Cara menjalankan inferensi batch di GPU menggunakan Tugas Cloud Run
Tentang codelab ini
subjectTerakhir diperbarui Mar 20, 2025
1. Pengantar
Ringkasan
Dalam contoh ini, Anda akan menyesuaikan model gemma-2b dengan set data text-to-sql yang dimaksudkan untuk membuat LLM membalas dengan kueri SQL saat ditanya dalam bahasa alami. Kemudian, Anda akan mengambil model yang telah dioptimalkan dan menayangkannya di Cloud Run menggunakan vLLM.
Yang akan Anda pelajari
- Cara melakukan penyesuaian menggunakan GPU Tugas Cloud Run
- Cara menggunakan konfigurasi VPC Langsung untuk Tugas GPU agar upload dan penayangan model lebih cepat
2. Sebelum memulai
Untuk menggunakan fitur GPU, Anda harus meminta penambahan kuota untuk wilayah yang didukung. Kuota yang diperlukan adalah nvidia_l4_gpu_allocation_no_zonal_redundancy, yang berada di Cloud Run Admin API. Berikut link langsung untuk meminta kuota.
3. Penyiapan dan Persyaratan
Tetapkan variabel lingkungan yang akan digunakan di seluruh codelab ini.
PROJECT_ID=<YOUR_PROJECT_ID>
REGION=<YOUR_REGION>
HF_TOKEN=<YOUR_HF_TOKEN>
AR_REPO=codelab-finetuning-jobs
IMAGE_NAME=finetune-to-gcs
JOB_NAME=finetuning-to-gcs-job
BUCKET_NAME=$PROJECT_ID-codelab-finetuning-jobs
SECRET_ID=HF_TOKEN
SERVICE_ACCOUNT="finetune-job-sa"
SERVICE_ACCOUNT_ADDRESS=$SERVICE_ACCOUNT@$PROJECT_ID.iam.gserviceaccount.com
Buat akun layanan dengan menjalankan perintah ini:
gcloud iam service-accounts create $SERVICE_ACCOUNT \
--display-name="Cloud Run job to access HF_TOKEN Secret ID"
Menggunakan Secret Manager untuk menyimpan token akses HuggingFace.
Anda dapat mempelajari lebih lanjut cara membuat dan menggunakan secret di dokumen Secret Manager.
gcloud secrets create $SECRET_ID \
--replication-policy="automatic"
printf $HF_TOKEN | gcloud secrets versions add $SECRET_ID --data-file=-
Anda akan melihat output yang mirip dengan
you'll see output similar to
Created secret [HF_TOKEN].
Created version [1] of the secret [HF_TOKEN].
Memberikan peran Secret Manager Secret Accessor ke akun layanan komputasi default Anda
gcloud secrets add-iam-policy-binding $SECRET_ID \
--member serviceAccount:$SERVICE_ACCOUNT_ADDRESS \
--role='roles/secretmanager.secretAccessor'
Membuat bucket yang akan menghosting model yang telah dioptimalkan
gsutil mb -l $REGION gs://$BUCKET_NAME
Kemudian, berikan akses ke bucket kepada SA.
gcloud storage buckets add-iam-policy-binding gs://$BUCKET_NAME \
--member=serviceAccount:$SERVICE_ACCOUNT_ADDRESS \
--role=roles/storage.objectAdmin
Membuat repositori registry artefak untuk Tugas
gcloud artifacts repositories create $AR_REPO \
--repository-format=docker \
--location=$REGION \
--description="codelab for finetuning using CR jobs" \
--project=$PROJECT_ID
Membuat bucket Cloud Storage untuk model yang dioptimalkan
gsutil mb -l $REGION gs://$BUCKET_NAME
Terakhir, buat repo Artifact Registry untuk tugas Cloud Run Anda.
gcloud artifacts repositories create $AR_REPO \
--repository-format=docker \
--location=$REGION \
--description="codelab for finetuning using cloud run jobs"
4. Membuat gambar tugas Cloud Run
Pada langkah berikutnya, Anda akan membuat kode yang melakukan hal berikut:
- Mengimpor gemma-2b dari huggingface
- Melakukan penyesuaian pada gemma-2b dengan set data text-to-sql menggunakan set data dari huggingface. Tugas menggunakan satu GPU L4 untuk penyesuaian.
