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

add swin norm param

Browse source
This commit is contained in:
nopperl 2024-07-15 21:19:27 +02:00
parent 0453f7d114
commit 654b1b35b3
3 changed files with 8 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

4
convert_hf_to_gguf.py
View file

@ -3427,6 +3427,10 @@ class ChatGLMModel(Model):
class ChameleonModel(Model): class ChameleonModel(Model):
model_arch = gguf.MODEL_ARCH.CHAMELEON model_arch = gguf.MODEL_ARCH.CHAMELEON
def set_gguf_parameters(self):
super().set_gguf_parameters()
self.gguf_writer.add_swin_norm(self.hparams.get("swin_norm", False))
def set_vocab(self): def set_vocab(self):
self._set_vocab_gpt2() self._set_vocab_gpt2()

1
gguf-py/gguf/constants.py
View file

@ -52,6 +52,7 @@ class Keys:
DECODER_START_TOKEN_ID = "{arch}.decoder_start_token_id" DECODER_START_TOKEN_ID = "{arch}.decoder_start_token_id"
ATTN_LOGIT_SOFTCAPPING = "{arch}.attn_logit_softcapping" ATTN_LOGIT_SOFTCAPPING = "{arch}.attn_logit_softcapping"
FINAL_LOGIT_SOFTCAPPING = "{arch}.final_logit_softcapping" FINAL_LOGIT_SOFTCAPPING = "{arch}.final_logit_softcapping"
SWIN_NORM = "{arch}.swin_norm"
class Attention: class Attention:
HEAD_COUNT = "{arch}.attention.head_count" HEAD_COUNT = "{arch}.attention.head_count"

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

@ -701,6 +701,9 @@ class GGUFWriter:
def add_eot_token_id(self, id: int) -> None: def add_eot_token_id(self, id: int) -> None:
self.add_uint32(Keys.Tokenizer.EOT_ID, id) self.add_uint32(Keys.Tokenizer.EOT_ID, id)
def add_swin_norm(self, value: bool) -> None:
self.add_bool(Keys.LLM.SWIN_NORM.format(arch=self.arch), value)
def _pack(self, fmt: str, value: Any, skip_pack_prefix: bool = False) -> bytes: def _pack(self, fmt: str, value: Any, skip_pack_prefix: bool = False) -> bytes:
pack_prefix = '' pack_prefix = ''
if not skip_pack_prefix: if not skip_pack_prefix: