Skip to content

Optimize first token of C++ NPU by adding npu_dpu_groups #12443

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

Merged
merged 3 commits into from
Nov 26, 2024
Merged

Optimize first token of C++ NPU by adding npu_dpu_groups #12443

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
5de18e3
add npu_dpu_groups
rnwang04 Nov 26, 2024
b740e4d
add check for env
rnwang04 Nov 26, 2024
f99e91f
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
5 changes: 4 additions & 1 deletion python/llm/src/ipex_llm/transformers/npu_pipeline_model/common.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,8 @@
import numpy as np


def update_names_of_IR_and_export_blob(model, model_name, dir, compile_blob=True, keep_ir=True):
def update_names_of_IR_and_export_blob(model, model_name, dir, compile_blob=True, keep_ir=True,
npu_dpu_groups=None):
xml_path = os.path.join(dir, model_name + ".xml")
bin_path = os.path.join(dir, model_name + ".bin")
model.save(xml_path)
Expand All @@ -35,6 +36,8 @@ def update_names_of_IR_and_export_blob(model, model_name, dir, compile_blob=True
core.set_property("NPU", {"NPU_COMPILATION_MODE_PARAMS":
"compute-layers-with-higher-precision=Sqrt,Power,ReduceMean,Add"})
core.set_property("NPU", {"PERFORMANCE_HINT": "LATENCY"})
if npu_dpu_groups is not None:
core.set_property("NPU", {"NPU_DPU_GROUPS": str(npu_dpu_groups)})

model = core.read_model(xml_path)
inputs = model.inputs
Expand Down
5 changes: 4 additions & 1 deletion python/llm/src/ipex_llm/transformers/npu_pipeline_model/llama.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -272,11 +272,13 @@ def convert_llama_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
input_len = 1
decoder_name = f"decoder_layer_{layer_idx}"
keep_position_ids = True
npu_dpu_groups = None
else:
input_len = kv_len
decoder_name = "decoder_layer_prefill"
layernorm_const = False
keep_position_ids = False
npu_dpu_groups = 6

single_decoder = LowBitLlamaMultiDecoderlayer(
[1, input_len, num_heads * head_dim],
Expand All @@ -303,7 +305,8 @@ def convert_llama_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
rest_blob_path = update_names_of_IR_and_export_blob(single_decoder,
decoder_name,
temp_dir,
True, False)
True, False,
npu_dpu_groups=npu_dpu_groups)

if mode == "decode":
if hasattr(curr_layer.self_attn.rotary_emb, "cos_cached"):
Expand Down