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

Fix eos tokens to glm4 and adapts to glm3

Browse source
This commit is contained in:
toyer 2024-06-20 08:43:33 +00:00
commit e773174052
7 changed files with 476 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files

17
gguf-py/gguf/constants.py
View file

@ -149,6 +149,7 @@ class MODEL_ARCH(IntEnum):
OLMO = auto()
ARCTIC = auto()
DEEPSEEK2 = auto()
CHATGLM = auto()
class MODEL_TENSOR(IntEnum):
@ -237,6 +238,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.OLMO: "olmo",
MODEL_ARCH.ARCTIC: "arctic",
MODEL_ARCH.DEEPSEEK2: "deepseek2",
MODEL_ARCH.CHATGLM: "chatglm",
}
TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
@ -808,6 +810,18 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN_SHEXP,
MODEL_TENSOR.FFN_UP_SHEXP,
],
MODEL_ARCH.CHATGLM : [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_QKV,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
# TODO
}
@ -845,6 +859,9 @@ MODEL_TENSOR_SKIP: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_ROT_EMBD,
],
MODEL_ARCH.CHATGLM: [
MODEL_TENSOR.ROPE_FREQS,
],
}
#