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

Implement the OLMo architecture (#6741)

Browse source 
* implement olmo architecture

* remove unused variable

* remove unused moe branch

* remove check for weight

* remove superfluous moe, bias and rope tensors

* clarified comment

* fix clamp_kqv setting

* remove obsolete parameter name filter
This commit is contained in:
nopperl 2024-04-19 09:35:54 +00:00 • committed by GitHub
parent 8b1b1f4982
commit 9958c81b79
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 271 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

13
gguf-py/gguf/constants.py
View file

@ -135,6 +135,7 @@ class MODEL_ARCH(IntEnum):
XVERSE = auto()
COMMAND_R = auto()
DBRX = auto()
OLMO = auto()
class MODEL_TENSOR(IntEnum):
@ -210,6 +211,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.XVERSE: "xverse",
MODEL_ARCH.COMMAND_R: "command-r",
MODEL_ARCH.DBRX: "dbrx",
MODEL_ARCH.OLMO: "olmo",
}
TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
@ -695,6 +697,17 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
],
MODEL_ARCH.OLMO: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT,
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,
],
# TODO
}