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

llama : add grok-1 support (#6204)

Browse source 
* Add support for Grok model architecture

* Revert convert-hf-to-gguf to default options

* Fixed f_norm_rms_eps bug

* Fix whitespaces

* llama : fix grok rope type

* llama : minor

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Julius Arkenberg 2024-03-23 17:41:53 +01:00 • committed by GitHub
parent 21cad01b6e
commit 476b0251b2
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 384 additions and 20 deletions
Download patch file Download diff file Expand all files Collapse all files

24
gguf-py/gguf/constants.py
View file

@ -100,6 +100,7 @@ class MODEL_ARCH(IntEnum):
LLAMA = auto()
FALCON = auto()
BAICHUAN = auto()
GROK = auto()
GPT2 = auto()
GPTJ = auto()
GPTNEOX = auto()
@ -167,6 +168,7 @@ 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",
@ -251,6 +253,28 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
],
MODEL_ARCH.GROK: [
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.ATTN_OUT_NORM,
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.LAYER_OUT_NORM,
],
MODEL_ARCH.GPTNEOX: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,