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

Add attention and final logit softcapping.

Browse source
This commit is contained in:
Andrei Betlen 2024-06-28 15:42:19 -04:00
parent 8748d8ac6f
commit 4d3f17b4ac
4 changed files with 38 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

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

@ -50,6 +50,8 @@ class Keys:
POOLING_TYPE = "{arch}.pooling_type"
LOGIT_SCALE = "{arch}.logit_scale"
DECODER_START_TOKEN_ID = "{arch}.decoder_start_token_id"
ATTN_LOGIT_SOFTCAPPING = "{arch}.attn_logit_softcapping"
FINAL_LOGIT_SOFTCAPPING = "{arch}.final_logit_softcapping"
class Attention:
HEAD_COUNT = "{arch}.attention.head_count"

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

@ -627,6 +627,9 @@ class GGUFWriter:
def add_mask_token_id(self, id: int) -> None:
self.add_uint32(Keys.Tokenizer.MASK_ID, id)
def add_eot_token_id(self, id: int) -> None:
self.add_uint32(Keys.Tokenizer.EOT_ID, id)
def add_add_bos_token(self, value: bool) -> None:
self.add_bool(Keys.Tokenizer.ADD_BOS, value)