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

Support baichuan2 for level0 pipeline (#12289)

Browse source
This commit is contained in:
binbin Deng 2024-10-29 19:24:16 +08:00 committed by GitHub
parent 546f455e8e
commit 3feb58d1e4
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
9 changed files with 472 additions and 174 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
python/llm/example/NPU/HF-Transformers-AutoModels/LLM/Pipeline-Models/README.md
View file

@ -8,6 +8,7 @@ In this directory, you will find examples on how to directly run HuggingFace `tr
|------------|----------------------------------------------------------------|
| Llama2 | [meta-llama/Llama-2-7b-chat-hf](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf) |
| Llama3 | [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct) |
| Baichuan2 | [baichuan-inc/Baichuan2-7B-Chat](https://huggingface.co/baichuan-inc/Baichuan-7B-Chat) |
## 0. Requirements
To run these examples with IPEX-LLM on Intel NPUs, make sure to install the newest driver version of Intel NPU.
@ -43,6 +44,9 @@ python llama2.py
:: to run Meta-Llama-3-8B-Instruct
python llama3.py
:: to run Baichuan2-7B-Chat
python baichuan2.py
```
Arguments info:

99
python/llm/example/NPU/HF-Transformers-AutoModels/LLM/Pipeline-Models/baichuan2.py Normal file
View file

@ -0,0 +1,99 @@
#
# 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.
#
import torch
import time
import argparse
from ipex_llm.transformers.npu_model import AutoModelForCausalLM
from transformers import AutoTokenizer
from transformers.utils import logging
logger = logging.get_logger(__name__)
def get_prompt(message: str, chat_history: list[tuple[str, str]],
system_prompt: str) -> str:
texts = [f'<s>[INST] <<SYS>>\n{system_prompt}\n<</SYS>>\n\n']
# The first user input is _not_ stripped
do_strip = False
for user_input, response in chat_history:
user_input = user_input.strip() if do_strip else user_input
do_strip = True
texts.append(f'{user_input} [/INST] {response.strip()} </s><s>[INST] ')
message = message.strip() if do_strip else message
texts.append(f'{message} [/INST]')
return ''.join(texts)
if __name__ == "__main__":
parser = argparse.ArgumentParser(
description="Predict Tokens using `generate()` API for npu model"
)
parser.add_argument(
"--repo-id-or-model-path",
type=str,
default="baichuan-inc/Baichuan2-7B-Chat",
help="The huggingface repo id for the Baichuan2 model to be downloaded"
", or the path to the huggingface checkpoint folder",
)
parser.add_argument('--prompt', type=str, default="What is AI?",
help='Prompt to infer')
parser.add_argument("--n-predict", type=int, default=32, help="Max tokens to predict")
parser.add_argument("--max-context-len", type=int, default=1024)
parser.add_argument("--max-prompt-len", type=int, default=960)
parser.add_argument("--disable-transpose-value-cache", action="store_true", default=False)
args = parser.parse_args()
model_path = args.repo_id_or_model_path
model = AutoModelForCausalLM.from_pretrained(model_path,
optimize_model=True,
pipeline=True,
max_context_len=args.max_context_len,
max_prompt_len=args.max_prompt_len,
torch_dtype=torch.float16,
attn_implementation="eager",
transpose_value_cache=not args.disable_transpose_value_cache,
trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
DEFAULT_SYSTEM_PROMPT = """\
"""
print("-" * 80)
print("done")
with torch.inference_mode():
print("finish to load")
for i in range(5):
prompt = get_prompt(args.prompt, [], system_prompt=DEFAULT_SYSTEM_PROMPT)
_input_ids = tokenizer.encode(prompt, return_tensors="pt")
