QLora

Finetune Llama-2-7b on a Google colab¶

Welcome to this Google Colab notebook that shows how to fine-tune the recent Llama-2-7b model on a single Google colab and turn it into a chatbot

We will leverage PEFT library from Hugging Face ecosystem, as well as QLoRA for more memory efficient finetuning

Setup¶

Run the cells below to setup and install the required libraries. For our experiment we will need accelerate, peft, transformers, datasets and TRL to leverage the recent SFTTrainer. We will use bitsandbytes to quantize the base model into 4bit. We will also install einops as it is a requirement to load Falcon models.

In [ ]:

!pip install -q -U trl transformers accelerate git+https://github.com/huggingface/peft.git
!pip install -q datasets bitsandbytes einops wandb

!pip install -q -U trl transformers accelerate git+https://github.com/huggingface/peft.git !pip install -q datasets bitsandbytes einops wandb

Dataset¶

In [ ]:

from datasets import load_dataset

#dataset_name = "timdettmers/openassistant-guanaco" ###Human ,.,,,,,, ###Assistant

dataset_name = 'AlexanderDoria/novel17_test' #french novels
dataset = load_dataset(dataset_name, split="train")

from datasets import load_dataset #dataset_name = "timdettmers/openassistant-guanaco" ###Human ,.,,,,,, ###Assistant dataset_name = 'AlexanderDoria/novel17_test' #french novels dataset = load_dataset(dataset_name, split="train")

Loading the model¶

In [ ]:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, AutoTokenizer

model_name = "TinyPixel/Llama-2-7B-bf16-sharded"

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.float16,
)

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    quantization_config=bnb_config,
    trust_remote_code=True
)
model.config.use_cache = False

import torch from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, AutoTokenizer model_name = "TinyPixel/Llama-2-7B-bf16-sharded" bnb_config = BitsAndBytesConfig( load_in_4bit=True, bnb_4bit_quant_type="nf4", bnb_4bit_compute_dtype=torch.float16, ) model = AutoModelForCausalLM.from_pretrained( model_name, quantization_config=bnb_config, trust_remote_code=True ) model.config.use_cache = False

Let's also load the tokenizer below

In [ ]:

tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token

tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True) tokenizer.pad_token = tokenizer.eos_token

In [ ]:

from peft import LoraConfig, get_peft_model

lora_alpha = 16
lora_dropout = 0.1
lora_r = 64

peft_config = LoraConfig(
    lora_alpha=lora_alpha,
    lora_dropout=lora_dropout,
    r=lora_r,
    bias="none",
    task_type="CAUSAL_LM"
)

from peft import LoraConfig, get_peft_model lora_alpha = 16 lora_dropout = 0.1 lora_r = 64 peft_config = LoraConfig( lora_alpha=lora_alpha, lora_dropout=lora_dropout, r=lora_r, bias="none", task_type="CAUSAL_LM" )

Loading the trainer¶

Here we will use the SFTTrainer from TRL library that gives a wrapper around transformers Trainer to easily fine-tune models on instruction based datasets using PEFT adapters. Let's first load the training arguments below.

In [ ]:

from transformers import TrainingArguments

output_dir = "./results"
per_device_train_batch_size = 4
gradient_accumulation_steps = 4
optim = "paged_adamw_32bit"
save_steps = 100
logging_steps = 10
learning_rate = 2e-4
max_grad_norm = 0.3
max_steps = 100
warmup_ratio = 0.03
lr_scheduler_type = "constant"

training_arguments = TrainingArguments(
    output_dir=output_dir,
    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,
    fp16=True,
    max_grad_norm=max_grad_norm,
    max_steps=max_steps,
    warmup_ratio=warmup_ratio,
    group_by_length=True,
    lr_scheduler_type=lr_scheduler_type,
)

from transformers import TrainingArguments output_dir = "./results" per_device_train_batch_size = 4 gradient_accumulation_steps = 4 optim = "paged_adamw_32bit" save_steps = 100 logging_steps = 10 learning_rate = 2e-4 max_grad_norm = 0.3 max_steps = 100 warmup_ratio = 0.03 lr_scheduler_type = "constant" training_arguments = TrainingArguments( output_dir=output_dir, 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, fp16=True, max_grad_norm=max_grad_norm, max_steps=max_steps, warmup_ratio=warmup_ratio, group_by_length=True, lr_scheduler_type=lr_scheduler_type, )

Then finally pass everthing to the trainer

In [ ]:

from trl import SFTTrainer

max_seq_length = 512

trainer = SFTTrainer(
    model=model,
    train_dataset=dataset,
    peft_config=peft_config,
    dataset_text_field="text",
    max_seq_length=max_seq_length,
    tokenizer=tokenizer,
    args=training_arguments,
)

from trl import SFTTrainer max_seq_length = 512 trainer = SFTTrainer( model=model, train_dataset=dataset, peft_config=peft_config, dataset_text_field="text", max_seq_length=max_seq_length, tokenizer=tokenizer, args=training_arguments, )

We will also pre-process the model by upcasting the layer norms in float 32 for more stable training

In [ ]:

for name, module in trainer.model.named_modules():
    if "norm" in name:
        module = module.to(torch.float32)

for name, module in trainer.model.named_modules(): if "norm" in name: module = module.to(torch.float32)

Train the model¶

Now let's train the model! Simply call trainer.train()

In [ ]:

trainer.train()

trainer.train()

During training, the model should converge nicely as follows:

The SFTTrainer also takes care of properly saving only the adapters during training instead of saving the entire model.

In [ ]:

model_to_save = trainer.model.module if hasattr(trainer.model, 'module') else trainer.model  # Take care of distributed/parallel training
model_to_save.save_pretrained("outputs")

model_to_save = trainer.model.module if hasattr(trainer.model, 'module') else trainer.model # Take care of distributed/parallel training model_to_save.save_pretrained("outputs")

In [ ]:

lora_config = LoraConfig.from_pretrained('outputs')
model = get_peft_model(model, lora_config)

lora_config = LoraConfig.from_pretrained('outputs') model = get_peft_model(model, lora_config)

In [ ]:

dataset['text']

dataset['text']

In [ ]:

text = "Écrire un texte dans un style baroque sur la glace et le feu ### Assistant: Si j'en luis éton"
device = "cuda:0"

inputs = tokenizer(text, return_tensors="pt").to(device)
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

text = "Écrire un texte dans un style baroque sur la glace et le feu ### Assistant: Si j'en luis éton" device = "cuda:0" inputs = tokenizer(text, return_tensors="pt").to(device) outputs = model.generate(**inputs, max_new_tokens=50) print(tokenizer.decode(outputs[0], skip_special_tokens=True))

In [ ]:

from huggingface_hub import login
login()

from huggingface_hub import login login()

In [ ]:

model.push_to_hub("llama2-qlora-finetunined-french")

model.push_to_hub("llama2-qlora-finetunined-french")

In [ ]: