Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[NPU] further fix of qwen2 int8 pipeline & C++ #12449

Merged
merged 2 commits into from
Nov 26, 2024
Merged

[NPU] further fix of qwen2 int8 pipeline & C++ #12449

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e86ec6a
fix
rnwang04 Nov 26, 2024
fc6e109
fix style
rnwang04 Nov 26, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion python/llm/src/ipex_llm/transformers/npu_model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -231,7 +231,7 @@ def optimize_npu_model(cls, *args, **kwargs):
from intel_npu_acceleration_library.compiler import create_npu_kernels

model = kwargs.pop("model")
qtype = kwargs.pop("qtype", "sym_int4")
qtype = kwargs.pop("qtype", "sym_int4_rtn")
mixed_precision = kwargs.pop("mixed_precision", False)
quantization_group_size = kwargs.pop("quantization_group_size", 0)
modules_to_not_convert = kwargs.pop("modules_to_not_convert", [])
Expand Down Expand Up @@ -280,6 +280,7 @@ def optimize_npu_model(cls, *args, **kwargs):
max_prompt_len=max_prompt_len,
transpose_value_cache=transpose_value_cache,
group_size=quantization_group_size,
qtype=qtype,
convert_model=convert_model,
save_directory=save_directory)
model.save_low_bit = types.MethodType(save_low_bit, model)
Expand Down
7 changes: 6 additions & 1 deletion python/llm/src/ipex_llm/transformers/npu_pipeline_model/convert_pipeline.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -193,13 +193,18 @@ def convert_llm(model: torch.nn.Module,
max_prompt_len: int,
transpose_value_cache: bool,
group_size: int,
qtype: str,
convert_model: bool=False,
save_directory: str=None):
# whether to set layernorm weight as const
layernorm_const = os.environ.get("IPEX_LLM_LAYERNORM_CONST", "1") == "1"
if group_size == 0:
n_splits_linear = 1
n_splits_down_proj = 2 if model.config.intermediate_size == 18944 else 1
if qtype == "sym_int8_rtn":
# do not split mlp down_proj for Qwen2-7B & sym_int8
n_splits_down_proj = 1
else:
n_splits_down_proj = 2 if model.config.intermediate_size == 18944 else 1
else:
n_splits_linear = model.config.hidden_size // group_size
n_splits_down_proj = model.config.intermediate_size // group_size
Expand Down