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

LISA Finetuning Example (#10743)

Browse source 
* enabling xetla only supports qtype=SYM_INT4 or FP8E5

* LISA Finetuning Example on gpu

* update readme

* add licence

* Explain parameters of lisa & Move backend codes to src dir

* fix style

* fix style

* update readme

* support chatglm

* fix style

* fix style

* update readme

* fix
This commit is contained in:
Ziteng Zhang 2024-04-18 13:48:10 +08:00 committed by GitHub
parent 581ebf6104
commit ff040c8f01
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 327 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

75
python/llm/example/GPU/LLM-Finetuning/LISA/README.md Normal file
View file

@ -0,0 +1,75 @@
# LISA Finetuning with IPEX-LLM
This example refers to [LISA with LMFLow's DynamicLayerActivationCallback Class](https://github.com/OptimalScale/LMFlow/blob/f3b3b007ea526009172c355e9d52ffa146b9dc0c/src/lmflow/pipeline/finetuner.py#L301), and adds [LISA fintuning](https://arxiv.org/abs/2403.17919) to IPEX-LLM on [Intel GPU](../../../GPU/README.md), based on [LORA finetuning with IPEX-LLM](../LoRA/alpaca_lora_finetuning.py).
### 0. Requirements
To run this example with IPEX-LLM on Intel GPUs, we have some recommended requirements for your machine, please refer to [here](../../../GPU/README.md#requirements) for more information.
### 1. Install
```bash
conda create -n llm python=3.11
conda activate llm
# below command will install intel_extension_for_pytorch==2.1.10+xpu as default
pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
pip install accelerate==0.23.0
pip install bitsandbytes==0.43.0
pip install datasets==2.18.0
pip install --upgrade transformers==4.36.0
pip install scipy fire
```
### 2. LISA Finetune
```bash
# Configures OneAPI environment variables
source /opt/intel/oneapi/setvars.sh
python ./lisa_finetuning.py \
--micro_batch_size 8 \
--batch_size 128 \
--base_model "meta-llama/Llama-2-7b-hf" \
--data_path "yahma/alpaca-cleaned" \
--output_dir "./ipex-llm-lisa-alpaca" \
--gradient_checkpointing True \
--lisa_activated_layers 1 \
--lisa_interval_steps 20
```
Optional parameters for `lisa_finetuning.py`:
**--repo-id-or-model-path** : default to `meta-llama/Llama-2-7b-hf`, and you can also specify your local model path.
**--data-path** : default to `yahma/alpaca-cleaned`, and you can also specify your local datal path, while note that changing to the other datasets will introduce code modification effort for yourself.
**--output-dir** : default to `./ipex-llm-lisa-alpaca` to save fine-tuned model, and you can change if needed.
**--lisa_activated_layers** : the number of self-attention layers randomly selected to activate.
**lisa_interval_steps** : the number of interval steps to switch active layers.
### 3. Sample Output
```log
......
{'loss': 1.8391, 'learning_rate': 1.9967238104745695e-05, 'epoch': 0.03}
{'loss': 1.8242, 'learning_rate': 1.9869167087338908e-05, 'epoch': 0.05}
5%|██████▉ | 20/388 [xx:xx<x:xx:xx, x.xxs/it]
Activating layers at indices: [10] for the next steps.
{'loss': 1.8128, 'learning_rate': 1.9706429546259592e-05, 'epoch': 0.08}
{'loss': 1.775, 'learning_rate': 1.9480091799562706e-05, 'epoch': 0.1}
10%|██████████████ | 40/388 [xx:xx<xx:xx, x.xxs/it]
Activating layers at indices: [30] for the next steps.
{'loss': 1.7669, 'learning_rate': 1.9191636897958123e-05, 'epoch': 0.13}
{'loss': 1.7749, 'learning_rate': 1.8842954907300236e-05, 'epoch': 0.15}
15%|█████████████████████ | 60/388 [xx:xx<xx:xx, x.xxs/it]
Activating layers at indices: [26] for the next steps.
{'loss': 1.7735, 'learning_rate': 1.8436330524160048e-05, 'epoch': 0.18}
{'loss': 1.7199, 'learning_rate': 1.797442810562721e-05, 'epoch': 0.21}
21%|████████████████████████████ | 80/388 [xx:xx<xx:xx, x.xxs/it]
Activating layers at indices: [17] for the next steps.
{'loss': 1.7328, 'learning_rate': 1.7460274211432463e-05, 'epoch': 0.23}
25%|█████████████████████████████████▋ | 96/388 [xx:xx<xx:xx, x.xxs/it]
......
```

