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

llama: Support MiniCPM-1B (with & w/o longrope) (#10559)

Browse source
This commit is contained in:
JFLFY2255 2024-12-04 17:42:50 +08:00 • committed by GitHub
parent 2759916d86
commit 8d0cfd554a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 61 additions and 183 deletions
Download patch file Download diff file Expand all files Collapse all files

9
gguf-py/gguf/constants.py
View file

@ -896,6 +896,8 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ROPE_FACTORS_LONG,
MODEL_TENSOR.ROPE_FACTORS_SHORT,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
@ -1388,9 +1390,10 @@ class TokenType(IntEnum):
class RopeScalingType(Enum):
NONE = 'none'
LINEAR = 'linear'
YARN = 'yarn'
NONE = 'none'
LINEAR = 'linear'
YARN = 'yarn'
LONGROPE = 'longrope'
class PoolingType(IntEnum):