print("input length:", len(_input_ids[0]))
st = time.time()
output = model.generate(
_input_ids, max_new_tokens=args.n_predict, do_print=True
)
end = time.time()
print(f"Inference time: {end-st} s")
input_str = tokenizer.decode(_input_ids[0], skip_special_tokens=False)
print("-" * 20, "Input", "-" * 20)
print(input_str)
output_str = tokenizer.decode(output[0], skip_special_tokens=False)
print("-" * 20, "Output", "-" * 20)
print(output_str)
print("-" * 80)
print("done")
print("success shut down")

66
python/llm/src/ipex_llm/transformers/npu_models/baichuan_mp.py
View file

@ -112,32 +112,14 @@ class LowBitBaichuanMultiDecoderlayer(LLMBaseNNFactory):
# Self Attention
if mode == "decode":
attention_mask = self.create_input_op((self.batch_size, 1, 1, self.max_seq_len + 1))
attention_mask = self.create_input_op((self.batch_size, 1, 1, self.max_seq_len + 1),
dtype=np.int64)
else:
attention_mask = self.create_input_op((self.batch_size, 1, self.seq_len, self.seq_len))
attention_mask = self.create_input_op((self.batch_size, 1, self.seq_len, self.seq_len),
dtype=np.int64)
position_ids = self.create_input_op((self.batch_size, self.seq_len))
position_ids = self.create_input_op((self.batch_size, self.seq_len), dtype=np.int64)
# self.num_key_value_heads = num_key_value_heads
past_keys = []
past_values = []
if mode == "decode":
for i in range(num_layers):
past_key = self.create_cache_op(
(self.batch_size, self.num_heads, self.max_seq_len, self.head_dim)
)
if transpose_value:
past_value = self.create_cache_op(
(self.batch_size, self.num_heads, self.head_dim, self.max_seq_len)
)
else:
past_value = self.create_cache_op(
(self.batch_size, self.num_heads, self.max_seq_len, self.head_dim)
)
past_keys.append(past_key)
past_values.append(past_value)
else:
past_keys = [None] * num_layers
past_values = [None] * num_layers
if input_layernorm_weights is None:
input_layernorm_weights = []
@ -163,6 +145,27 @@ class LowBitBaichuanMultiDecoderlayer(LLMBaseNNFactory):
input_layernorm_weights = [self.constant(w) for w in input_layernorm_weights]
post_attn_layernorm_weights = [self.constant(w) for w in post_attn_layernorm_weights]
past_keys = []
past_values = []
if mode == "decode":
for i in range(num_layers):
past_key = self.create_cache_op(
(self.batch_size, self.num_heads, self.max_seq_len, self.head_dim)
)
if transpose_value:
past_value = self.create_cache_op(
(self.batch_size, self.num_heads, self.head_dim, self.max_seq_len)
)
else:
past_value = self.create_cache_op(
(self.batch_size, self.num_heads, self.max_seq_len, self.head_dim)
)
past_keys.append(past_key)
past_values.append(past_value)
else:
past_keys = [None] * num_layers
past_values = [None] * num_layers
hidden_states = input
curr_key_values = []
@ -251,6 +254,7 @@ class LowBitBaichuanMultiDecoderlayer(LLMBaseNNFactory):
attn_weight = self.matmul(query_states, key_states, False, True) / (
math.sqrt(self.head_dim))
attention_mask = self.convert_to_fp16(attention_mask)
attn_weight = self.eltwise_add(attn_weight, attention_mask)
attn_weight = self.convert_to_fp32(attn_weight)
attn_weight = self.softmax(attn_weight, -1)
@ -395,8 +399,8 @@ class FusedBaichuanLowBitMultiDecoderlayer(torch.nn.Module):
inputs = (
hidden_states.to(torch.float16),
attention_mask,
position_ids.to(torch.float16),
attention_mask.to(torch.int64),
position_ids.to(torch.int64),
)
for i in range(self.intra_stages):
@ -502,7 +506,9 @@ class FusedBaichuanLowBitDecoderlayer(torch.nn.Module):
seq_len = hidden_states.shape[1]
backend_cls = self.backend_cls_prefill
inputs = (hidden_states.to(torch.float16), attention_mask, position_ids.to(torch.float16))
