vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge branch 'master' into compilade/bitnet-ternary

Browse source
This commit is contained in:
Francis Couture-Harpin 2024-08-22 16:42:24 -04:00
commit cb6d9962c4
77 changed files with 4681 additions and 2212 deletions
Download patch file Download diff file Expand all files Collapse all files

41
gguf-py/gguf/constants.py
View file

@ -130,6 +130,7 @@ class Keys:
INNER_SIZE = "{arch}.ssm.inner_size"
STATE_SIZE = "{arch}.ssm.state_size"
TIME_STEP_RANK = "{arch}.ssm.time_step_rank"
DT_B_C_RMS = "{arch}.ssm.dt_b_c_rms"
class Tokenizer:
MODEL = "tokenizer.ggml.model"
@ -219,6 +220,8 @@ class MODEL_ARCH(IntEnum):
T5 = auto()
T5ENCODER = auto()
JAIS = auto()
NEMOTRON = auto()
EXAONE = auto()
class MODEL_TENSOR(IntEnum):
@ -347,6 +350,8 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.T5: "t5",
MODEL_ARCH.T5ENCODER: "t5encoder",
MODEL_ARCH.JAIS: "jais",
MODEL_ARCH.NEMOTRON: "nemotron",
MODEL_ARCH.EXAONE: "exaone",
}
TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
@ -1065,6 +1070,37 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.NEMOTRON: [
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_NORM,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.EXAONE: [
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_NORM,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
# TODO
}
@ -1105,6 +1141,10 @@ MODEL_TENSOR_SKIP: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_ARCH.CHATGLM: [
MODEL_TENSOR.ROPE_FREQS,
],
MODEL_ARCH.NEMOTRON: [
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_ROT_EMBD,
],
}
#
@ -1339,6 +1379,7 @@ KEY_SSM_CONV_KERNEL = Keys.SSM.CONV_KERNEL
KEY_SSM_INNER_SIZE = Keys.SSM.INNER_SIZE
KEY_SSM_STATE_SIZE = Keys.SSM.STATE_SIZE
KEY_SSM_TIME_STEP_RANK = Keys.SSM.TIME_STEP_RANK
KEY_SSM_DT_B_C_RMS = Keys.SSM.DT_B_C_RMS
# tokenization
KEY_TOKENIZER_MODEL = Keys.Tokenizer.MODEL