- Mengupload model yang telah dioptimalkan yang disebut new_model ke bucket GCS pengguna
Buat direktori untuk kode tugas finetuning Anda.
mkdir codelab-finetuning-job
cd codelab-finetuning-job
Buat file bernama finetune.py
# Copyright 2024 Google LLC
#
# 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
import torch
from datasets import load_dataset, Dataset
from transformers import (
AutoModelForCausalLM,
AutoTokenizer,
BitsAndBytesConfig,
TrainingArguments,
)
from peft import LoraConfig, PeftModel
from trl import SFTTrainer
from pathlib import Path
# GCS bucket to upload the model
bucket_name = os.getenv("BUCKET_NAME", "YOUR_BUCKET_NAME")
# The model that you want to train from the Hugging Face hub
model_name = os.getenv("MODEL_NAME", "google/gemma-2b")
# The instruction dataset to use
dataset_name = "b-mc2/sql-create-context"
# Fine-tuned model name
new_model = os.getenv("NEW_MODEL", "gemma-2b-sql")
################################################################################
# QLoRA parameters
################################################################################
# LoRA attention dimension
lora_r = int(os.getenv("LORA_R", "4"))
# Alpha parameter for LoRA scaling
lora_alpha = int(os.getenv("LORA_ALPHA", "8"))
# Dropout probability for LoRA layers
lora_dropout = 0.1
################################################################################
# bitsandbytes parameters
################################################################################
# Activate 4-bit precision base model loading
use_4bit = True
# Compute dtype for 4-bit base models
bnb_4bit_compute_dtype = "float16"
# Quantization type (fp4 or nf4)
bnb_4bit_quant_type = "nf4"
# Activate nested quantization for 4-bit base models (double quantization)
use_nested_quant = False
################################################################################
# TrainingArguments parameters
################################################################################
# Output directory where the model predictions and checkpoints will be stored
output_dir = "./results"
# Number of training epochs
num_train_epochs = 1
# Enable fp16/bf16 training (set bf16 to True with an A100)
fp16 = True
bf16 = False
# Batch size per GPU for training
per_device_train_batch_size = int(os.getenv("TRAIN_BATCH_SIZE", "1"))
# Batch size per GPU for evaluation
per_device_eval_batch_size = int(os.getenv("EVAL_BATCH_SIZE", "2"))
# Number of update steps to accumulate the gradients for
gradient_accumulation_steps = int(os.getenv("GRADIENT_ACCUMULATION_STEPS", "1"))
# Enable gradient checkpointing
gradient_checkpointing = True
# Maximum gradient normal (gradient clipping)
max_grad_norm = 0.3
# Initial learning rate (AdamW optimizer)
learning_rate = 2e-4
# Weight decay to apply to all layers except bias/LayerNorm weights
weight_decay = 0.001
# Optimizer to use
optim = "paged_adamw_32bit"
# Learning rate schedule
lr_scheduler_type = "cosine"
# Number of training steps (overrides num_train_epochs)
max_steps = -1
# Ratio of steps for a linear warmup (from 0 to learning rate)
warmup_ratio = 0.03
# Group sequences into batches with same length
# Saves memory and speeds up training considerably
group_by_length = True
# Save checkpoint every X updates steps
save_steps = 0
# Log every X updates steps
logging_steps = int(os.getenv("LOGGING_STEPS", "50"))
################################################################################
