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

convert : fix grok tensor names

Browse source
This commit is contained in:
Georgi Gerganov 2024-04-02 21:35:13 +03:00
parent f421b32d5a
commit c704c778f6
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
1 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
convert-hf-to-gguf.py
View file

@ -1351,7 +1351,7 @@ class GrokModel(Model):
for wid in ["linear", "linear_1", "linear_v"]: for wid in ["linear", "linear_1", "linear_v"]:
full = True full = True
for xid in range(n_experts): for xid in range(n_experts):
ename = f"transformer.decoder_layer.{bid}.moe.{xid}.{wid}" ename = f"transformer.decoder_layer.{bid}.moe.{xid}.{wid}.weight"
if ename not in experts: if ename not in experts:
full = False full = False
break break
@ -1360,7 +1360,7 @@ class GrokModel(Model):
datas = [] datas = []
for xid in range(n_experts): for xid in range(n_experts):
ename = f"transformer.decoder_layer.{bid}.moe.{xid}.{wid}" ename = f"transformer.decoder_layer.{bid}.moe.{xid}.{wid}.weight"
datas.append(experts[ename]) datas.append(experts[ename])
del experts[ename] del experts[ename]