types"] = [t.name for t in field.types][1:]
if not args.json_array:
continue
itype = field.types[-1]
if itype == GGUFValueType.STRING:
curr["value"] = [str(bytes(field.parts[idx]), encoding="utf-8") for idx in field.data]
else:
curr["value"] = [pv for idx in field.data for pv in field.parts[idx].tolist()]
elif field.types[0] == GGUFValueType.STRING:
curr["value"] = str(bytes(field.parts[-1]), encoding="utf-8")
curr["value"] = field.contents()
else:
curr["value"] = field.parts[-1].tolist()[0]
curr["value"] = field.contents()
if not args.no_tensors:
for idx, tensor in enumerate(reader.tensors):
tensors[tensor.name] = {
@ -181,7 +181,7 @@ def element_count_rounded_notation(count: int) -> str:
def translate_tensor_name(name):
words = name.split(".")
# Source: https://github.com/ggerganov/ggml/blob/master/docs/gguf.md#standardized-tensor-names
# Source: https://github.com/ggml-org/ggml/blob/master/docs/gguf.md#standardized-tensor-names
abbreviation_dictionary = {
'token_embd': 'Token embedding',
'pos_embd': 'Position embedding',

4
gguf-py/scripts/gguf_hash.py → gguf-py/gguf/scripts/gguf_hash.py
View File

@ -13,8 +13,8 @@ from pathlib import Path
from tqdm import tqdm
# Necessary to load the local gguf package
if "NO_LOCAL_GGUF" not in os.environ and (Path(__file__).parent.parent.parent / 'gguf-py').exists():
sys.path.insert(0, str(Path(__file__).parent.parent))
if "NO_LOCAL_GGUF" not in os.environ and (Path(__file__).parent.parent.parent.parent / 'gguf-py').exists():
sys.path.insert(0, str(Path(__file__).parent.parent.parent))
from gguf import GGUFReader # noqa: E402

47
gguf-py/scripts/gguf_new_metadata.py → gguf-py/gguf/scripts/gguf_new_metadata.py
View File

@ -8,13 +8,12 @@ import sys
import json
from pathlib import Path
import numpy as np
from tqdm import tqdm
from typing import Any, Sequence, NamedTuple
# Necessary to load the local gguf package
if "NO_LOCAL_GGUF" not in os.environ and (Path(__file__).parent.parent.parent / 'gguf-py').exists():
sys.path.insert(0, str(Path(__file__).parent.parent))
if "NO_LOCAL_GGUF" not in os.environ and (Path(__file__).parent.parent.parent.parent / 'gguf-py').exists():
sys.path.insert(0, str(Path(__file__).parent.parent.parent))
import gguf
@ -27,45 +26,10 @@ class MetadataDetails(NamedTuple):
description: str = ''
def get_byteorder(reader: gguf.GGUFReader) -> gguf.GGUFEndian:
if np.uint32(1) == np.uint32(1).newbyteorder("<"):
# Host is little endian
host_endian = gguf.GGUFEndian.LITTLE
swapped_endian = gguf.GGUFEndian.BIG
else:
# Sorry PDP or other weird systems that don't use BE or LE.
host_endian = gguf.GGUFEndian.BIG
swapped_endian = gguf.GGUFEndian.LITTLE
if reader.byte_order == "S":
return swapped_endian
else:
return host_endian
def decode_field(field: gguf.ReaderField | None) -> Any:
if field and field.types:
main_type = field.types[0]
if main_type == gguf.GGUFValueType.ARRAY:
sub_type = field.types[-1]
if sub_type == gguf.GGUFValueType.STRING:
return [str(bytes(field.parts[idx]), encoding='utf-8') for idx in field.data]
else:
return [pv for idx in field.data for pv in field.parts[idx].tolist()]
if main_type == gguf.GGUFValueType.STRING:
return str(bytes(field.parts[-1]), encoding='utf-8')
else:
return field.parts[-1][0]
return None
def get_field_data(reader: gguf.GGUFReader, key: str) -> Any:
field = reader.get_field(key)
return decode_field(field)
return field.contents() if field else None
def find_token(token_list: Sequence[int], token: str) -> Sequence[int]:
@ -93,7 +57,7 @@ def copy_with_new_metadata(reader: gguf.GGUFReader, writer: gguf.GGUFWriter, new
logger.debug(f'Removing {field.name}')
continue
old_val = MetadataDetails(field.types[0], decode_field(field))
old_val = MetadataDetails(field.types[0], field.contents())
val = new_metadata.get(field.name, old_val)
if field.name in new_metadata:
@ -192,7 +156,6 @@ def main() -> None:
reader = gguf.GGUFReader(args.input, 'r')
arch = get_field_data(reader, gguf.Keys.General.ARCHITECTURE)
endianess = get_byteorder(reader)
token_list = get_field_data(reader, gguf.Keys.Tokenizer.LIST) or []
@ -230,7 +193,7 @@ def main() -> None:
sys.exit(0)
logger.info(f'* Writing: {args.output}')
writer = gguf.GGUFWriter(args.output, arch=arch, endianess=endianess)
writer = gguf.GGUFWriter(args.output, arch=arch, endianess=reader.endianess)
alignment = get_field_data(reader, gguf.Keys.General.ALIGNMENT)
if alignment is not None:

