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

mpt : do not duplicate token_embd.weight on disk (#5670)

Browse source
This commit is contained in:
Jared Van Bortel 2024-02-22 17:05:23 -05:00 • committed by GitHub
parent 96633eeca1
commit 15499eb942
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 4 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

5
convert-hf-to-gguf.py
View file

@ -622,11 +622,6 @@ class MPTModel(Model):
self.gguf_writer.add_tensor(new_name, data)
# note: MPT output is tied to (same as) wte in original model;
# for easier implementation in llama.cpp it's duplicated in GGUF, though :/
if new_name == "token_embd.weight":
self.gguf_writer.add_tensor("output.weight", data)
class OrionModel(Model):
def set_vocab(self):