# SFT parameters
################################################################################
# Maximum sequence length to use
max_seq_length = int(os.getenv("MAX_SEQ_LENGTH", "512"))
# Pack multiple short examples in the same input sequence to increase efficiency
packing = False
# Load the entire model on the GPU 0
device_map = {'':torch.cuda.current_device()}
# Set limit to a positive number
limit = int(os.getenv("DATASET_LIMIT", "5000"))
dataset = load_dataset(dataset_name, split="train")
if limit != -1:
dataset = dataset.shuffle(seed=42).select(range(limit))
def transform(data):
question = data['question']
context = data['context']
answer = data['answer']
template = "Question: {question}\nContext: {context}\nAnswer: {answer}"
return {'text': template.format(question=question, context=context, answer=answer)}
transformed = dataset.map(transform)
# Load tokenizer and model with QLoRA configuration
compute_dtype = getattr(torch, bnb_4bit_compute_dtype)
bnb_config = BitsAndBytesConfig(
load_in_4bit=use_4bit,
bnb_4bit_quant_type=bnb_4bit_quant_type,
bnb_4bit_compute_dtype=compute_dtype,
bnb_4bit_use_double_quant=use_nested_quant,
)
# Check GPU compatibility with bfloat16
if compute_dtype == torch.float16 and use_4bit:
major, _ = torch.cuda.get_device_capability()
if major >= 8:
print("=" * 80)
print("Your GPU supports bfloat16")
print("=" * 80)
# Load base model
# model = AutoModelForCausalLM.from_pretrained("google/gemma-7b")
model = AutoModelForCausalLM.from_pretrained(
model_name,
quantization_config=bnb_config,
device_map=device_map,
torch_dtype=torch.float16,
)
model.config.use_cache = False
model.config.pretraining_tp = 1
# Load LLaMA tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = "right" # Fix weird overflow issue with fp16 training
# Load LoRA configuration
peft_config = LoraConfig(
lora_alpha=lora_alpha,
lora_dropout=lora_dropout,
r=lora_r,
bias="none",
task_type="CAUSAL_LM",
target_modules=["q_proj", "v_proj"]
)
# Set training parameters
training_arguments = TrainingArguments(
output_dir=output_dir,
num_train_epochs=num_train_epochs,
per_device_train_batch_size=per_device_train_batch_size,
gradient_accumulation_steps=gradient_accumulation_steps,
optim=optim,
save_steps=save_steps,
logging_steps=logging_steps,
learning_rate=learning_rate,
weight_decay=weight_decay,
fp16=fp16,
bf16=bf16,
max_grad_norm=max_grad_norm,
max_steps=max_steps,
warmup_ratio=warmup_ratio,
group_by_length=group_by_length,
lr_scheduler_type=lr_scheduler_type,
)
trainer = SFTTrainer(
model=model,
train_dataset=transformed,
peft_config=peft_config,
dataset_text_field="text",
max_seq_length=max_seq_length,
tokenizer=tokenizer,
args=training_arguments,
packing=packing,
)
trainer.train()
trainer.model.save_pretrained(new_model)
# Reload model in FP16 and merge it with LoRA weights
base_model = AutoModelForCausalLM.from_pretrained(
model_name,
low_cpu_mem_usage=True,
return_dict=True,
torch_dtype=torch.float16,
device_map=device_map,
)
model = PeftModel.from_pretrained(base_model, new_model)
model = model.merge_and_unload()
# Push to HF
# model.push_to_hub(new_model, check_pr=True)
# tokenizer.push_to_hub(new_model, check_pr=True)
# push to GCS
file_path_to_save_the_model = '/finetune/new_model'
model.save_pretrained(file_path_to_save_the_model)
tokenizer.save_pretrained(file_path_to_save_the_model)
Buat file requirements.txt.
