ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-05-01 03:41:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

bitnet: python + llama

Browse Source
This commit is contained in:
Iwan Kawrakow
2024-06-16 14:25:12 +03:00
parent 58756ef03f
commit f20b28558b
5 changed files with 779 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -149,6 +149,7 @@ class MODEL_ARCH(IntEnum):
OLMO = auto()
ARCTIC = auto()
DEEPSEEK2 = auto()
BITNET = auto()
class MODEL_TENSOR(IntEnum):
@@ -200,6 +201,8 @@ class MODEL_TENSOR(IntEnum):
ATTN_KV_B = auto()
ATTN_Q_A_NORM = auto()
ATTN_KV_A_NORM = auto()
FFN_SUB_NORM = auto()
ATTN_SUB_NORM = auto()
MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
@@ -237,6 +240,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.OLMO: "olmo",
MODEL_ARCH.ARCTIC: "arctic",
MODEL_ARCH.DEEPSEEK2: "deepseek2",
MODEL_ARCH.BITNET: "bitnet",
}
TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
@@ -288,6 +292,8 @@ TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
MODEL_TENSOR.ATTN_KV_B: "blk.{bid}.attn_kv_b",
MODEL_TENSOR.ATTN_Q_A_NORM: "blk.{bid}.attn_q_a_norm",
MODEL_TENSOR.ATTN_KV_A_NORM: "blk.{bid}.attn_kv_a_norm",
MODEL_TENSOR.ATTN_SUB_NORM: "blk.{bid}.attn_sub_norm",
MODEL_TENSOR.FFN_SUB_NORM: "blk.{bid}.ffn_sub_norm",
}
MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
@@ -808,6 +814,24 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_TENSOR.FFN_DOWN_SHEXP,
MODEL_TENSOR.FFN_UP_SHEXP,
],
MODEL_ARCH.BITNET: [
MODEL_TENSOR.ATTN_Q,
MODEL_TENSOR.ATTN_K,
MODEL_TENSOR.ATTN_V,
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.ROPE_FREQS,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_QKV,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.ATTN_ROT_EMBD,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_GATE,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
MODEL_TENSOR.ATTN_SUB_NORM,
MODEL_TENSOR.FFN_SUB_NORM,
],
# TODO
}

8
gguf-py/gguf/tensor_mapping.py
View File

@@ -413,6 +413,14 @@ class TensorNameMap:
MODEL_TENSOR.ATTN_KV_A_NORM: (
"model.layers.{bid}.self_attn.kv_a_layernorm", # deepseek2
),
MODEL_TENSOR.ATTN_SUB_NORM: (
"model.layers.{bid}.self_attn.inner_attn_ln", # bitnet
),
MODEL_TENSOR.FFN_SUB_NORM: (
"model.layers.{bid}.mlp.ffn_layernorm", # bitnet
),
}
# architecture-specific block mappings