Error in finetuning
import os
import sys
from typing import List
import re
import fire
import torch
import torch.nn as nn
import bitsandbytes as bnb
from datasets import load_dataset
import transformers
from peft import LoraConfig, PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, BitsAndBytesConfig
from peft import (
prepare_model_for_int8_training,
LoraConfig,
get_peft_model,
get_peft_model_state_dict,
)
def train(
# model/data params
base_model: str = "", # the only required argument
data_path: str = "./alpaca_data_cleaned.json",
output_dir: str = "./lora-alpaca",
# training hyperparams
batch_size: int = 128,
micro_batch_size: int = 4,
num_epochs: int = 3,
learning_rate: float = 3e-4,
cutoff_len: int = 512,
val_set_size: int = 0,
# lora hyperparams
lora_r: int = 8,
lora_alpha: int = 16,
lora_dropout: float = 0.05,
lora_target_modules: List[str] = [
"q_proj",
"v_proj",
],
# llm hyperparams
train_on_inputs: bool = True, # if False, masks out inputs in loss
group_by_length: bool = True, # faster, but produces an odd training loss curve
# other
mask: bool = False,
):
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"lora_r: {lora_r}\n"
f"lora_alpha: {lora_alpha}\n"
f"lora_dropout: {lora_dropout}\n"
f"lora_target_modules: {lora_target_modules}\n"
f"train_on_inputs: {train_on_inputs}\n"
f"group_by_length: {group_by_length}\n"
)
assert (
base_model
), "Please specify a --base_model, e.g. --base_model='decapoda-research/llama-7b-hf'"
gradient_accumulation_steps = batch_size // micro_batch_size
#quantization_config = BitsAndBytesConfig(llm_int8_enable_fp32_cpu_offload=True,load_in_4bit=True, bnb_4bit_compute_dtype=torch.float16)
device_map = "auto"
world_size = int(os.environ.get("WORLD_SIZE", 1))
ddp = world_size != 1
if ddp:
device_map = {"": int(os.environ.get("LOCAL_RANK") or 0)}
gradient_accumulation_steps = gradient_accumulation_steps // world_size
model = AutoModelForCausalLM.from_pretrained(
base_model,
return_dict=True,
torch_dtype=torch.float32,
device_map=device_map
)
tokenizer = AutoTokenizer.from_pretrained(base_model)
tokenizer.pad_token_id = 0 # unk. we want this to be different from the eos token
tokenizer.padding_side = "left" # Allow batched inference
def tokenize(prompt, add_eos_token=True):
# there's probably a way to do this with the tokenizer settings
# but again, gotta move fast
if mask:
print('Masking Obersevation')
tokenizer.mask_token = "~"
# Split the input text into lines
lines = prompt.split('\n')
# Initialize an empty list to store the modified lines
masked_lines = []
# Initialize a flag to indicate if we are between "Observation:" and "Thought:"
between_observation_and_thought = False
# Iterate through each line
for line in lines:
if "Observation:" in line:
between_observation_and_thought = True
#split the line and mask all but the first word
line = line.split()
line[1:] = [tokenizer.mask_token] * len(line[1:])
line = " ".join(line)
masked_lines.append(line) # Add the line as-is
else:
masked_lines.append(line) # Add the line as-is
# Concatenate the modified lines to form the masked text
masked_text = '\n'.join(masked_lines)
prompt = masked_text
result = tokenizer(
prompt,
truncation=True,
max_length=cutoff_len,
padding=False,
return_tensors=None
)
if (
result["input_ids"][-1] != tokenizer.eos_token_id
and len(result["input_ids"]) < cutoff_len
and add_eos_token
):
result["input_ids"].append(tokenizer.eos_token_id)
result["attention_mask"].append(1)
else:
if len(result["input_ids"]) >= cutoff_len:
print("WARNING: input too long, truncating")
masked_token_id = tokenizer.mask_token_id
ids = [-100 if token_id == 3695 else token_id for token_id in result["input_ids"]]
result["labels"] = result["input_ids"].copy()
return result
def generate_and_tokenize_prompt(data_point):
full_prompt = generate_prompt(data_point)
tokenized_full_prompt = tokenize(full_prompt)
if not train_on_inputs:
user_prompt = generate_prompt({**data_point, "output": ""})
tokenized_user_prompt = tokenize(user_prompt, add_eos_token=False)
user_prompt_len = len(tokenized_user_prompt["input_ids"])
tokenized_full_prompt["labels"] = [
-100
] * user_prompt_len + tokenized_full_prompt["labels"][
user_prompt_len:
] # could be sped up, probably
return tokenized_full_prompt
model = prepare_model_for_int8_training(model)
config = LoraConfig(
r=lora_r,
lora_alpha=lora_alpha,
target_modules=lora_target_modules,
lora_dropout=lora_dropout,
bias="none",
task_type="CAUSAL_LM",
)
model = get_peft_model(model, config)
data = load_dataset("json", data_files=data_path)
if val_set_size > 0:
train_val = data["train"].train_test_split(
test_size=val_set_size, shuffle=True, seed=42
)
train_data = train_val["train"].shuffle().map(generate_and_tokenize_prompt)
val_data = train_val["test"].shuffle().map(generate_and_tokenize_prompt)
else:
train_data = data["train"].shuffle().map(generate_and_tokenize_prompt)
val_data = None
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,
warmup_steps=100,
num_train_epochs=num_epochs,
learning_rate=learning_rate,
logging_steps=10,
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,
save_total_limit=3,
load_best_model_at_end=True if val_set_size > 0 else False,
ddp_find_unused_parameters=False if ddp else None,
group_by_length=group_by_length,
),
data_collator=transformers.DataCollatorForSeq2Seq(
tokenizer, pad_to_multiple_of=8, return_tensors="pt", padding=True
),
)
model.config.use_cache = False
if torch.__version__ >= "2" and sys.platform != "win32":
model = torch.compile(model)
trainer.train()
model.save_pretrained(output_dir)
merge_model = AutoModelForCausalLM.from_pretrained(
base_model,
low_cpu_mem_usage=True,
return_dict=True,
torch_dtype=torch.float16,
device_map=device_map,
)
model = PeftModel.from_pretrained(merge_model, output_dir)
model = model.merge_and_unload()
# Reload tokenizer to save it
tokenizer = AutoTokenizer.from_pretrained(base_model, trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = "right"
model.push_to_hub(output_dir, use_temp_dir=False)
tokenizer.push_to_hub(output_dir, use_temp_dir=False)
print("\n If there's a warning about missing keys above, please disregard :)")
while True:
prompt = input("Enter a prompt: ")
if prompt=="exit":
break
pipe = pipeline(task="text-generation", model=model, tokenizer=tokenizer, max_length=200)
result = pipe(f"<s>[INST] {prompt} [/INST]")
print(result[0]['generated_text'])
print("You can end the session by typing exit")
def generate_prompt(data_point):
print(data_point)
# sorry about the formatting disaster gotta move fast
if data_point["input"]:
return f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
Instruction:
{data_point["instruction"]}
Input:
{data_point["input"]}
Response:
{data_point["output"]}"""
else:
return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.
Instruction:
{data_point["instruction"]}
Response:
{data_point["output"]}"""
if name == "main":
fire.Fire(train)
Above is my finetuning code. I am trying to finetune meta-llama/Llama-2-13b-chat-hf, but running into error:
self.is_model_parallel = self.args.device != torch.device(devices[0])
IndexError: list index out of range
When I remove device_map="auto", there are other error line NotImplementedError: Cannot copy out of meta tensor; no data!
And when I set device_map completely on GPU, using {"":0}, I am getting error:RuntimeError: No CUDA GPUs are available
Is there any particular solution around it as I am new to all this. I am running this in gcp g2-standard-12 VM with nvidia L4 GPU. GPU memory is maybe around 24gb