tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-19 14:31:06 -04:00
Code Issues Packages Projects Releases Wiki Activity

llama : add gemma model (#5631)

Browse Source 
There are couple things in this architecture:

1. Shared input and output embedding parameters.
2. Key length and value length are not derived from `n_embd`.

More information about the models can be found at
https://ai.google.dev/gemma. GGUFs can be downloaded from
https://huggingface.co/google.
This commit is contained in:
postmasters
2024-02-21 05:08:22 -08:00
committed by GitHub
parent 88c46cbdac
commit 580111d42b
3 changed files with 186 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
gguf-py/gguf/constants.py
View File

@@ -111,6 +111,7 @@ class MODEL_ARCH(IntEnum):
ORION = auto()
INTERNLM2 = auto()
MINICPM = auto()
GEMMA = auto()
class MODEL_TENSOR(IntEnum):
@@ -167,6 +168,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.ORION: "orion",
MODEL_ARCH.INTERNLM2: "internlm2",
MODEL_ARCH.MINICPM: "minicpm",
MODEL_ARCH.GEMMA: "gemma",
}
TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
@@ -511,6 +513,19 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
],
MODEL_ARCH.GEMMA: [
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_TENSOR.FFN_NORM,
],
# TODO
}