4
gguf-py/scripts/gguf_set_metadata.py → gguf-py/gguf/scripts/gguf_set_metadata.py
View File

@ -6,8 +6,8 @@ import sys
from pathlib import Path
# Necessary to load the local gguf package
if "NO_LOCAL_GGUF" not in os.environ and (Path(__file__).parent.parent.parent / 'gguf-py').exists():
sys.path.insert(0, str(Path(__file__).parent.parent))
if "NO_LOCAL_GGUF" not in os.environ and (Path(__file__).parent.parent.parent.parent / 'gguf-py').exists():
sys.path.insert(0, str(Path(__file__).parent.parent.parent))
from gguf import GGUFReader # noqa: E402

160
gguf-py/gguf/tensor_mapping.py
View File

@ -13,7 +13,7 @@ class TensorNameMap:
"transformer.wte", # gpt2 gpt-j mpt refact qwen dbrx jais exaone
"transformer.word_embeddings", # falcon
"word_embeddings", # bloom
"model.embed_tokens", # llama-hf nemotron olmoe olmo_1124
"model.embed_tokens", # llama-hf nemotron olmoe olmo2 rwkv6qwen2
"tok_embeddings", # llama-pth
"embeddings.word_embeddings", # bert nomic-bert
"language_model.embedding.word_embeddings", # persimmon
@ -42,6 +42,7 @@ class TensorNameMap:
"emb_ln", # nomic-bert
"transformer.norm", # openelm
"rwkv.blocks.0.pre_ln", # rwkv
"backbone.norm", # wavtokenizer
),
# Position embeddings
@ -54,19 +55,20 @@ class TensorNameMap:
# Output
MODEL_TENSOR.OUTPUT: (
"embed_out", # gptneox
"lm_head", # gpt2 mpt falcon llama-hf baichuan qwen mamba dbrx jais nemotron exaone olmoe olmo_1124
"lm_head", # gpt2 mpt falcon llama-hf baichuan qwen mamba dbrx jais nemotron exaone olmoe olmo2 phimoe
"output", # llama-pth bloom internlm2
"word_embeddings_for_head", # persimmon
"lm_head.linear", # phi2
"output_layer", # chatglm
"head", # rwkv
"head.out", # wavtokenizer
),
# Output norm
MODEL_TENSOR.OUTPUT_NORM: (
"gpt_neox.final_layer_norm", # gptneox
"transformer.ln_f", # gpt2 gpt-j falcon jais exaone
"model.norm", # llama-hf baichuan internlm2 olmoe olmo_1124
"model.norm", # llama-hf baichuan internlm2 olmoe olmo2 phimoe
"norm", # llama-pth
"transformer.norm_f", # mpt dbrx
"ln_f", # refact bloom qwen gpt2
@ -80,6 +82,7 @@ class TensorNameMap:
"transformer.norm", # openelm
"model.norm", # nemotron
"rwkv.ln_out", # rwkv
"backbone.final_layer_norm", # wavtokenizer
),
# Rope frequencies
@ -90,6 +93,10 @@ class TensorNameMap:
MODEL_TENSOR.ROPE_FACTORS_LONG: (),
MODEL_TENSOR.ROPE_FACTORS_SHORT: (),
MODEL_TENSOR.CONV1D: (
"backbone.embed", # roberta
),
}
block_mappings_cfg: dict[MODEL_TENSOR, tuple[str, ...]] = {
@ -101,7 +108,7 @@ class TensorNameMap:
"transformer.h.{bid}.input_layernorm", # falcon7b
"h.{bid}.input_layernorm", # bloom
"transformer.h.{bid}.ln_mlp", # falcon40b
"model.layers.{bid}.input_layernorm", # llama-hf nemotron olmoe
"model.layers.{bid}.input_layernorm", # llama-hf nemotron olmoe phimoe
"layers.{bid}.attention_norm", # llama-pth
"language_model.encoder.layers.{bid}.input_layernorm", # persimmon
"model.layers.{bid}.ln1", # yi
@ -145,7 +152,8 @@ class TensorNameMap:
# Attention query
MODEL_TENSOR.ATTN_Q: (
"model.layers.{bid}.self_attn.q_proj", # llama-hf nemotron olmoe olmo_1124
