Skip to content

Commit 725108b

Browse files
authored
add jina qkv tensors
1 parent 2b13162 commit 725108b

File tree

2 files changed

+3
-0
lines changed

2 files changed

+3
-0
lines changed

gguf-py/gguf/constants.py

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1036,6 +1036,7 @@ class MODEL_TENSOR(IntEnum):
10361036
MODEL_TENSOR.POS_EMBD,
10371037
MODEL_TENSOR.OUTPUT_NORM,
10381038
MODEL_TENSOR.ATTN_OUT_NORM,
1039+
MODEL_TENSOR.ATTN_QKV,
10391040
MODEL_TENSOR.ATTN_Q,
10401041
MODEL_TENSOR.ATTN_K,
10411042
MODEL_TENSOR.ATTN_V,

gguf-py/gguf/tensor_mapping.py

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -157,6 +157,7 @@ class TensorNameMap:
157157
"h.{bid}.attn.c_attn", # gpt2
158158
"transformer.h.{bid}.mixer.Wqkv", # phi2
159159
"encoder.layers.{bid}.attn.Wqkv", # nomic-bert
160+
"encoder.layers.{bid}.mixer.Wqkv", # jina
160161
"model.layers.{bid}.self_attn.qkv_proj", # phi3
161162
"encoder.layers.{bid}.self_attention.query_key_value", # chatglm
162163
"transformer.layers.{bid}.attn.qkv_proj", # openelm
@@ -224,6 +225,7 @@ class TensorNameMap:
224225
"model.layers.layers.{bid}.self_attn.o_proj", # plamo
225226
"model.layers.{bid}.attention.wo", # internlm2
226227
"encoder.layers.{bid}.attn.out_proj", # nomic-bert
228+
"encoder.layers.{bid}.mixer.out_proj", # jina
227229
"transformer.decoder_layer.{bid}.multi_head_attention.linear", # Grok
228230
"transformer.blocks.{bid}.norm_attn_norm.attn.out_proj", # dbrx
229231
"encoder.layers.{bid}.self_attention.dense", # chatglm

0 commit comments

Comments
 (0)