Open
Description
Describe the bug
I found that the CPU memory increase happens when accelerate calls "loss.backward()" (https://github.com/huggingface/accelerate/blob/main/src/accelerate/accelerator.py#L1989) when doing LoRA finetuing on Intel GPU Max 1100.
Reproduce memory leak using ipex an transformers.
python finetune.py \
--micro_batch_size 1 \
--batch_size 4 \
--gradient_checkpointing False \
--base_model meta-llama/Llama-2-7b-hfmemory trend
finetune.py
import os
import sys
os.environ["ACCELERATE_USE_IPEX"] = "1"
os.environ["ACCELERATE_USE_XPU"] = "1"
from typing import List
import fire
import torch
import transformers
from datasets import load_dataset
import accelerate
"""
Unused imports:
import torch.nn as nn
import bitsandbytes as bnb
"""
from peft import (
LoraConfig,
get_peft_model_state_dict,
set_peft_model_state_dict,
)
from transformers import LlamaTokenizer, AutoTokenizer
from transformers import AutoModelForCausalLM
import intel_extension_for_pytorch as ipex
from peft import get_peft_model, prepare_model_for_kbit_training
from utils.prompter import Prompter
def train(
# model/data params
base_model: str = "meta-llama/Llama-2-7b-hf", # the only required argument
saved_low_bit_model: str = None,
data_path: str = "yahma/alpaca-cleaned",
output_dir: str = "./lora-alpaca",
# training hyperparams
batch_size: int = 128,
micro_batch_size: int = 2,
num_epochs: int = 10,
learning_rate: float = 3e-5,
cutoff_len: int = 256,
val_set_size: int = 0,
# lora hyperparams
lora_r: int = 8,
lora_alpha: int = 16,
lora_dropout: float = 0.05,
lora_target_modules: List[str] = [
"q_proj",
"v_proj",
"k_proj",
"o_proj",
"up_proj",
"down_proj",
"gate_proj"
],
# llm hyperparams
train_on_inputs: bool = True, # if False, masks out inputs in loss
add_eos_token: bool = False,
group_by_length: bool = False, # faster, but produces an odd training loss curve
# wandb params
wandb_project: str = "",
wandb_run_name: str = "",
wandb_watch: str = "", # options: false | gradients | all
wandb_log_model: str = "", # options: false | true
resume_from_checkpoint: str = None, # either training checkpoint or final adapter
prompt_template_name: str = "alpaca", # The prompt template to use, will default to alpaca.
gradient_checkpointing: bool = False,
deepspeed: str = None,
):
if int(os.environ.get("LOCAL_RANK", 0)) == 0:
print(
f"Training Alpaca-LoRA model with params:\n"
f"base_model: {base_model}\n"
f"data_path: {data_path}\n"
f"output_dir: {output_dir}\n"
f"batch_size: {batch_size}\n"
f"micro_batch_size: {micro_batch_size}\n"
f"num_epochs: {num_epochs}\n"
f"learning_rate: {learning_rate}\n"
f"cutoff_len: {cutoff_len}\n"
f"val_set_size: {val_set_size}\n"
f"lora_r: {lora_r}\n"
f"lora_alpha: {lora_alpha}\n"
f"lora_dropout: {lora_dropout}\n"
f"lora_target_modules: {lora_target_modules}\n"
f"train_on_inputs: {train_on_inputs}\n"
f"add_eos_token: {add_eos_token}\n"
f"group_by_length: {group_by_length}\n"
f"wandb_project: {wandb_project}\n"
f"wandb_run_name: {wandb_run_name}\n"
f"wandb_watch: {wandb_watch}\n"
f"wandb_log_model: {wandb_log_model}\n"
f"resume_from_checkpoint: {resume_from_checkpoint or False}\n"
f"prompt template: {prompt_template_name}\n"
)
assert (
base_model
), "Please specify a --base_model, e.g. --base_model='huggyllama/llama-7b'"
gradient_accumulation_steps = batch_size // micro_batch_size
prompter = Prompter(prompt_template_name)
device_map = "auto"
world_size = int(os.environ.get("WORLD_SIZE", 1))
ddp = world_size != 1
if ddp:
device_map = {"": int(os.environ.get("LOCAL_RANK") or 0)}
gradient_accumulation_steps = gradient_accumulation_steps // world_size
# Check if parameter passed or if set within environ
use_wandb = len(wandb_project) > 0 or (