inputs = (hidden_states.to(torch.float16),
attention_mask.to(torch.int64),
position_ids.to(torch.int64))
inputs += (self.layer_norm_0, self.layer_norm_1)
hidden_states, past_key, past_value = run_model(
inputs, self.op_parameters, backend_cls, self.op_id, replica=2
@ -625,9 +631,9 @@ def run_decode(
pad_mask = (0, pad_len)
padded_causal_mask = F.pad(
attention_mask.to(torch.float16), pad_mask, value=torch.finfo(torch.float16).min
attention_mask.to(torch.int64), pad_mask, value=torch.iinfo(torch.int64).min
)
padded_causal_mask[:, :, :, -1] = 0.0
padded_causal_mask[:, :, :, -1] = 0
dist.recv(hidden_states, src=rank - 1)
layer_outputs = multi_decoder(
hidden_states,
@ -869,9 +875,9 @@ class PrefillRunner:
hidden_states = F.pad(hidden_states.to(torch.float16), (0, 0, 0, pad_len), value=0.0)
position_ids = F.pad(position_ids, (0, pad_len), value=0)
attention_mask = F.pad(
attention_mask.to(torch.float16),
attention_mask.to(torch.int64),
(0, pad_len, 0, pad_len),
value=torch.finfo(torch.float16).min,
value=torch.iinfo(torch.int64).min,
)
args = (hidden_states, position_ids, attention_mask, past_key_value)

64
python/llm/src/ipex_llm/transformers/npu_models/convert_mp.py
View file

@ -192,6 +192,41 @@ def convert_llama(
convert_forward(model, LlamaForCausalLM, llama2_casullm_forward)
def convert_baichuan(
model: torch.nn.Module,
max_output_len=1024,
max_prompt_len=1024,
decoder=False,
inter_pp=None,
intra_pp=None,
transpose_value_cache=True,
):
from ipex_llm.transformers.npu_models.baichuan_mp import gen_baichuan_fused_model_forward
from ipex_llm.transformers.npu_models.baichuan_mp import DecodeRunner, PrefillRunner
if decoder:
decode_runner = DecodeRunner(
model,
max_seq_len=max_output_len,
inter_pp=inter_pp,
intra_pp=intra_pp,
transpose_value_cache=transpose_value_cache,
)
else:
decode_runner = None
prefill_runner = PrefillRunner(
model,
max_output_len=max_output_len,
max_prompt_len=max_prompt_len,
transpose_value_cache=transpose_value_cache,
)
baichuan_model_forward = gen_baichuan_fused_model_forward(
prefill_runner=prefill_runner, decode_runner=decode_runner
)
modeling_module_name = model.__class__.__module__
module = importlib.import_module(modeling_module_name)
convert_forward(model, module.BaichuanModel, baichuan_model_forward)
def optimize_llm(
model: torch.nn.Module,
max_context_len=1024,
@ -297,28 +332,13 @@ def optimize_llm(
intra_pp = 2
if inter_pp is None:
inter_pp = 2
from ipex_llm.transformers.npu_models.baichuan_mp import gen_baichuan_fused_model_forward
from ipex_llm.transformers.npu_models.baichuan_mp import DecodeRunner, PrefillRunner
decode_runner = DecodeRunner(
model,
max_seq_len=max_context_len,
inter_pp=inter_pp,
intra_pp=intra_pp,
transpose_value_cache=transpose_value_cache,
)
prefill_runner = PrefillRunner(
model,
max_output_len=max_context_len,
max_prompt_len=max_prompt_len,
transpose_value_cache=transpose_value_cache,
)
baichuan_model_forward = gen_baichuan_fused_model_forward(
prefill_runner=prefill_runner, decode_runner=decode_runner
)
modeling_module_name = model.__class__.__module__
module = importlib.import_module(modeling_module_name)
convert_forward(model, module.BaichuanModel, baichuan_model_forward)
convert_baichuan(model,
max_output_len=max_context_len,
max_prompt_len=max_prompt_len,
inter_pp=inter_pp,
intra_pp=intra_pp,
decoder=True,
transpose_value_cache=transpose_value_cache)
if isinstance(model.lm_head, SlicedLMHead):
model.lm_head.get_fused_lm_head()

131
python/llm/src/ipex_llm/transformers/npu_pipeline_model/baichuan.py Normal file
View file