"model.layers.{bid}.self_attn.q_proj", # llama-hf nemotron olmoe olmo2 phimoe
"model.layers.{bid}.self_attn.q_proj_no_perm", # llama-custom
"layers.{bid}.attention.wq", # llama-pth
"encoder.layer.{bid}.attention.self.query", # bert
"transformer.h.{bid}.attn.q_proj", # gpt-j
@ -157,7 +165,8 @@ class TensorNameMap:
# Attention key
MODEL_TENSOR.ATTN_K: (
"model.layers.{bid}.self_attn.k_proj", # llama-hf nemotron olmoe olmo_1124
"model.layers.{bid}.self_attn.k_proj", # llama-hf nemotron olmoe olmo2 phimoe
"model.layers.{bid}.self_attn.k_proj_no_perm", # llama-custom
"layers.{bid}.attention.wk", # llama-pth
"encoder.layer.{bid}.attention.self.key", # bert
"transformer.h.{bid}.attn.k_proj", # gpt-j
@ -170,7 +179,7 @@ class TensorNameMap:
# Attention value
MODEL_TENSOR.ATTN_V: (
"model.layers.{bid}.self_attn.v_proj", # llama-hf nemotron olmoe olmo_1124
"model.layers.{bid}.self_attn.v_proj", # llama-hf nemotron olmoe olmo2 phimoe
"layers.{bid}.attention.wv", # llama-pth
"encoder.layer.{bid}.attention.self.value", # bert
"transformer.h.{bid}.attn.v_proj", # gpt-j
@ -188,7 +197,8 @@ class TensorNameMap:
"transformer.blocks.{bid}.attn.out_proj", # mpt
"transformer.h.{bid}.self_attention.dense", # falcon
"h.{bid}.self_attention.dense", # bloom
"model.layers.{bid}.self_attn.o_proj", # llama-hf nemotron olmoe olmo_1124
"model.layers.{bid}.self_attn.o_proj", # llama-hf nemotron olmoe olmo2 phimoe
"model.layers.{bid}.self_attn.linear_attn", # deci
"layers.{bid}.attention.wo", # llama-pth
"encoder.layer.{bid}.attention.output.dense", # bert
"transformer.h.{bid}.attn.out_proj", # gpt-j
@ -215,7 +225,7 @@ class TensorNameMap:
),
MODEL_TENSOR.ATTN_POST_NORM: (
"model.layers.{bid}.post_attention_layernorm", # gemma2 olmo_1124
"model.layers.{bid}.post_attention_layernorm", # gemma2 olmo2
),
# Rotary embeddings
@ -232,7 +242,7 @@ class TensorNameMap:
"transformer.h.{bid}.ln_2", # gpt2 refact qwen jais exaone
"h.{bid}.post_attention_layernorm", # bloom
"transformer.blocks.{bid}.norm_2", # mpt
"model.layers.{bid}.post_attention_layernorm", # llama-hf nemotron olmoe
"model.layers.{bid}.post_attention_layernorm", # llama-hf nemotron olmoe phimoe
"layers.{bid}.ffn_norm", # llama-pth
"language_model.encoder.layers.{bid}.post_attention_layernorm", # persimmon
"model.layers.{bid}.ln2", # yi
@ -250,12 +260,12 @@ class TensorNameMap:
# Post feed-forward norm
MODEL_TENSOR.FFN_POST_NORM: (
"model.layers.{bid}.post_feedforward_layernorm", # gemma2 olmo_1124
"model.layers.{bid}.post_feedforward_layernorm", # gemma2 olmo2
),
MODEL_TENSOR.FFN_GATE_INP: (
"layers.{bid}.feed_forward.gate", # mixtral
"model.layers.{bid}.block_sparse_moe.gate", # mixtral
"model.layers.{bid}.block_sparse_moe.gate", # mixtral phimoe
"model.layers.{bid}.mlp.gate", # qwen2moe olmoe
"transformer.decoder_layer.{bid}.router", # Grok
"transformer.blocks.{bid}.ffn.router.layer", # dbrx
@ -266,6 +276,10 @@ class TensorNameMap:
"model.layers.{bid}.mlp.shared_expert_gate", # qwen2moe
),
MODEL_TENSOR.FFN_EXP_PROBS_B: (
"model.layers.{bid}.mlp.gate.e_score_correction", # deepseek-v3
