CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

convert : fix reflinks for stacked MoE tensors

Browse Source
This commit is contained in:
Francis Couture-Harpin
2025-09-02 15:22:01 -04:00
parent f7394cdaf4
commit 7724bf9e4f
4 changed files with 31 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
convert_hf_to_gguf.py
View File

@@ -462,7 +462,9 @@ class ModelBase:
# workaround BF16 not being supported by Numpy
if data_torch.dtype == torch.bfloat16:
data_torch = data_torch.view(torch.uint8)
# Need a contiguous last dimension otherwise byte view doesn't work
# (problem can be reproduced with DeepSeek-V2-Lite-Chat)
data_torch = data_torch.contiguous().view(torch.uint8)
# if data ends up empty, it means data_torch was a scalar tensor -> restore
if len(data_torch.shape) == 0: