Skip to content

Commit

Permalink
[NPU L0] update layernorm & code refactor (#12287)
Browse files Browse the repository at this point in the history 
* update layernorm & code refactor

* fix style

* add common utils

* change to Pool()

* remove print
  • Loading branch information
@rnwang04
rnwang04 authored Oct 29, 2024
1 parent 4467645 commit 821b003
Show file tree
Hide file tree
Showing 3 changed files with 234 additions and 196 deletions.
54 changes: 54 additions & 0 deletions python/llm/src/ipex_llm/transformers/npu_pipeline_model/common.py
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,54 @@
#
# Copyright 2016 The BigDL Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#


from openvino.runtime import Core, serialize
import os


def update_names_of_IR_and_export_blob(model, model_name, dir):
xml_path = os.path.join(dir, model_name + ".xml")
model.save(xml_path)
new_ir_path = os.path.join(dir, model_name + "_new.xml")
blob_path = os.path.join(dir, model_name + ".blob")

core = Core()
core.set_property("NPU", {"NPU_COMPILATION_MODE_PARAMS":
"compute-layers-with-higher-precision=Sqrt,Power,ReduceMean,Add"})
core.set_property("NPU", {"PERFORMANCE_HINT": "LATENCY"})
model = core.read_model(xml_path)
inputs = model.inputs
for idx, input in enumerate(inputs):
if len(input.names) == 0:
model.inputs[idx].set_names({f"input_{idx}"})
outputs = model.outputs
for idx, input in enumerate(outputs):
if len(input.names) == 0:
model.outputs[idx].set_names({f"output_{idx}"})
# rewrite this model to a new IR path
if new_ir_path is not None:
serialize(model, new_ir_path)

if blob_path is not None:
compiledModel = core.compile_model(model, device_name="NPU")
model_stream = compiledModel.export_model()
with open(blob_path, 'wb') as f:
f.write(model_stream)

os.remove(xml_path)
os.remove(new_ir_path)

return blob_path
218 changes: 22 additions & 196 deletions python/llm/src/ipex_llm/transformers/npu_pipeline_model/convert_pipeline.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,6 @@
#


from openvino.runtime import Core, serialize
import os
import torch
from ipex_llm.utils.common import invalidInputError
Expand All @@ -31,6 +30,7 @@
import tempfile
import numpy as np
from ipex_llm.transformers.npu_models.lm_head import SlicedLMHead
from multiprocessing import Pool


def generate(
Expand Down Expand Up @@ -188,41 +188,6 @@ def generate(
return output


def update_names_of_IR_and_export_blob(model, model_name, dir):
xml_path = os.path.join(dir, model_name + ".xml")
model.save(xml_path)
new_ir_path = os.path.join(dir, model_name + "_new.xml")
blob_path = os.path.join(dir, model_name + ".blob")

core = Core()
core.set_property("NPU", {"NPU_COMPILATION_MODE_PARAMS":
"compute-layers-with-higher-precision=Sqrt,Power,ReduceMean,Add"})
core.set_property("NPU", {"PERFORMANCE_HINT": "LATENCY"})
model = core.read_model(xml_path)
inputs = model.inputs
for idx, input in enumerate(inputs):
if len(input.names) == 0:
model.inputs[idx].set_names({f"input_{idx}"})
outputs = model.outputs
for idx, input in enumerate(outputs):
if len(input.names) == 0:
model.outputs[idx].set_names({f"output_{idx}"})
# rewrite this model to a new IR path
if new_ir_path is not None:
serialize(model, new_ir_path)

if blob_path is not None:
compiledModel = core.compile_model(model, device_name="NPU")
model_stream = compiledModel.export_model()
with open(blob_path, 'wb') as f:
f.write(model_stream)

os.remove(xml_path)
os.remove(new_ir_path)

return blob_path


def convert_llm(model: torch.nn.Module,
kv_len: int,
max_prompt_len: int,
Expand All @@ -235,180 +200,41 @@ def convert_llm(model: torch.nn.Module,
n_splits_linear = model.config.hidden_size // group_size
n_splits_down_proj = model.config.intermediate_size // group_size
if model.config.model_type == "llama":
from ipex_llm.transformers.npu_models.convert_mp import convert_llama
convert_llama(model,
max_output_len=kv_len,
max_prompt_len=max_prompt_len,
decoder=False,
transpose_value_cache=transpose_value_cache)
from .llama import LowBitLlamaLMHead, LlamaEmbedding
with tempfile.TemporaryDirectory() as temp_dir:
# generate lm_head blob
weight_dir = os.path.join(temp_dir, "model_weights")
os.mkdir(weight_dir)
num_heads = model.model.layers[0].self_attn.num_heads
num_key_value_heads = model.model.layers[0].self_attn.num_key_value_heads
head_dim = model.model.layers[0].self_attn.head_dim
intermediate_size = model.config.intermediate_size
layer_num = len(model.model.layers)
rms_norm_eps = model.config.rms_norm_eps
vocab_size = model.config.vocab_size
model_norm = model.model.norm
lm_head = model.lm_head
if n_splits_linear == 1:
weights = [(lm_head.weight, lm_head.scale)]
else:
lm_heads = lm_head.lm_heads
lm_head_weights = []
scales = []
for i in range(n_splits_linear):
lm_head_weights.append(lm_heads[i].weight)
scales.append(lm_heads[i].scale)
weights = [(torch.stack(lm_head_weights, axis=0),
torch.stack(scales, axis=0))]
if isinstance(weights[0], tuple):
np_dtype = np.int8 if weights[0][0].dtype == torch.int8 else np.uint8
else: # FP16 Linear
np_dtype = np.float16

new_lm_head = LowBitLlamaLMHead(
[1, 1, num_heads * head_dim],
num_heads=num_heads,
num_key_value_heads=num_key_value_heads,
max_seq_len=kv_len,
rms_norm_eps=rms_norm_eps,
mode="decode",
transpose_value=False,
dtype=np_dtype,
model_norm_weight=model_norm.weight.to(torch.float16),
vocab_size=vocab_size,
n_splits=n_splits_linear
)
last_blob_path = update_names_of_IR_and_export_blob(new_lm_head, "lm_head", temp_dir)

# save weights bins files
if n_splits_linear == 1:
weight_numpy = [
lm_head.weight.data.numpy(), lm_head.scale.data.numpy(),
]
else:
weight_numpy = [v.numpy() for v in weights[0]]
from .llama import convert_llama_layer, convert_lm_head_and_embedding
first_blob_path, last_blob_path = convert_lm_head_and_embedding(model, n_splits_linear,
temp_dir, weight_dir)

for idx, weight in enumerate(weight_numpy):
bin_file = os.path.join(weight_dir, f"model_lm_head_input_{1+idx}.bin")
weight.tofile(bin_file)

embedding_layer = model.model.embed_tokens
new_embedding = LlamaEmbedding(
vocab_size=model.config.vocab_size,
embedding_dim=model.config.hidden_size,
embedding_weight=embedding_layer.weight.to(torch.float16).detach().numpy(),
padding_idx=model.config.pad_token_id,
dtype=np.float16,
)
first_blob_path = update_names_of_IR_and_export_blob(new_embedding, "embedding",
temp_dir)

# generate decoder layer blob
from ipex_llm.transformers.npu_models.llama_mp import LowBitLlamaMultiDecoderlayer
param_list = []
for layer_idx in range(0, layer_num):
curr_layer = model.model.layers[layer_idx]
attn_layer = curr_layer.self_attn
mlp_layer = curr_layer.mlp

weights = []
if n_splits_linear == 1:
for q, k, v, o, g, u in zip(attn_layer.q_proj_dq_list,
attn_layer.k_proj_dq_list,
attn_layer.v_proj_dq_list,
attn_layer.o_proj_dq_list,
mlp_layer.gate_proj_dq_list,
mlp_layer.up_proj_dq_list):
weights.append((q.weight, q.scale))
weights.append((k.weight, k.scale))
weights.append((v.weight, v.scale))
weights.append((o.weight, o.scale))
weights.append((g.weight, g.scale))
weights.append((u.weight, u.scale))
else:
for layer_list in [attn_layer.q_proj_dq_list, attn_layer.k_proj_dq_list,
attn_layer.v_proj_dq_list, attn_layer.o_proj_dq_list,
mlp_layer.gate_proj_dq_list, mlp_layer.up_proj_dq_list]:
l_weights = []
scales = []
for l in layer_list:
l_weights.append(l.weight)
scales.append(l.scale)
weights.append((torch.stack(l_weights, axis=0),
torch.stack(scales, axis=0)))

