Open zihaohe123 opened 9 months ago
What llama model size are you using?
NousResearch/Llama-2-7b-hf
Did you make any other change to the code than the model id? What GPU are you using?
Below is the code I was using. I don't think I made any changes from yours. And I was using an A40.
from datasets import load_dataset
from random import randrange
# Load dataset from the hub
dataset = load_dataset("databricks/databricks-dolly-15k", split="train")
print(f"dataset size: {len(dataset)}")
print(dataset[randrange(len(dataset))])
# dataset size: 15011
def format_instruction(sample):
return f"""### Instruction:
Use the Input below to create an instruction, which could have been used to generate the input using an LLM.
### Input:
{sample['response']}
### Response:
{sample['instruction']}
"""
from random import randrange
print(format_instruction(dataset[randrange(len(dataset))]))
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
use_flash_attention = False
# COMMENT IN TO USE FLASH ATTENTION
# replace attention with flash attention
if torch.cuda.get_device_capability()[0] >= 8:
from utils.llama_patch import replace_attn_with_flash_attn
print("Using flash attention")
replace_attn_with_flash_attn()
use_flash_attention = True
# Hugging Face model id
model_id = "NousResearch/Llama-2-7b-hf" # non-gated
# model_id = "meta-llama/Llama-2-7b-hf" # gated
# BitsAndBytesConfig int-4 config
bnb_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_use_double_quant=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_compute_dtype=torch.bfloat16
)
# Load model and tokenizer
model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=bnb_config, use_cache=False,
device_map="auto")
model.config.pretraining_tp = 1
# Validate that the model is using flash attention, by comparing doc strings
if use_flash_attention:
from utils.llama_patch import forward
assert model.model.layers[0].self_attn.forward.__doc__ == forward.__doc__, "Model is not using flash attention"
tokenizer = AutoTokenizer.from_pretrained(model_id)
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = "right"
from peft import LoraConfig, prepare_model_for_kbit_training, get_peft_model
# LoRA config based on QLoRA paper
peft_config = LoraConfig(
lora_alpha=16,
lora_dropout=0.1,
r=64,
bias="none",
task_type="CAUSAL_LM",
)
# prepare model for training
model = prepare_model_for_kbit_training(model)
model = get_peft_model(model, peft_config)
from transformers import TrainingArguments
args = TrainingArguments(
output_dir="llama-7-int4-dolly",
num_train_epochs=3,
per_device_train_batch_size=6 if use_flash_attention else 8,
gradient_accumulation_steps=2,
gradient_checkpointing=True,
optim="paged_adamw_32bit",
logging_steps=10,
save_strategy="epoch",
learning_rate=2e-4,
bf16=True,
fp16=False,
tf32=True,
max_grad_norm=0.3,
warmup_ratio=0.03,
lr_scheduler_type="constant",
disable_tqdm=True # disable tqdm since with packing values are in correct
)
from trl import SFTTrainer
max_seq_length = 2048 # max sequence length for model and packing of the dataset
trainer = SFTTrainer(
model=model,
train_dataset=dataset,
peft_config=peft_config,
max_seq_length=max_seq_length,
tokenizer=tokenizer,
packing=True,
formatting_func=format_instruction,
args=args,
)
# train
trainer.train() # there will not be a progress bar since tqdm is disabled
# save model
trainer.save_model()
Hi Philipp!
Thanks for this great repo!
I was trying to run llama2 instruction tuning following the tutorial. The code went well without flash attention. But after I commented in flash attention, I got this error:
"Runtime Error: Flash Attention only support fp16 and bf16 data type", from line 87 in llama_patch.py.
It seems to have something to do with data precision. Could you help me figure it out? Thanks a lot!