"WANDB_PROJECT" in os.environ and len(os.environ["WANDB_PROJECT"]) > 0
)
# Only overwrite environ if wandb param passed
if len(wandb_project) > 0:
os.environ["WANDB_PROJECT"] = wandb_project
if len(wandb_watch) > 0:
os.environ["WANDB_WATCH"] = wandb_watch
if len(wandb_log_model) > 0:
os.environ["WANDB_LOG_MODEL"] = wandb_log_model
if saved_low_bit_model is not None:
model = AutoModelForCausalLM.load_low_bit(
saved_low_bit_model,
optimize_model=False,
torch_dtype=torch.bfloat16,
modules_to_not_convert=["lm_head"],
)
else:
model = AutoModelForCausalLM.from_pretrained(
base_model,
# load_in_low_bit="nf4",
# load_in_4bit=True,
torch_dtype=torch.bfloat16,
# device_map=device_map,
# modules_to_not_convert=["lm_head"],
)
print(f"Model loaded on rank {os.environ.get('LOCAL_RANK')}")
model = model.to(f'xpu:{os.environ.get("LOCAL_RANK", 0)}')
print(f"Model moved to rank {os.environ.get('LOCAL_RANK')}")
tokenizer = AutoTokenizer.from_pretrained(base_model)
print(f"Tokenizer loaded on rank {os.environ.get('LOCAL_RANK')}")
tokenizer = LlamaTokenizer.from_pretrained(base_model)
tokenizer.pad_token_id = (
0 # unk. we want this to be different from the eos token
)
tokenizer.padding_side = "left" # Allow batched inference
print(model)
def tokenize(prompt, add_eos_token=True):
# there's probably a way to do this with the tokenizer settings
# but again, gotta move fast
result = tokenizer(
prompt,
truncation=True,
max_length=cutoff_len,
padding="max_length",
return_tensors=None,
)
if (
result["input_ids"][-1] != tokenizer.eos_token_id
and len(result["input_ids"]) < cutoff_len
and add_eos_token
):
result["input_ids"].append(tokenizer.eos_token_id)
result["attention_mask"].append(1)
result["labels"] = result["input_ids"].copy()
return result
def generate_and_tokenize_prompt(data_point):
full_prompt = prompter.generate_prompt(
data_point["instruction"],
data_point["input"],
data_point["output"],
)
tokenized_full_prompt = tokenize(full_prompt, )
if not train_on_inputs:
user_prompt = prompter.generate_prompt(
data_point["instruction"], data_point["input"]
)
tokenized_user_prompt = tokenize(
user_prompt, add_eos_token=add_eos_token
)
user_prompt_len = len(tokenized_user_prompt["input_ids"])
if add_eos_token:
user_prompt_len -= 1
tokenized_full_prompt["labels"] = [
-100
] * user_prompt_len + tokenized_full_prompt["labels"][
user_prompt_len:
] # could be sped up, probably
return tokenized_full_prompt
model = prepare_model_for_kbit_training(model, use_gradient_checkpointing=gradient_checkpointing)
config = LoraConfig(
r=lora_r,
lora_alpha=lora_alpha,
target_modules=lora_target_modules,
lora_dropout=lora_dropout,
bias="none",
task_type="CAUSAL_LM",
)
model = get_peft_model(model, config)
if data_path.endswith(".json") or data_path.endswith(".jsonl"):
data = load_dataset("json", data_files=data_path, split="train[:100]")
else:
data = load_dataset(data_path, split="train[:100]")
if resume_from_checkpoint:
# Check the available weights and load them
checkpoint_name = os.path.join(
resume_from_checkpoint, "pytorch_model.bin"
) # Full checkpoint
if not os.path.exists(checkpoint_name):
checkpoint_name = os.path.join(
resume_from_checkpoint, "adapter_model.bin"
) # only LoRA model - LoRA config above has to fit
resume_from_checkpoint = (
False # So the trainer won't try loading its state
)
# The two files above have a different name depending on how they were saved, but are actually the same.
if os.path.exists(checkpoint_name):
print(f"Restarting from {checkpoint_name}")
adapters_weights = torch.load(checkpoint_name)
set_peft_model_state_dict(model, adapters_weights)
else:
print(f"Checkpoint {checkpoint_name} not found")
