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

llama : add phi3 support (#6852)

Browse source 
* add explicit phi3 support

* add explicit phi3 support

* remove unused code

* convert : add BOS token

* llama : match EOT token <|end|>

* llama : minor / style

* llama : tabs -> spaces

* convert : fix lint checks

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
liuwei-git 2024-04-24 15:00:37 +08:00 • committed by GitHub
parent 4e96a812b3
commit c8297c6af5
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 294 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

16
gguf-py/gguf/constants.py
View file

@ -124,6 +124,7 @@ class MODEL_ARCH(IntEnum):
QWEN2 = auto()
QWEN2MOE = auto()
PHI2 = auto()
PHI3 = auto()
PLAMO = auto()
CODESHELL = auto()
ORION = auto()
@ -200,6 +201,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.QWEN2: "qwen2",
MODEL_ARCH.QWEN2MOE: "qwen2moe",
MODEL_ARCH.PHI2: "phi2",
MODEL_ARCH.PHI3: "phi3",
MODEL_ARCH.PLAMO: "plamo",
MODEL_ARCH.CODESHELL: "codeshell",
MODEL_ARCH.ORION: "orion",
@ -550,6 +552,20 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.PHI3: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_QKV,
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
],
MODEL_ARCH.CODESHELL: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.POS_EMBD,

2
gguf-py/gguf/tensor_mapping.py
View file

@ -117,6 +117,7 @@ class TensorNameMap:
"h.{bid}.attn.c_attn", # gpt2
"transformer.h.{bid}.mixer.Wqkv", # phi2
"encoder.layers.{bid}.attn.Wqkv", # nomic-bert
"model.layers.{bid}.self_attn.qkv_proj" # phi3
),
# Attention query
@ -234,6 +235,7 @@ class TensorNameMap:
"h.{bid}.mlp.c_fc", # gpt2
"transformer.h.{bid}.mlp.fc1", # phi2
"model.layers.{bid}.mlp.fc1", # phi2
"model.layers.{bid}.mlp.gate_up_proj", # phi3
"model.layers.layers.{bid}.mlp.up_proj", # plamo
"model.layers.{bid}.feed_forward.w3", # internlm2
"encoder.layers.{bid}.mlp.fc11", # nomic-bert