),
# Feed-forward up
MODEL_TENSOR.FFN_UP: (
"gpt_neox.layers.{bid}.mlp.dense_h_to_4h", # gptneox
@ -273,7 +287,7 @@ class TensorNameMap:
"transformer.blocks.{bid}.ffn.up_proj", # mpt
"transformer.h.{bid}.mlp.dense_h_to_4h", # falcon
"h.{bid}.mlp.dense_h_to_4h", # bloom
"model.layers.{bid}.mlp.up_proj", # llama-hf refact nemotron olmo_1124
"model.layers.{bid}.mlp.up_proj", # llama-hf refact nemotron olmo2
"layers.{bid}.feed_forward.w3", # llama-pth
"encoder.layer.{bid}.intermediate.dense", # bert
"transformer.h.{bid}.mlp.fc_in", # gpt-j
@ -296,15 +310,16 @@ class TensorNameMap:
),
MODEL_TENSOR.FFN_UP_EXP: (
"layers.{bid}.feed_forward.experts.w3", # mixtral (merged)
"transformer.decoder_layer.{bid}.moe.linear_v", # Grok (merged)
"transformer.blocks.{bid}.ffn.experts.mlp.v1", # dbrx
"model.layers.{bid}.mlp.experts.up_proj", # qwen2moe olmoe (merged)
"layers.{bid}.feed_forward.experts.w3", # mixtral (merged)
"transformer.decoder_layer.{bid}.moe.linear_v", # Grok (merged)
"transformer.blocks.{bid}.ffn.experts.mlp.v1", # dbrx
"model.layers.{bid}.mlp.experts.up_proj", # qwen2moe olmoe (merged)
"model.layers.{bid}.block_sparse_moe.experts.w3", # phimoe (merged)
),
MODEL_TENSOR.FFN_UP_SHEXP: (
"model.layers.{bid}.mlp.shared_expert.up_proj", # qwen2moe
"model.layers.{bid}.mlp.shared_experts.up_proj", # deepseek2
"model.layers.{bid}.mlp.shared_experts.up_proj", # deepseek deepseek2
),
# AWQ-activation gate
@ -314,7 +329,7 @@ class TensorNameMap:
# Feed-forward gate
MODEL_TENSOR.FFN_GATE: (
"model.layers.{bid}.mlp.gate_proj", # llama-hf refact olmo_1124
"model.layers.{bid}.mlp.gate_proj", # llama-hf refact olmo2
"layers.{bid}.feed_forward.w1", # llama-pth
"transformer.h.{bid}.mlp.w2", # qwen
"transformer.h.{bid}.mlp.c_fc2", # jais
@ -328,15 +343,16 @@ class TensorNameMap:
),
MODEL_TENSOR.FFN_GATE_EXP: (
"layers.{bid}.feed_forward.experts.w1", # mixtral (merged)
"transformer.decoder_layer.{bid}.moe.linear", # Grok (merged)
"transformer.blocks.{bid}.ffn.experts.mlp.w1", # dbrx
"model.layers.{bid}.mlp.experts.gate_proj", # qwen2moe olmoe (merged)
"layers.{bid}.feed_forward.experts.w1", # mixtral (merged)
"transformer.decoder_layer.{bid}.moe.linear", # Grok (merged)
"transformer.blocks.{bid}.ffn.experts.mlp.w1", # dbrx
"model.layers.{bid}.mlp.experts.gate_proj", # qwen2moe olmoe (merged)
"model.layers.{bid}.block_sparse_moe.experts.w1", # phimoe (merged)
),
MODEL_TENSOR.FFN_GATE_SHEXP: (
"model.layers.{bid}.mlp.shared_expert.gate_proj", # qwen2moe
"model.layers.{bid}.mlp.shared_experts.gate_proj", # deepseek2
"model.layers.{bid}.mlp.shared_experts.gate_proj", # deepseek deepseek2
),
# Feed-forward down
@ -346,7 +362,7 @@ class TensorNameMap:
"transformer.blocks.{bid}.ffn.down_proj", # mpt
"transformer.h.{bid}.mlp.dense_4h_to_h", # falcon
"h.{bid}.mlp.dense_4h_to_h", # bloom
"model.layers.{bid}.mlp.down_proj", # llama-hf nemotron olmo_1124
"model.layers.{bid}.mlp.down_proj", # llama-hf nemotron olmo2
"layers.{bid}.feed_forward.w2", # llama-pth
"encoder.layer.{bid}.output.dense", # bert