if n_splits_down_proj == 1:
for l in mlp_layer.down_proj_dq_list:
weights.append((l.weight, l.scale))
else:
l_weights = []
scales = []
for l in mlp_layer.down_proj_dq_list:
l_weights.append(l.weight)
scales.append(l.scale)
weights.append((torch.stack(l_weights, axis=0), torch.stack(scales, axis=0)))

cached_cos = curr_layer.self_attn.rotary_emb.cos_cached.to(torch.float16)
cached_sin = curr_layer.self_attn.rotary_emb.sin_cached.to(torch.float16)
layer_norm_0 = curr_layer.input_layernorm.weight.to(torch.float16)
layer_norm_1 = curr_layer.post_attention_layernorm.weight.to(torch.float16)

if isinstance(weights[0], tuple):
np_dtype = np.int8 if weights[0][0].dtype == torch.int8 else np.uint8
else: # FP16 Linear
np_dtype = np.float16

if layer_idx == 0:
single_decoder = LowBitLlamaMultiDecoderlayer(
[1, 1, num_heads * head_dim],
input_layernorm_weights=None,
post_attn_layernorm_weights=None,
cached_cos=cached_cos,
cached_sin=cached_sin,
num_heads=num_heads,
num_key_value_heads=num_key_value_heads,
num_layers=1,
max_seq_len=kv_len,
rms_norm_eps=rms_norm_eps,
intermediate_size=intermediate_size,
mode="decode",
transpose_value=transpose_value_cache,
dtype=np_dtype,
n_splits_linear=n_splits_linear,
n_splits_down_proj=n_splits_down_proj,
group_size=group_size
)
rest_blob_path = update_names_of_IR_and_export_blob(single_decoder,
"decoder_layer",
temp_dir)

input_lm_bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_3.bin")
post_lm_bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_4.bin")
layer_norm_0.data.numpy().tofile(input_lm_bin_file)
layer_norm_1.data.numpy().tofile(post_lm_bin_file)

for idx, (weight, scale) in enumerate(weights):
bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_{7+idx*2}.bin")
weight.numpy().tofile(bin_file)
bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_{7+idx*2+1}.bin")
scale.numpy().tofile(bin_file)
param_list.append((model, layer_idx, n_splits_linear, n_splits_down_proj,
temp_dir, weight_dir, transpose_value_cache, kv_len, group_size))
with Pool() as pool:
result = pool.starmap(convert_llama_layer, param_list)

# Prefill Runner
from ipex_llm.transformers.npu_models.convert_mp import convert_llama
convert_llama(model,
max_output_len=kv_len,
max_prompt_len=max_prompt_len,
decoder=False,
transpose_value_cache=transpose_value_cache)

# patch attrs for generate
model.kv_len = kv_len
model.num_head = num_heads
model.head_dim = head_dim
model.num_head = model.model.layers[0].self_attn.num_heads
model.head_dim = model.model.layers[0].self_attn.head_dim
model.num_layers = layer_num
model.transpose_value_cache = transpose_value_cache

try:
res = InitLLMPipeline(kv_len, num_heads, head_dim, layer_num,
res = InitLLMPipeline(kv_len, model.num_head, model.head_dim, layer_num,
model.vocab_size, weight_dir, "model",
first_blob_path, last_blob_path, rest_blob_path)
first_blob_path, last_blob_path,
os.path.join(temp_dir, "decoder_layer"))
except:
invalidInputError(False,
"False to InitLLMPipeline.")
Expand Down
Loading

0 comments on commit 821b003

Please sign in to comment.