accelerate==0.30.1
bitsandbytes==0.43.1
datasets==2.19.1
transformers==4.41.0
peft==0.11.1
trl==0.8.6
torch==2.3.0
Membuat Dockerfile
FROM nvidia/cuda:12.6.2-runtime-ubuntu22.04
RUN apt-get update && \
apt-get -y --no-install-recommends install python3-dev gcc python3-pip git && \
rm -rf /var/lib/apt/lists/*
RUN pip3 install -r requirements.txt --no-cache-dir
COPY finetune.py /finetune.py
ENV PYTHONUNBUFFERED 1
CMD python3 /finetune.py --device cuda
Mem-build container di repo Artifact Registry Anda
gcloud builds submit --tag $REGION-docker.pkg.dev/$PROJECT_ID/$AR_REPO/$IMAGE_NAME
5. Men-deploy dan menjalankan tugas
Pada langkah ini, Anda akan membuat konfigurasi YAML Tugas dengan traffic keluar VPC langsung untuk upload yang lebih cepat ke Google Cloud Storage.
Perhatikan bahwa file ini berisi variabel yang akan Anda perbarui pada langkah berikutnya.
Pertama, buat file bernama finetune-job.yaml
apiVersion: run.googleapis.com/v1
kind: Job
metadata:
name: finetuning-to-gcs-job
labels:
cloud.googleapis.com/location: us-central1
annotations:
run.googleapis.com/launch-stage: ALPHA
spec:
template:
metadata:
annotations:
run.googleapis.com/execution-environment: gen2
run.googleapis.com/network-interfaces: '[{"network":"default","subnetwork":"default"}]'
spec:
parallelism: 1
taskCount: 1
template:
spec:
serviceAccountName: YOUR_SERVICE_ACCOUNT_NAME@YOUR_PROJECT_ID.iam.gserviceaccount.com
containers:
- name: finetune-to-gcs
image: YOUR_REGION-docker.pkg.dev/YOUR_PROJECT_ID/YOUR_AR_REPO/YOUR_IMAGE_NAME
env:
- name: MODEL_NAME
value: "google/gemma-2b"
- name: NEW_MODEL
value: "gemma-2b-sql-finetuned"
- name: LORA_R
value: "8"
- name: LORA_ALPHA
value: "16"
- name: TRAIN_BATCH_SIZE
value: "1"
- name: EVAL_BATCH_SIZE
value: "2"
- name: GRADIENT_ACCUMULATION_STEPS
value: "2"
- name: DATASET_LIMIT
value: "1000"
- name: MAX_SEQ_LENGTH
value: "512"
- name: LOGGING_STEPS
value: "5"
- name: HF_TOKEN
valueFrom:
secretKeyRef:
key: 'latest'
name: HF_TOKEN
resources:
limits:
cpu: 8000m
nvidia.com/gpu: '1'
memory: 32Gi
volumeMounts:
- mountPath: /finetune/new_model
name: finetuned_model
volumes:
- name: finetuned_model
csi:
driver: gcsfuse.run.googleapis.com
readOnly: false
volumeAttributes:
bucketName: YOUR_RPOJECT_ID-codelab-finetuning-jobs
maxRetries: 3
timeoutSeconds: '3600'
nodeSelector:
run.googleapis.com/accelerator: nvidia-l4
Sekarang, ganti placeholder dengan variabel lingkungan Anda untuk image dengan menjalankan perintah berikut:
sed -i "s/YOUR_SERVICE_ACCOUNT_NAME/$SERVICE_ACCOUNT/; s/YOUR_PROJECT_ID/$PROJECT_ID/; s/YOUR_PROJECT_ID/$PROJECT_ID/; s/YOUR_REGION/$REGION/; s/YOUR_AR_REPO/$AR_REPO/; s/YOUR_IMAGE_NAME/$IMAGE_NAME/; s/YOUR_PROJECT_ID/$PROJECT_ID/" finetune-job.yaml
Selanjutnya, buat Tugas Cloud Run
gcloud alpha run jobs replace finetune-job.yaml
Lalu, jalankan tugas. Proses ini akan memerlukan waktu sekitar 10 menit.
gcloud alpha run jobs execute $JOB_NAME --region $REGION
6. Menggunakan layanan Cloud Run untuk menayangkan model yang dioptimalkan dengan vLLM
Buat folder untuk kode layanan Cloud Run yang akan menayangkan model yang telah dioptimalkan
cd ..
mkdir codelab-finetuning-service
cd codelab-finetuning-service
Membuat file service.yaml
Konfigurasi ini menggunakan VPC langsung untuk mengakses bucket GCS melalui jaringan pribadi agar download lebih cepat.
