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

convert-persimmon : typing fixup

Browse source
This commit is contained in:
Jared Van Bortel 2024-03-27 12:36:08 -04:00
parent 9803bb7206
commit 03f0c2e8ce
1 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

6
convert-persimmon-to-gguf.py
View file

@ -106,12 +106,12 @@ def main():
tensor_map = gguf.get_tensor_name_map(arch, block_count) tensor_map = gguf.get_tensor_name_map(arch, block_count)
print(tensor_map) print(tensor_map)
for name in tensors.keys(): for name in tensors.keys():
data = tensors[name] data_torch = tensors[name]
if name.endswith(".self_attention.rotary_emb.inv_freq"): if name.endswith(".self_attention.rotary_emb.inv_freq"):
continue continue
old_dtype = data.dtype old_dtype = data_torch.dtype
# TODO: FP16 conversion produces garbage outputs. (Q8_0 does not, so..?) # TODO: FP16 conversion produces garbage outputs. (Q8_0 does not, so..?)
data = data.to(torch.float32).squeeze().numpy() data = data_torch.to(torch.float32).squeeze().numpy()
new_name = tensor_map.get_name(name, try_suffixes = (".weight", ".bias")) new_name = tensor_map.get_name(name, try_suffixes = (".weight", ".bias"))
if new_name is None: if new_name is None:
print("Can not map tensor '" + name + "'") print("Can not map tensor '" + name + "'")