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

init (#8933)

Browse source
This commit is contained in:
Yina Chen 2023-09-11 14:30:55 +08:00 committed by GitHub
parent 4dce238867
commit df165ad165
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_SPR = is_spr()
TORCH_LINEAR_THRESHOLD = 96 TORCH_LINEAR_THRESHOLD = 96
SYM_INT4 = ggml_tensor_qtype["sym_int4"] SYM_INT4 = ggml_tensor_qtype["sym_int4"]
SYM_INT8 = ggml_tensor_qtype["sym_int8"] SYM_INT8 = ggml_tensor_qtype["sym_int8"]
NF4 = ggml_tensor_qtype["nf4"]
def ggml_convert_qtype(tensor: torch.Tensor, qtype: int, device=None): def ggml_convert_qtype(tensor: torch.Tensor, qtype: int, device=None):
@ -99,7 +100,7 @@ def ggml_q_format_convet_cpu2xpu(tensor: torch.Tensor, num_elem: int, qtype: int
src = ctypes.c_void_p(tensor.data.data_ptr()) src = ctypes.c_void_p(tensor.data.data_ptr())
if qtype in [SYM_INT4, SYM_INT8]: if qtype in [SYM_INT4, SYM_INT8, NF4]:
dst_tensor = torch.empty_like(tensor) dst_tensor = torch.empty_like(tensor)
elif qtype == ggml_tensor_qtype["sym_int5"]: elif qtype == ggml_tensor_qtype["sym_int5"]:
QK = ggml.ggml_qk_size(qtype) QK = ggml.ggml_qk_size(qtype)
@ -124,7 +125,7 @@ def ggml_q_format_convet_xpu2cpu(tensor: torch.Tensor, num_elem: int, qtype: int
src = ctypes.c_void_p(tensor.data.data_ptr()) src = ctypes.c_void_p(tensor.data.data_ptr())
if qtype in [SYM_INT4, SYM_INT8]: if qtype in [SYM_INT4, SYM_INT8, NF4]:
dst_tensor = torch.empty_like(tensor) dst_tensor = torch.empty_like(tensor)
elif qtype == ggml_tensor_qtype["sym_int5"]: elif qtype == ggml_tensor_qtype["sym_int5"]:
QK = ggml.ggml_qk_size(ggml_tensor_qtype["asym_int5"]) QK = ggml.ggml_qk_size(ggml_tensor_qtype["asym_int5"])