"transformer.h.{bid}.mlp.fc_out", # gpt-j
@ -373,17 +389,18 @@ class TensorNameMap:
"transformer.blocks.{bid}.ffn.experts.mlp.w2", # dbrx
"model.layers.{bid}.mlp.experts.down_proj", # qwen2moe olmoe (merged)
"model.layers.{bid}.block_sparse_moe.output_linear", # granitemoe
"model.layers.{bid}.block_sparse_moe.experts.w2", # phimoe (merged)
),
MODEL_TENSOR.FFN_DOWN_SHEXP: (
"model.layers.{bid}.mlp.shared_expert.down_proj", # qwen2moe
"model.layers.{bid}.mlp.shared_experts.down_proj", # deepseek2
"model.layers.{bid}.mlp.shared_experts.down_proj", # deepseek deepseek2
),
MODEL_TENSOR.ATTN_Q_NORM: (
"language_model.encoder.layers.{bid}.self_attention.q_layernorm",
"model.layers.{bid}.self_attn.q_layernorm", # persimmon
"model.layers.{bid}.self_attn.q_norm", # cohere olmoe chameleon olmo_1124
"model.layers.{bid}.self_attn.q_norm", # cohere olmoe chameleon olmo2
"transformer.blocks.{bid}.attn.q_ln", # sea-lion
"encoder.layer.{bid}.attention.self.layer_norm_q", # jina-bert-v2
"transformer.layers.{bid}.attn.q_norm", # openelm
@ -392,7 +409,7 @@ class TensorNameMap:
MODEL_TENSOR.ATTN_K_NORM: (
"language_model.encoder.layers.{bid}.self_attention.k_layernorm",
"model.layers.{bid}.self_attn.k_layernorm", # persimmon
"model.layers.{bid}.self_attn.k_norm", # cohere olmoe chameleon olmo_1124
"model.layers.{bid}.self_attn.k_norm", # cohere olmoe chameleon olmo2
"transformer.blocks.{bid}.attn.k_ln", # sea-lion
"encoder.layer.{bid}.attention.self.layer_norm_k", # jina-bert-v2
"transformer.layers.{bid}.attn.k_norm", # openelm
@ -451,34 +468,42 @@ class TensorNameMap:
MODEL_TENSOR.TIME_MIX_W1: (
"rwkv.blocks.{bid}.attention.time_maa_w1", # rwkv v6
"model.layers.{bid}.self_attn.time_maa_w1", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_W2: (
"rwkv.blocks.{bid}.attention.time_maa_w2", # rwkv v6
"model.layers.{bid}.self_attn.time_maa_w2", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_LERP_X: (
"rwkv.blocks.{bid}.attention.time_maa_x", # rwkv v6
"model.layers.{bid}.self_attn.time_maa_x", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_LERP_K: (
"rwkv.blocks.{bid}.attention.time_maa_k", # rwkv v6
"model.layers.{bid}.self_attn.time_maa_k", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_LERP_V: (
"rwkv.blocks.{bid}.attention.time_maa_v", # rwkv v6
"model.layers.{bid}.self_attn.time_maa_v", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_LERP_R: (
"rwkv.blocks.{bid}.attention.time_maa_r", # rwkv v6
"model.layers.{bid}.self_attn.time_maa_r", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_LERP_G: (
"rwkv.blocks.{bid}.attention.time_maa_g", # rwkv v6
"model.layers.{bid}.self_attn.time_maa_g", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_LERP_W: (
"rwkv.blocks.{bid}.attention.time_maa_w", # rwkv v6
"model.layers.{bid}.self_attn.time_maa_w", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_FIRST: (
@ -487,30 +512,37 @@ class TensorNameMap:
MODEL_TENSOR.TIME_MIX_DECAY: (
"rwkv.blocks.{bid}.attention.time_decay", # rwkv v6
"model.layers.{bid}.self_attn.time_decay", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_DECAY_W1: (