@ -0,0 +1,131 @@
#
# 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.
#
import torch
import numpy as np
import os
from .common import update_names_of_IR_and_export_blob, LLMEmbedding, LowBitLLMLMHead
def convert_lm_head_and_embedding(model, n_splits_linear, temp_dir, weight_dir):
num_heads = model.model.layers[0].self_attn.num_heads
head_dim = model.model.layers[0].self_attn.head_dim
rms_norm_eps = model.config.rms_norm_eps
vocab_size = model.config.vocab_size
model_norm = model.model.norm
lm_head = model.lm_head
weights = [(lm_head.weight, lm_head.scale)]
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 = LowBitLLMLMHead(
[1, 1, num_heads * head_dim],
num_heads=num_heads,
max_seq_len=1,
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,
)
last_blob_path = update_names_of_IR_and_export_blob(new_lm_head, "lm_head", temp_dir)
# save weights bins files
weight_numpy = [
lm_head.weight.data.numpy(), lm_head.scale.data.numpy(),
]
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 = LLMEmbedding(
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)
return first_blob_path, last_blob_path
def convert_baichuan_layer(model, layer_idx, n_splits_linear, n_splits_down_proj,
temp_dir, weight_dir, transpose_value_cache, kv_len, group_size):
num_heads = model.model.layers[0].self_attn.num_heads
head_dim = model.model.layers[0].self_attn.head_dim
intermediate_size = model.config.intermediate_size
rms_norm_eps = model.config.rms_norm_eps
from ipex_llm.transformers.npu_models.baichuan_mp import LowBitBaichuanMultiDecoderlayer
curr_layer = model.model.layers[layer_idx]
attn_layer = curr_layer.self_attn
mlp_layer = curr_layer.mlp
weights = []
if n_splits_linear == 1:
weights = [
(attn_layer.W_pack.weight, attn_layer.W_pack.scale),
(attn_layer.o_proj.weight, attn_layer.o_proj.scale),
(mlp_layer.gate_proj.weight, mlp_layer.gate_proj.scale),
(mlp_layer.up_proj.weight, mlp_layer.up_proj.scale),
(mlp_layer.down_proj.weight, mlp_layer.down_proj.scale),
]
else:
# TODO
pass
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
single_decoder = LowBitBaichuanMultiDecoderlayer(
[1, 1, num_heads * head_dim],
input_layernorm_weights=[layer_norm_0],
post_attn_layernorm_weights=[layer_norm_1],
cached_cos=cached_cos,
cached_sin=cached_sin,
num_heads=num_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,
)
rest_blob_path = update_names_of_IR_and_export_blob(single_decoder,
f"decoder_layer_{layer_idx}",
temp_dir)
for idx, (weight, scale) in enumerate(weights):
bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_{5+idx*2}.bin")
weight.numpy().tofile(bin_file)
bin_file = os.path.join(weight_dir, f"model_{layer_idx}_input_{5+idx*2+1}.bin")
scale.numpy().tofile(bin_file)
del single_decoder

102
python/llm/src/ipex_llm/transformers/npu_pipeline_model/common.py
View file

@ -17,6 +17,10 @@
from openvino.runtime import Core, serialize
import os
from ipex_llm.transformers.npu_models.mp_models_base import LLMBaseNNFactory
from typing import Sequence
from intel_npu_acceleration_library.backend.factory import NNFactory
import numpy as np
def update_names_of_IR_and_export_blob(model, model_name, dir):
@ -52,3 +56,101 @@ def update_names_of_IR_and_export_blob(model, model_name, dir):
os.remove(new_ir_path)
return blob_path
class LowBitLLMLMHead(LLMBaseNNFactory):
def __init__(
self,
hidden_shape: Sequence[int],
num_heads: int,
