À propos de cet atelier de programmation
1. Introduction
Présentation
Dans cet exemple, vous allez affiner un modèle gemma-2b avec un ensemble de données de conversion texte-SQL destiné à faire répondre le LLM par une requête SQL lorsqu'il reçoit une question en langage naturel. Vous allez ensuite utiliser le modèle affiné sur Cloud Run à l'aide de vLLM.
Points abordés
- Effectuer un ajustement précis à l'aide du GPU des jobs Cloud Run
- Utiliser la configuration VPC directe pour un job GPU afin d'accélérer l'importation et la diffusion du modèle
2. Avant de commencer
Pour utiliser la fonctionnalité GPU, vous devez demander une augmentation de quota pour une région compatible. Le quota requis est nvidia_l4_gpu_allocation_no_zonal_redundancy, qui se trouve dans l'API Cloud Run Admin. Voici le lien direct pour demander un quota.
3. Préparation
Définissez les variables d'environnement qui seront utilisées tout au long de cet atelier de programmation.
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
Créez le compte de service en exécutant la commande suivante:
gcloud iam service-accounts create $SERVICE_ACCOUNT \
--display-name="Cloud Run job to access HF_TOKEN Secret ID"
Utilisez Secret Manager pour stocker le jeton d'accès Hugging Face.
Pour en savoir plus sur la création et l'utilisation de secrets, consultez la documentation de Secret Manager.
gcloud secrets create $SECRET_ID \
--replication-policy="automatic"
printf $HF_TOKEN | gcloud secrets versions add $SECRET_ID --data-file=-
Un résultat semblable à ce qui suit s'affiche :
you'll see output similar to
Created secret [HF_TOKEN].
Created version [1] of the secret [HF_TOKEN].
Attribuer le rôle "Accesseur de secrets du gestionnaire de secrets" à votre compte de service Compute par défaut
gcloud secrets add-iam-policy-binding $SECRET_ID \
--member serviceAccount:$SERVICE_ACCOUNT_ADDRESS \
--role='roles/secretmanager.secretAccessor'
Créer un bucket qui hébergera votre modèle affiné
gsutil mb -l $REGION gs://$BUCKET_NAME
Accordez ensuite à l'administrateur de service l'accès au bucket.
gcloud storage buckets add-iam-policy-binding gs://$BUCKET_NAME \
--member=serviceAccount:$SERVICE_ACCOUNT_ADDRESS \
--role=roles/storage.objectAdmin
Créer un dépôt Artifact Registry pour le job
gcloud artifacts repositories create $AR_REPO \
--repository-format=docker \
--location=$REGION \
--description="codelab for finetuning using CR jobs" \
--project=$PROJECT_ID
4. Créer l'image de la tâche Cloud Run
À l'étape suivante, vous allez créer le code qui effectue les opérations suivantes:
- Importe gemma-2b depuis huggingface
- Effectue un affinage sur gemma-2b avec un ensemble de données textuelles vers SQL à l'aide de l'ensemble de données de huggingface. La tâche utilise un seul GPU L4 pour le paramétrage fin.
- Importe le modèle affiné appelé "new_model" dans le bucket GCS de l'utilisateur.
Créez un répertoire pour votre code de tâche de réglage fin.
mkdir codelab-finetuning-job
cd codelab-finetuning-job
Créez un fichier appelé 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)
Créez un fichier 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
Créer un objet 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/*
COPY requirements.txt /requirements.txt
RUN pip3 install -r requirements.txt --no-cache-dir
COPY finetune.py /finetune.py
ENV PYTHONUNBUFFERED 1
CMD python3 /finetune.py --device cuda
Créer le conteneur dans votre dépôt Artifact Registry
gcloud builds submit --tag $REGION-docker.pkg.dev/$PROJECT_ID/$AR_REPO/$IMAGE_NAME
5. Déployer et exécuter la tâche
Dans cette étape, vous allez créer la configuration YAML des jobs avec une sortie VPC directe pour des importations plus rapides dans Google Cloud Storage.
Notez que ce fichier contient des variables que vous mettrez à jour lors d'une étape ultérieure.
Commencez par créer un fichier appelé 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_PROJECT_ID-codelab-finetuning-jobs
maxRetries: 3
timeoutSeconds: '3600'
nodeSelector:
run.googleapis.com/accelerator: nvidia-l4
Remplacez maintenant les espaces réservés par vos variables d'environnement pour l'image en exécutant la commande suivante:
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
Créez ensuite le job Cloud Run.
gcloud alpha run jobs replace finetune-job.yaml
Exécutez la tâche. L'opération prendra environ 10 minutes.
gcloud alpha run jobs execute $JOB_NAME --region $REGION
6. Utiliser un service Cloud Run pour diffuser votre modèle affiné avec vLLM
Créez un dossier pour le code du service Cloud Run qui diffusera le modèle affiné.
cd ..
mkdir codelab-finetuning-service
cd codelab-finetuning-service
Créez un fichier service.yaml
.
Cette configuration utilise un VPC direct pour accéder au bucket GCS via un réseau privé afin d'accélérer les téléchargements.
Notez que ce fichier contient des variables que vous mettrez à jour lors d'une étape ultérieure.
apiVersion: serving.knative.dev/v1
kind: Service
metadata:
name: serve-gemma2b-sql
labels:
cloud.googleapis.com/location: us-central1
annotations:
run.googleapis.com/launch-stage: BETA
run.googleapis.com/ingress: all
run.googleapis.com/ingress-status: all
spec:
template:
metadata:
labels:
annotations:
autoscaling.knative.dev/maxScale: '5'
run.googleapis.com/cpu-throttling: 'false'
run.googleapis.com/network-interfaces: '[{"network":"default","subnetwork":"default"}]'
spec:
containers:
- name: serve-finetuned
image: us-docker.pkg.dev/vertex-ai/vertex-vision-model-garden-dockers/pytorch-vllm-serve:20240220_0936_RC01
ports:
- name: http1
containerPort: 8000
resources:
limits:
cpu: 8000m
nvidia.com/gpu: '1'
memory: 32Gi
volumeMounts:
- name: fuse
mountPath: /finetune/new_model
command: ["python3", "-m", "vllm.entrypoints.api_server"]
args:
- --model=/finetune/new_model
- --tensor-parallel-size=1
env:
- name: MODEL_ID
value: 'new_model'
- name: HF_HUB_OFFLINE
value: '1'
volumes:
- name: fuse
csi:
driver: gcsfuse.run.googleapis.com
volumeAttributes:
bucketName: YOUR_BUCKET_NAME
nodeSelector:
run.googleapis.com/accelerator: nvidia-l4
Remplacez le nom de votre bucket dans le fichier service.yaml
.
sed -i "s/YOUR_BUCKET_NAME/$BUCKET_NAME/" service.yaml
Déployez maintenant votre service Cloud Run
gcloud alpha run services replace service.yaml
7. Tester votre modèle affiné
Commencez par obtenir l'URL de service de votre service Cloud Run.
SERVICE_URL=$(gcloud run services describe serve-gemma2b-sql --platform managed --region $REGION --format 'value(status.url)')
Créez votre requête pour votre modèle.
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)"
Curl votre service
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
Un résultat semblable à celui-ci doit s'afficher :
{"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"]}