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

llama : add phixtral support (wip)

Browse source
This commit is contained in:
Georgi Gerganov 2024-01-13 14:19:13 +02:00
parent 15ebe59210
commit 9998ecd191
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
4 changed files with 107 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

3
gguf-py/gguf/constants.py
View file

@ -393,9 +393,12 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.FFN_GATE_INP,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
MODEL_TENSOR.FFN_DOWN_EXP,
MODEL_TENSOR.FFN_UP_EXP,
]
# TODO
}

3
gguf-py/gguf/tensor_mapping.py
View file

@ -173,6 +173,7 @@ class TensorNameMap:
MODEL_TENSOR.FFN_GATE_INP: (
"layers.{bid}.feed_forward.gate", # mixtral
"model.layers.{bid}.block_sparse_moe.gate", # mixtral
"transformer.h.{bid}.moe.gate", # phixtral
),
# Feed-forward up
@ -198,6 +199,7 @@ class TensorNameMap:
MODEL_TENSOR.FFN_UP_EXP: (
"layers.{bid}.feed_forward.experts.{xid}.w3", # mixtral
"model.layers.{bid}.block_sparse_moe.experts.{xid}.w3", # mixtral
"transformer.h.{bid}.moe.mlp.{xid}.fc1", # phixtral
),
# AWQ-activation gate
@ -240,6 +242,7 @@ class TensorNameMap:
MODEL_TENSOR.FFN_DOWN_EXP: (
"layers.{bid}.feed_forward.experts.{xid}.w2", # mixtral
"model.layers.{bid}.block_sparse_moe.experts.{xid}.w2", # mixtral
"transformer.h.{bid}.moe.mlp.{xid}.fc2", # phixtral
),
MODEL_TENSOR.ATTN_Q_NORM: (