rms_norm_eps: float,
model_norm_weight,
vocab_size: int,
mode: str = "decode",
dtype: np.dtype = np.int8,
max_seq_len: int = 1024,
transpose_value: bool = False,
profile: bool = False,
device: str = "NPU",
n_splits: int = 1,
):
super().__init__(max_seq_len=max_seq_len,
transpose_value=transpose_value,
dtype=dtype,
profile=profile,
device=device)
self.max_seq_len = max_seq_len
self.dtype = dtype
self.batch_size, self.seq_len, self.hidden_size = hidden_shape
self.mode = mode
self.rms_norm_eps = rms_norm_eps
self.transpose_value = transpose_value
self.vocab_size = vocab_size
self.num_heads = num_heads
self.head_dim = self.hidden_size // self.num_heads
# define input, the order self.parameter matters
input = self.create_input_op((self.batch_size, self.seq_len, self.hidden_size))
hidden_states = input
# model norm and lm head
model_norm_weight = self.constant(model_norm_weight)
hidden_states = self.layer_norm(hidden_states, model_norm_weight)
if n_splits == 1:
hidden_states = self.linear(
hidden_states, self.vocab_size, self.hidden_size, bias=False, wt_dtype=self.dtype
)
else:
hidden_states = self.dq_split_linear(
hidden_states, self.vocab_size, self.hidden_size, n_splits,
wt_dtype=dtype, scale_factor=False
)
# define outputs
hidden_states = self.convert_to_fp32(hidden_states)
print("start compiling")
self.compile()
class LLMEmbedding(NNFactory):
def __init__(
self,
vocab_size,
embedding_dim,
embedding_weight,
padding_idx,
dtype, # fp16
device: str = "NPU",
):
super().__init__(False, device)
self.vocab_size = vocab_size
self.embedding_dim = embedding_dim
self.padding_idx = padding_idx
self.dtype = dtype
# define input
weight = self.constant(embedding_weight)
input = self.parameter((1, 1), dtype=np.int32)
if padding_idx == -1:
padding_idx += vocab_size
axis_node = self.constant(np.array([0], dtype=np.int64))
if padding_idx is not None:
masked_embeddings = np.ones(weight.shape, dtype=np.float16)
masked_embeddings[padding_idx, :] = 0.0 # mask
node_mask = self.constant(masked_embeddings)
node_masked_w = self.eltwise_mul(weight, node_mask)
res = self.gather(node_masked_w, input, axis_node, 0)
else:
res = self.gather(weight, input, axis_node, 0)
# define outputs
res = self.convert_to_fp16(res)
print("start compiling")
self.compile()

49
python/llm/src/ipex_llm/transformers/npu_pipeline_model/convert_pipeline.py
View file

@ -122,6 +122,7 @@ def generate(
thread = threading.Thread(target=generate_serve,
args=(self.kv_len, self.num_head,
self.head_dim, self.num_layers,
self.vocab_size,
self.transpose_value_cache,
new_tokens - 2))
thread.start()
@ -163,11 +164,11 @@ def generate(
break
token = int.from_bytes(data, sys.byteorder)
idx += 1
if token == eos:
break
output_tokens.append(torch.tensor([token]))
if streamer is not None:
streamer.put(torch.tensor([token]))
if token == eos:
break
output = torch.stack(output_tokens, dim=1)
output = torch.cat((inputs, output), dim=1)
@ -231,7 +232,47 @@ def convert_llm(model: torch.nn.Module,
model.transpose_value_cache = transpose_value_cache
try:
res = InitLLMPipeline(kv_len, model.num_head, model.head_dim, layer_num,
res = InitLLMPipeline("llama", kv_len, model.num_head, model.head_dim, layer_num,
model.vocab_size, weight_dir, "model",
first_blob_path, last_blob_path,
os.path.join(temp_dir, "decoder_layer"))
except:
invalidInputError(False,
"False to InitLLMPipeline.")