"rwkv.blocks.{bid}.attention.time_decay_w1", # rwkv v6
"model.layers.{bid}.self_attn.time_decay_w1", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_DECAY_W2: (
"rwkv.blocks.{bid}.attention.time_decay_w2", # rwkv v6
"model.layers.{bid}.self_attn.time_decay_w2", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_KEY: (
"rwkv.blocks.{bid}.attention.key", # rwkv
"rwkv.blocks.{bid}.attention.key", # rwkv
"model.layers.{bid}.self_attn.k_proj", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_VALUE: (
"rwkv.blocks.{bid}.attention.value", # rwkv
"rwkv.blocks.{bid}.attention.value", # rwkv
"model.layers.{bid}.self_attn.v_proj", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_RECEPTANCE: (
"rwkv.blocks.{bid}.attention.receptance", # rwkv
"model.layers.{bid}.self_attn.q_proj", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_GATE: (
"rwkv.blocks.{bid}.attention.gate", # rwkv
"rwkv.blocks.{bid}.attention.gate", # rwkv
"model.layers.{bid}.self_attn.gate", # rwkv6qwen2
),
MODEL_TENSOR.TIME_MIX_LN: (
@ -518,7 +550,8 @@ class TensorNameMap:
),
MODEL_TENSOR.TIME_MIX_OUTPUT: (
"rwkv.blocks.{bid}.attention.output", # rwkv
"rwkv.blocks.{bid}.attention.output", # rwkv
"model.layers.{bid}.self_attn.o_proj", # rwkv6qwen2
),
MODEL_TENSOR.CHANNEL_MIX_LERP_K: (
@ -683,6 +716,8 @@ class TensorNameMap:
"encoder.block.{bid}.layer.1.DenseReluDense.wo", # t5
),
############################################################################
# TODO: these do not belong to block_mappings_cfg - move them to mappings_cfg
MODEL_TENSOR.ENC_OUTPUT_NORM: (
"encoder.final_layer_norm", # t5
),
@ -695,6 +730,67 @@ class TensorNameMap:
MODEL_TENSOR.CLS_OUT: (
"classifier.out_proj", # roberta
),
#############################################################################
MODEL_TENSOR.CONVNEXT_DW: (
"backbone.convnext.{bid}.dwconv", # wavtokenizer
),
MODEL_TENSOR.CONVNEXT_NORM: (
"backbone.convnext.{bid}.norm", # wavtokenizer
),
MODEL_TENSOR.CONVNEXT_PW1: (
"backbone.convnext.{bid}.pwconv1", # wavtokenizer
),
MODEL_TENSOR.CONVNEXT_PW2: (
"backbone.convnext.{bid}.pwconv2", # wavtokenizer
),
MODEL_TENSOR.CONVNEXT_GAMMA: (
"backbone.convnext.{bid}.gamma", # wavtokenizer
),
MODEL_TENSOR.POSNET_CONV1: (
"backbone.posnet.{bid}.conv1", # wavtokenizer
),
MODEL_TENSOR.POSNET_CONV2: (
"backbone.posnet.{bid}.conv2", # wavtokenizer
),
MODEL_TENSOR.POSNET_NORM: (
"backbone.posnet.{bid}.norm", # wavtokenizer
),
MODEL_TENSOR.POSNET_NORM1: (
"backbone.posnet.{bid}.norm1", # wavtokenizer
),
MODEL_TENSOR.POSNET_NORM2: (
"backbone.posnet.{bid}.norm2", # wavtokenizer
),
MODEL_TENSOR.POSNET_ATTN_NORM: (
"backbone.posnet.{bid}.norm", # wavtokenizer
),
MODEL_TENSOR.POSNET_ATTN_Q: (
"backbone.posnet.{bid}.q", # wavtokenizer
),
MODEL_TENSOR.POSNET_ATTN_K: (
"backbone.posnet.{bid}.k", # wavtokenizer
),
MODEL_TENSOR.POSNET_ATTN_V: (
"backbone.posnet.{bid}.v", # wavtokenizer
),
MODEL_TENSOR.POSNET_ATTN_OUT: (
"backbone.posnet.{bid}.proj_out", # wavtokenizer
),
}
# architecture-specific block mappings

