ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

enable new q4_1 (#9479)

Browse source
This commit is contained in:
Xin Qiu 2023-11-17 14:58:57 +08:00 committed by GitHub
parent 3dac21ac7b
commit 50b01058f1
1 changed files with 3 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
python/llm/src/bigdl/llm/transformers/low_bit_linear.py
View file

@ -61,6 +61,7 @@ IS_SERVER = is_server()
IS_SPR = is_spr()
TORCH_LINEAR_THRESHOLD = 96
SYM_INT4 = ggml_tensor_qtype["sym_int4"]
ASYM_INT4 = ggml_tensor_qtype["asym_int4"]
SYM_INT8 = ggml_tensor_qtype["sym_int8"]
NF4 = ggml_tensor_qtype["nf4"]
NF3 = ggml_tensor_qtype["nf3"]
@ -111,7 +112,7 @@ def ggml_q_format_convet_cpu2xpu(tensor: torch.Tensor, num_elem: int, qtype: int
src = ctypes.c_void_p(tensor.data.data_ptr())
if qtype in [SYM_INT4, SYM_INT8, NF4, NF3, FP4, FP8]:
if qtype in [SYM_INT4, ASYM_INT4, SYM_INT8, NF4, NF3, FP4, FP8]:
dst_tensor = torch.empty_like(tensor)
elif qtype == ggml_tensor_qtype["sym_int5"]:
QK = ggml.ggml_qk_size(qtype)
@ -136,7 +137,7 @@ def ggml_q_format_convet_xpu2cpu(tensor: torch.Tensor, num_elem: int, qtype: int
src = ctypes.c_void_p(tensor.data.data_ptr())
if qtype in [SYM_INT4, SYM_INT8, NF4, NF3, FP4, FP8]:
if qtype in [SYM_INT4, ASYM_INT4, SYM_INT8, NF4, NF3, FP4, FP8]:
dst_tensor = torch.empty_like(tensor)
elif qtype == ggml_tensor_qtype["sym_int5"]:
QK = ggml.ggml_qk_size(ggml_tensor_qtype["asym_int5"])