elif model.config.model_type == "baichuan":
with tempfile.TemporaryDirectory() as temp_dir:
weight_dir = os.path.join(temp_dir, "model_weights")
os.mkdir(weight_dir)
layer_num = len(model.model.layers)
from .baichuan import convert_baichuan_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)
param_list = []
for layer_idx in range(0, layer_num):
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_baichuan_layer, param_list)
# Prefill Runner
from ipex_llm.transformers.npu_models.convert_mp import convert_baichuan
convert_baichuan(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 = 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
model.vocab_size = model.config.vocab_size
try:
res = InitLLMPipeline("baichuan", kv_len, model.num_head, model.head_dim, layer_num,
model.vocab_size, weight_dir, "model",
first_blob_path, last_blob_path,
os.path.join(temp_dir, "decoder_layer"))
@ -240,7 +281,7 @@ def convert_llm(model: torch.nn.Module,
"False to InitLLMPipeline.")
else:
invalidInputError(False,
"Now we only support Llama2 for pipeline running.")
"Now we only support Llama2 / Llama3 / Baichuan2 for pipeline running.")
if isinstance(model.lm_head, SlicedLMHead):
model.lm_head.get_fused_lm_head()

111
python/llm/src/ipex_llm/transformers/npu_pipeline_model/llama.py
View file

@ -17,112 +17,8 @@
import torch
import numpy as np
from ipex_llm.transformers.npu_models.mp_models_base import LLMBaseNNFactory
from typing import Sequence
from intel_npu_acceleration_library.backend.factory import NNFactory
import os
from .common import update_names_of_IR_and_export_blob
class LowBitLlamaLMHead(LLMBaseNNFactory):
def __init__(
self,
hidden_shape: Sequence[int],
num_heads: int,
num_key_value_heads: int,
rms_norm_eps: float,
model_norm_weight,
vocab_size: int,
mode: str = "decode",
dtype: np.dtype = np.int8,
max_seq_len: int = 1024,
transpose_value: bool = False,
profile: bool = False,
device: str = "NPU",
n_splits: int = 1,
):
super().__init__(max_seq_len=max_seq_len,
transpose_value=transpose_value,
dtype=dtype,
profile=profile,
device=device)
self.max_seq_len = max_seq_len
self.dtype = dtype
self.batch_size, self.seq_len, self.hidden_size = hidden_shape
self.mode = mode
self.rms_norm_eps = rms_norm_eps
self.transpose_value = transpose_value
self.vocab_size = vocab_size
self.num_heads = num_heads
self.num_key_value_heads = num_key_value_heads
self.head_dim = self.hidden_size // self.num_heads
self.num_key_value_groups = self.num_heads // self.num_key_value_heads
# define input, the order self.parameter matters
input = self.create_input_op((self.batch_size, self.seq_len, self.hidden_size))
hidden_states = input
# model norm and lm head
model_norm_weight = self.constant(model_norm_weight)
hidden_states = self.layer_norm(hidden_states, model_norm_weight)
if n_splits == 1:
hidden_states = self.linear(
hidden_states, self.vocab_size, self.hidden_size, bias=False, wt_dtype=self.dtype
)
else:
hidden_states = self.dq_split_linear(
hidden_states, self.vocab_size, self.hidden_size, n_splits,
wt_dtype=dtype, scale_factor=False
)
# define outputs
hidden_states = self.convert_to_fp32(hidden_states)
print("start compiling")
self.compile()
class LlamaEmbedding(NNFactory):
def __init__(
self,
vocab_size,
embedding_dim,
embedding_weight,
padding_idx,
dtype, # fp16
device: str = "NPU",
):
super().__init__(False, device)
self.vocab_size = vocab_size
self.embedding_dim = embedding_dim
