Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-25 22:08:46 +01:00
Code Issues Packages Projects Releases Wiki Activity

convert : use f32 outtype for bf16 tensors (#6106)

Browse Source 
The old behaviour is to use f16, but bf16 to f16 is not a lossless conversion.
Change the outtype to f32 to default to a lossless conversion.
This commit is contained in:
Romain D 2024-03-18 09:04:41 +01:00 committed by GitHub
parent d01b3c4c32
commit 3a6efdd03c
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
convert.py
View File

@ -1167,9 +1167,9 @@ class OutputFile:
def pick_output_type(model: LazyModel, output_type_str: str | None) -> GGMLFileType: def pick_output_type(model: LazyModel, output_type_str: str | None) -> GGMLFileType:
wq_type = model[gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.ATTN_Q].format(bid=0) + ".weight"].data_type wq_type = model[gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.ATTN_Q].format(bid=0) + ".weight"].data_type
if output_type_str == "f32" or (output_type_str is None and wq_type == DT_F32): if output_type_str == "f32" or (output_type_str is None and wq_type in (DT_F32, DT_BF16)):
return GGMLFileType.AllF32 return GGMLFileType.AllF32
if output_type_str == "f16" or (output_type_str is None and wq_type in (DT_F16, DT_BF16)): if output_type_str == "f16" or (output_type_str is None and wq_type == DT_F16):
return GGMLFileType.MostlyF16 return GGMLFileType.MostlyF16
if output_type_str == "q8_0": if output_type_str == "q8_0":
return GGMLFileType.MostlyQ8_0 return GGMLFileType.MostlyQ8_0