2
gguf-py/gguf/utility.py
View File

@ -47,7 +47,7 @@ def size_label(total_params: int, shared_params: int, expert_params: int, expert
def naming_convention(model_name: str | None, base_name: str | None, finetune_string: str | None, version_string: str | None, size_label: str | None, output_type: str | None, model_type: Literal['vocab', 'LoRA'] | None = None) -> str:
# Reference: https://github.com/ggerganov/ggml/blob/master/docs/gguf.md#gguf-naming-convention
# Reference: https://github.com/ggml-org/ggml/blob/master/docs/gguf.md#gguf-naming-convention
if base_name is not None:
name = base_name.strip().replace(' ', '-').replace('/', '-')

2
gguf-py/gguf/vocab.py
View File

@ -127,7 +127,7 @@ class SpecialVocab:
self.merges = merges
elif isinstance(merges[0], list) and len(merges[0]) == 2 and isinstance(merges[0][0], str):
# New format since transformers 4.45 to support spaces in merges
# ref: https://github.com/ggerganov/llama.cpp/issues/9692
# ref: https://github.com/ggml-org/llama.cpp/issues/9692
# TODO: internally store as the new format instead of converting to old
if any(' ' in s for pair in merges for s in pair):
logger.warning(f'Spaces in merges detected, encoding as {chr(ord(" ") + 256)!r}')

13
gguf-py/pyproject.toml
View File

@ -1,16 +1,15 @@
[tool.poetry]
name = "gguf"
version = "0.10.0"
version = "0.16.0"
description = "Read and write ML models in GGUF for GGML"
authors = ["GGML <ggml@ggml.ai>"]
packages = [
{include = "gguf"},
{include = "gguf/py.typed"},
{include = "scripts"},
]
readme = "README.md"
homepage = "https://ggml.ai"
repository = "https://github.com/ggerganov/llama.cpp"
repository = "https://github.com/ggml-org/llama.cpp"
keywords = ["ggml", "gguf", "llama.cpp"]
classifiers = [
"Programming Language :: Python :: 3",
@ -33,7 +32,7 @@ requires = ["poetry-core>=1.0.0"]
build-backend = "poetry.core.masonry.api"
[tool.poetry.scripts]
gguf-convert-endian = "scripts:gguf_convert_endian_entrypoint"
gguf-dump = "scripts:gguf_dump_entrypoint"
gguf-set-metadata = "scripts:gguf_set_metadata_entrypoint"
gguf-new-metadata = "scripts:gguf_new_metadata_entrypoint"
gguf-convert-endian = "gguf.scripts:gguf_convert_endian_entrypoint"
gguf-dump = "gguf.scripts:gguf_dump_entrypoint"
gguf-set-metadata = "gguf.scripts:gguf_set_metadata_entrypoint"
gguf-new-metadata = "gguf.scripts:gguf_new_metadata_entrypoint"

2
gguf-py/tests/test_quants.py
View File

@ -136,7 +136,7 @@ def compare_tensors(t1: np.ndarray, t2: np.ndarray, qtype: GGMLQuantizationType)
logger.debug(f"Sample bad block ({diff_bits[bad_block_id]} differing bits):\n{t1[bad_block_id]}\nReference:\n{t2[bad_block_id]}")
sum_diff_bits = np.sum(diff_bits)
logger.debug(f"{sum_diff_bits} bits differ ({100 * sum_diff_bits/(x.size * 8):.6f}%)")
logger.debug(f"{sum_diff_bits} bits differ ({100 * sum_diff_bits / (x.size * 8):.6f}%)")
return False