self.padding_idx = padding_idx
self.dtype = dtype
# define input
weight = self.constant(embedding_weight)
input = self.parameter((1, 1), dtype=np.int32)
if padding_idx == -1:
padding_idx += vocab_size
if padding_idx is not None:
masked_embeddings = np.ones(weight.shape, dtype='int64')
masked_embeddings[padding_idx, :] = 0 # mask
node_mask = self.constant(masked_embeddings)
node_masked_w = self.matmul(weight, node_mask, False, True)
axis_node = self.constant(np.array([0], dtype=np.int64))
res = self.gather(node_masked_w if padding_idx else weight, input, axis_node, 0)
# define outputs
res = self.convert_to_fp16(res)
print("start compiling")
self.compile()
from .common import update_names_of_IR_and_export_blob, LLMEmbedding, LowBitLLMLMHead
def convert_lm_head_and_embedding(model, n_splits_linear, temp_dir, weight_dir):
@ -149,10 +45,9 @@ def convert_lm_head_and_embedding(model, n_splits_linear, temp_dir, weight_dir):
else: # FP16 Linear
np_dtype = np.float16
new_lm_head = LowBitLlamaLMHead(
new_lm_head = LowBitLLMLMHead(
[1, 1, num_heads * head_dim],
num_heads=num_heads,
num_key_value_heads=num_key_value_heads,
max_seq_len=1,
rms_norm_eps=rms_norm_eps,
mode="decode",
@ -177,7 +72,7 @@ def convert_lm_head_and_embedding(model, n_splits_linear, temp_dir, weight_dir):
weight.tofile(bin_file)
embedding_layer = model.model.embed_tokens
new_embedding = LlamaEmbedding(
new_embedding = LLMEmbedding(
vocab_size=model.config.vocab_size,
embedding_dim=model.config.hidden_size,
embedding_weight=embedding_layer.weight.to(torch.float16).detach().numpy(),

20
python/llm/src/ipex_llm/transformers/npu_pipeline_model/pipeline_cpp.py
View file

@ -43,23 +43,23 @@ _, _lib_path = get_shared_lib_info("pipeline")
# Load the library
_lib = ctypes.cdll.LoadLibrary(_lib_path)
_lib.InitLLMPipeline.argtypes = [ctypes.c_int] * 5 + [ctypes.c_char_p] * 5
_lib.InitLLMPipeline.argtypes = [ctypes.c_char_p] + [ctypes.c_int] * 5 + [ctypes.c_char_p] * 5
_lib.InitLLMPipeline.restype = ctypes.c_int
_lib.generate_serve.argtypes = [ctypes.c_int] * 4 + [ctypes.c_bool] + [ctypes.c_int]
_lib.generate_serve.argtypes = [ctypes.c_int] * 5 + [ctypes.c_bool] + [ctypes.c_int]
_lib.generate_serve.restype = ctypes.c_int
def InitLLMPipeline(kv_len: int, num_head: int, head_dim: int, num_layers: int, vocab_size: int,
model_weight_dir: str, model_name: str,
def InitLLMPipeline(model_type: str, kv_len: int, num_head: int, head_dim: int, num_layers: int,
vocab_size: int, model_weight_dir: str, model_name: str,
first_blob_name: str, last_blob_name: str, rest_blob_name: str):
return _lib.InitLLMPipeline(kv_len, num_head, head_dim, num_layers, vocab_size,
model_weight_dir.encode('utf-8'), model_name.encode('utf-8'),
first_blob_name.encode('utf-8'), last_blob_name.encode('utf-8'),
rest_blob_name.encode('utf-8'))
return _lib.InitLLMPipeline(model_type.encode('utf-8'), kv_len, num_head, head_dim, num_layers,
vocab_size, model_weight_dir.encode('utf-8'),
model_name.encode('utf-8'), first_blob_name.encode('utf-8'),
last_blob_name.encode('utf-8'), rest_blob_name.encode('utf-8'))
def generate_serve(kv_len: int, num_head: int, head_dim: int, num_layers: int,
transpose_value_cache: bool, param_n_output: int):
vocab_size: int, transpose_value_cache: bool, param_n_output: int):
_lib.generate_serve(kv_len, num_head, head_dim, num_layers,
transpose_value_cache, param_n_output)
vocab_size, transpose_value_cache, param_n_output)