Perhatikan bahwa file ini berisi variabel yang akan Anda perbarui pada langkah berikutnya.
apiVersion: serving.knative.dev/v1
kind: Service
metadata:
name: serve-gemma2b-sql
labels:
cloud.googleapis.com/location: us-central1
annotations:
run.googleapis.com/launch-stage: BETA
run.googleapis.com/ingress: all
run.googleapis.com/ingress-status: all
spec:
template:
metadata:
labels:
annotations:
autoscaling.knative.dev/maxScale: '5'
run.googleapis.com/cpu-throttling: 'false'
run.googleapis.com/network-interfaces: '[{"network":"default","subnetwork":"default"}]'
spec:
containers:
- name: serve-finetuned
image: us-docker.pkg.dev/vertex-ai/vertex-vision-model-garden-dockers/pytorch-vllm-serve:20240220_0936_RC01
ports:
- name: http1
containerPort: 8000
resources:
limits:
cpu: 8000m
nvidia.com/gpu: '1'
memory: 32Gi
volumeMounts:
- name: fuse
mountPath: /finetune/new_model
command: ["python3", "-m", "vllm.entrypoints.api_server"]
args:
- --model=/finetune/new_model
- --tensor-parallel-size=1
env:
- name: MODEL_ID
value: 'new_model'
- name: HF_HUB_OFFLINE
value: '1'
volumes:
- name: fuse
csi:
driver: gcsfuse.run.googleapis.com
volumeAttributes:
bucketName: YOUR_BUCKET_NAME
nodeSelector:
run.googleapis.com/accelerator: nvidia-l4
Perbarui file service.yaml dengan nama bucket Anda.
sed -i "s/YOUR_BUCKET_NAME/$BUCKET_NAME/" finetune-job.yaml
Sekarang, deploy Layanan Cloud Run Anda
gcloud alpha run services replace service.yaml
7. Menguji model yang telah di-fine-tune
Pertama, dapatkan URL layanan untuk layanan Cloud Run Anda.
SERVICE_URL=$(gcloud run services describe serve-gemma2b-sql --platform managed --region $REGION --format 'value(status.url)')
Buat perintah untuk model Anda.
USER_PROMPT="Question: What are the first name and last name of all candidates? Context: CREATE TABLE candidates (candidate_id VARCHAR); CREATE TABLE people (first_name VARCHAR, last_name VARCHAR, person_id VARCHAR)"
Sekarang, curl Layanan Anda
curl -X POST $SERVICE_URL/generate \
-H "Content-Type: application/json" \
-H "Authorization: bearer $(gcloud auth print-identity-token)" \
-d @- <<EOF
{
"prompt": "${USER_PROMPT}",
"temperature": 0.1,
"top_p": 1.0,
"max_tokens": 56
}
EOF
Anda akan melihat respons seperti berikut:
{"predictions":["Prompt:\nQuestion: What are the first name and last name of all candidates? Context: CREATE TABLE candidates (candidate_id VARCHAR); CREATE TABLE people (first_name VARCHAR, last_name VARCHAR, person_id VARCHAR)\nOutput:\n CREATE TABLE people_to_candidates (candidate_id VARCHAR, person_id VARCHAR) CREATE TABLE people_to_people (person_id VARCHAR, person_id VARCHAR) CREATE TABLE people_to_people_to_candidates (person_id VARCHAR, candidate_id"]}