167
python/llm/example/GPU/LLM-Finetuning/LISA/lisa_finetuning.py Normal file
View file

@ -0,0 +1,167 @@
#
# 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 os
from typing import List
import fire
import torch
import transformers
from datasets import load_dataset
import accelerate
from ipex_llm.transformers.lisa import DynamicLayerActivationCallback
from transformers import AutoTokenizer
current_dir = os.path.dirname(os.path.realpath(__file__))
common_util_path = os.path.join(current_dir, '..')
import sys
sys.path.append(common_util_path)
from common.utils import Prompter, get_train_val_data
from ipex_llm.transformers import AutoModelForCausalLM
from ipex_llm.utils.common import invalidInputError
def train(
# model/data params
base_model: str = "meta-llama/Llama-2-7b-hf", # the only required argument, default to be "meta-llama/Llama-2-7b-hf"
data_path: str = "yahma/alpaca-cleaned",
output_dir: str = "./ipex-llm-lisa-alpaca",
# training hyperparams
bf16: bool = True, # default to bf16
batch_size: int = 128,
micro_batch_size: int = 8, # default to be 8, limited by GPU memory
num_epochs: int = 1,
learning_rate: float = 2e-5,
cutoff_len: int = 256,
val_set_size: int = 2000,
# 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
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,
training_mode: str = "lisa",
lisa_activated_layers: int = 1,
lisa_interval_steps: int = 20,
):
invalidInputError(training_mode == "lisa",
f"This example is for lisa training mode, but got training_mode={training_mode}.")
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"train_on_inputs: {train_on_inputs}\n"
f"add_eos_token: {add_eos_token}\n"
f"group_by_length: {group_by_length}\n"
f"resume_from_checkpoint: {resume_from_checkpoint or False}\n"
f"prompt template: {prompt_template_name}\n"
f"training_mode: {training_mode}\n"
f"lisa_activated_layers: {lisa_activated_layers}\n"
f"lisa_interval_steps: {lisa_interval_steps}\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)
model = AutoModelForCausalLM.from_pretrained(
base_model,
load_in_low_bit="bf16",
optimize_model=True,
torch_dtype=torch.bfloat16,
trust_remote_code=True,
enable_xetla=False
)
model = model.to("xpu")
tokenizer = AutoTokenizer.from_pretrained(base_model, trust_remote_code=True)
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)
if data_path.endswith(".json") or data_path.endswith(".jsonl"):
data = load_dataset("json", data_files=data_path)
else:
data = load_dataset(data_path)
train_data, val_data = get_train_val_data(data, tokenizer, prompter, train_on_inputs,
add_eos_token, cutoff_len, val_set_size, seed=42)
trainer_callbacks = []
# Instantiate the callback
dynamic_layer_activation_callback = DynamicLayerActivationCallback(
n_layers=lisa_activated_layers, # Number of layers to activate
interval_steps = lisa_interval_steps, # Step interval to update active layers
model = model
)
trainer_callbacks.append(dynamic_layer_activation_callback)
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,
max_grad_norm=0.3,
num_train_epochs=num_epochs,
learning_rate=learning_rate,
lr_scheduler_type="cosine",
bf16=bf16, # ensure training more stable
logging_steps=10,
optim="adamw_hf",
evaluation_strategy="steps" if val_set_size > 0 else "no",
save_strategy="steps",
eval_steps=200 if val_set_size > 0 else None,
save_steps=200,
output_dir=output_dir,
load_best_model_at_end=True if val_set_size > 0 else False,
group_by_length=group_by_length,
gradient_checkpointing=gradient_checkpointing,
deepspeed=deepspeed,
save_safetensors=False,
),
data_collator=transformers.DataCollatorForSeq2Seq(
tokenizer, pad_to_multiple_of=8, return_tensors="pt", padding=True
),
callbacks=trainer_callbacks
)
model.config.use_cache = False
trainer.train(resume_from_checkpoint=resume_from_checkpoint)
# model.save_pretrained(output_dir)
trainer.save_model()
if __name__ == "__main__":
fire.Fire(train)