# model.print_trainable_parameters() # Be more transparent about the % of trainable params.
if val_set_size > 0:
train_val = data["train"].train_test_split(
test_size=val_set_size, shuffle=True, seed=42
)
train_data = (
train_val["train"].shuffle().map(generate_and_tokenize_prompt)
)
val_data = (
train_val["test"].shuffle().map(generate_and_tokenize_prompt)
)
else:
train_data = data.shuffle().map(generate_and_tokenize_prompt)
val_data = None
if not ddp and torch.cuda.device_count() > 1:
# keeps Trainer from trying its own DataParallelism when more than 1 gpu is available
model.is_parallelizable = True
model.model_parallel = True
trainer = transformers.Trainer(
model=model,
train_dataset=train_data,
eval_dataset=val_data,
args=transformers.TrainingArguments(
per_device_train_batch_size=micro_batch_size,
gradient_accumulation_steps=gradient_accumulation_steps,
# warmup_ratio=0.03,
# warmup_steps=100,
max_grad_norm=0.3,
num_train_epochs=num_epochs,
learning_rate=learning_rate,
lr_scheduler_type="cosine",
bf16=True,
logging_steps=1,
optim="adamw_torch",
evaluation_strategy="steps" if val_set_size > 0 else "no",
save_strategy="steps",
eval_steps=100 if val_set_size > 0 else None,
save_steps=100,
output_dir=output_dir,
save_total_limit=100,
load_best_model_at_end=True if val_set_size > 0 else False,
ddp_find_unused_parameters=False if ddp else None,
group_by_length=group_by_length,
report_to="wandb" if use_wandb else None,
run_name=wandb_run_name if use_wandb else None,
gradient_checkpointing=gradient_checkpointing,
ddp_backend="ccl",
deepspeed=deepspeed,
),
data_collator=transformers.DataCollatorForSeq2Seq(
tokenizer, pad_to_multiple_of=8, return_tensors="pt", padding=True
),
)
model.config.use_cache = False
trainer.train(resume_from_checkpoint=resume_from_checkpoint)
model.save_pretrained(output_dir)
print(
"\n If there's a warning about missing keys above, please disregard :)"
)
if __name__ == "__main__":
fire.Fire(train)Versions
Collecting environment information...
PyTorch version: N/A
PyTorch CXX11 ABI: N/A
IPEX version: N/A
IPEX commit: N/A
Build type: N/A
OS: Ubuntu 22.04.3 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: N/A
IGC version: N/A
CMake version: version 3.22.1
Libc version: glibc-2.35
Python version: 3.9.18 (main, Sep 11 2023, 13:41:44) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-5.15.0-73-generic-x86_64-with-glibc2.35
Is XPU available: N/A
DPCPP runtime version: N/A
MKL version: N/A
GPU models and configuration:
N/A
Intel OpenCL ICD version: 23.17.26241.33-647~22.04
Level Zero version: 1.3.26241.33-647~22.04
CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 52 bits physical, 57 bits virtual
Byte Order: Little Endian
CPU(s): 224
On-line CPU(s) list: 0-223
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) Platinum 8480+
CPU family: 6
Model: 143
Thread(s) per core: 2
Core(s) per socket: 56
Socket(s): 2
Stepping: 8
CPU max MHz: 3800.0000
CPU min MHz: 800.0000
BogoMIPS: 4000.00
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cat_l2 cdp_l3 invpcid_single intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
Virtualization: VT-x
L1d cache: 5.3 MiB (112 instances)
L1i cache: 3.5 MiB (112 instances)
L2 cache: 224 MiB (112 instances)
L3 cache: 210 MiB (2 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0-55,112-167
NUMA node1 CPU(s): 56-111,168-223
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Enhanced IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS SW sequence
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Versions of relevant libraries:
[pip3] intel-extension-for-pytorch==2.0.110+xpu
[pip3] numpy==1.26.0
[pip3] torch==2.0.1a0+cxx11.abi
[pip3] torchvision==0.15.2a0+cxx11.abi
[conda] intel-extension-for-pytorch 2.0.110+xpu pypi_0 pypi
[conda] numpy 1.26.0 pypi_0 pypi
[conda] torch 2.0.1a0+cxx11.abi pypi_0 pypi
[conda] torchvision 0.15.2a0+cxx11.abi pypi_0 pypi
Other relevant libraries:
pip install transformers==4.34.0
pip install peft==0.5.0
pip install oneccl_bind_pt==2.0.100 -f https://developer.intel.com/ipex-whl-stable-xpu
pip install git+https://github.com/microsoft/DeepSpeed.git@78c518e
pip install git+https://github.com/intel/intel-extension-for-deepspeed.git@ec33277
pip install accelerate==0.23.0
pip install datasets
pip install fire