81
python/llm/src/ipex_llm/transformers/lisa.py Normal file
View file

@ -0,0 +1,81 @@
#
# 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 transformers import TrainerCallback
import numpy as np
from ipex_llm.utils.common import invalidInputError
# source: https://github.com/OptimalScale/LMFlow/blob/main/src/lmflow/pipeline/finetuner.py
class DynamicLayerActivationCallback(TrainerCallback):
def __init__(self, n_layers, interval_steps, model):
super().__init__()
self.n_layers = n_layers
self.interval_steps = interval_steps
self.model = model
# Determine the way to access layers based on the model type
class_to_layers_map = {
'LlamaForCausalLM': 'model.model.layers',
'Qwen2ForCausalLM': 'model.model.layers',
'MistralForCausalLM': 'model.model.layers',
'MixtralForCausalLM': 'model.model.layers',
'GemmaForCausalLM': 'model.model.layers',
'GPT2LMHeadModel': 'model.transformer.h',
'ChatGLMModel': 'model.transformer.encoder.layers',
}
model_class_name = self.model.__class__.__name__
if model_class_name in class_to_layers_map:
self.layers_attribute = class_to_layers_map[model_class_name]
else:
# self.layers_attribute = training_args.lisa_layers_attribute
invalidInputError(False, f"Model {model_class_name} not supported.")
# Dynamically execute to get the number of layers
self.total_layers = len(eval('self.' + self.layers_attribute))
self.active_layers_indices = []
def freeze_all_layers(self):
layers = eval('self.' + self.layers_attribute) # Dynamically execute to get layers
for layer in layers:
for param in layer.parameters():
param.requires_grad = False
def on_step_begin(self, args, state, control, **kwargs):
# Check if it's time to switch active layers, including at step 0
if state.global_step % self.interval_steps == 0:
self.switch_active_layers()
def switch_active_layers(self):
# First, disable gradients for all layers
self.freeze_all_layers()
# Randomly select n_layers to activate
layers = eval('self.' + self.layers_attribute) # Re-fetch layer references
self.active_layers_indices = np.random.choice(
range(self.total_layers),
self.n_layers,
replace=False
)
print(
f"Activating layers at indices: {self.active_layers_indices} for the next steps.",
flush=True
)
# Enable gradients only for the selected layers
for idx in self.active_layers_indices:
for param in layers[idx].parameters():
param.requires_grad = True

5
python/llm/src/ipex_llm/transformers/models/llama.py
View file

@ -313,7 +313,10 @@ def should_use_xetla_mm_qkv(self, device):
full_attn = self.q_proj.out_len == self.k_proj.out_len == self.v_proj.out_len
supported_qtype = self.q_proj.qtype == SYM_INT4 and full_attn
supported_qtype = supported_qtype or self.q_proj.qtype == FP8E5
enable_xetla = self.q_proj.enable_xetla
if self.q_proj.qtype == SYM_INT4 or self.q_proj.qtype == FP8E5:
enable_xetla = self.q_proj.enable_xetla
else:
enable_xetla = False
return device.type == "xpu" and enable_xetla and supported_qtype