Spaces:
Sleeping
Sleeping
File size: 9,258 Bytes
6376749 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 |
import copy
import json
import os
import re
import sys
import argparse
import fire
import torch
sys.path.append(os.path.join(os.getcwd(), "peft/src/"))
from peft import PeftModel
from tqdm import tqdm
from transformers import GenerationConfig, LlamaForCausalLM, LlamaTokenizer, AutoModelForCausalLM, AutoTokenizer
if torch.cuda.is_available():
device = "cuda"
else:
device = "cpu"
try:
if torch.backends.mps.is_available():
device = "mps"
except: # noqa: E722
pass
def main(
load_8bit: bool = False,
base_model: str = "",
lora_weights: str = "tloen/alpaca-lora-7b",
share_gradio: bool = False,
):
args = parse_args()
def evaluate(
instructions,
input=None,
temperature=0.1,
top_p=0.75,
top_k=40,
num_beams=4,
max_new_tokens=32,
**kwargs,
):
prompts = [generate_prompt(instruction, input) for instruction in instructions]
inputs = tokenizer(prompts, return_tensors="pt", padding=True)
input_ids = inputs["input_ids"].to(device)
generation_config = GenerationConfig(
temperature=temperature,
top_p=top_p,
top_k=top_k,
num_beams=num_beams,
**kwargs,
)
with torch.no_grad():
generation_output = model.generate(
input_ids=input_ids,
generation_config=generation_config,
return_dict_in_generate=True,
output_scores=True,
max_new_tokens=max_new_tokens,
)
s = generation_output.sequences
outputs = tokenizer.batch_decode(s, skip_special_tokens=True)
outputs = [o.split("### Response:")[1].strip() for o in outputs]
print(outputs)
return outputs
save_file = f'experiment/{args.model}-{args.adapter}-{args.dataset}.json'
create_dir('experiment/')
dataset = load_data(args)
batches = create_batch(dataset, args.batch_size)
tokenizer, model = load_model(args)
total = len(batches)
correct = 0
current = 0
output_data = []
pbar = tqdm(total=total)
for idx, batch in enumerate(batches):
current += len(batch)
instructions = [data.get('instruction') for data in batch]
outputs = evaluate(instructions)
for data, output in zip(batch, outputs):
label = data.get('answer')
flag = False
predict = extract_answer(args, output)
if label == predict:
correct += 1
flag = True
new_data = copy.deepcopy(data)
new_data['output_pred'] = output
new_data['pred'] = predict
new_data['flag'] = flag
output_data.append(new_data)
print(data["instruction"])
print(output)
print('prediction:', predict)
print('label:', label)
print('---------------')
print(f'\rtest:{idx + 1}/{total} | accuracy {correct} {correct / current}')
print('---------------')
with open(save_file, 'w+') as f:
json.dump(output_data, f, indent=4)
pbar.update(1)
pbar.close()
print('\n')
print('test finished')
def create_dir(dir_path):
if not os.path.exists(dir_path):
os.mkdir(dir_path)
return
def generate_prompt(instruction, input=None):
if 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:
{instruction}
### Input:
{input}
### Response:
""" # noqa: E501
else:
return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.
### Instruction:
{instruction}
### Response:
""" # noqa: E501
def load_data(args) -> list:
"""
read data from dataset file
Args:
args:
Returns:
"""
file_path = f'dataset/{args.dataset}/test.json'
if not os.path.exists(file_path):
raise FileNotFoundError(f"can not find dataset file : {file_path}")
json_data = json.load(open(file_path, 'r'))
return json_data
def create_batch(dataset, batch_size):
batches = []
num_batch = len(dataset)//batch_size if len(dataset) % batch_size == 0 else len(dataset)//batch_size + 1
for i in range(num_batch):
batch = dataset[i*batch_size: min((i+1)*batch_size, len(dataset))]
batches.append(batch)
return batches
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument('--dataset', choices=["boolq", "piqa", "social_i_qa", "hellaswag", "winogrande", "ARC-Challenge", "ARC-Easy", "openbookqa"],
required=True)
parser.add_argument('--model', choices=['LLaMA-7B', "LLaMA-13B",'BLOOM-7B', 'GPT-j-6B'], required=True)
parser.add_argument('--adapter', choices=['LoRA', 'AdapterP', 'AdapterH', 'Parallel'],
required=True)
parser.add_argument('--base_model', required=True)
parser.add_argument('--lora_weights', required=True)
parser.add_argument('--batch_size', type=int, required=True)
parser.add_argument('--load_8bit', action='store_true', default=False)
return parser.parse_args()
def load_model(args) -> tuple:
"""
load tuned model
Args:
args:
Returns:
tuple(tokenizer, model)
"""
base_model = args.base_model
if not base_model:
raise ValueError(f'can not find base model name by the value: {args.model}')
lora_weights = args.lora_weights
if not lora_weights:
raise ValueError(f'can not find lora weight, the value is: {lora_weights}')
load_8bit = args.load_8bit
if "LLaMA" in args.model:
tokenizer = LlamaTokenizer.from_pretrained(base_model)
else:
tokenizer = AutoTokenizer.from_pretrained(base_model)
tokenizer.padding_side = "left"
tokenizer.pad_token_id = (
0 # unk. we want this to be different from the eos token
)
if device == "cuda":
model = AutoModelForCausalLM.from_pretrained(
base_model,
load_in_8bit=load_8bit,
torch_dtype=torch.float16,
device_map="auto",
trust_remote_code=True,
) # fix zwq
model = PeftModel.from_pretrained(
model,
lora_weights,
torch_dtype=torch.float16,
device_map={"":0}
)
elif device == "mps":
model = AutoModelForCausalLM.from_pretrained(
base_model,
device_map={"": device},
torch_dtype=torch.float16,
)
model = PeftModel.from_pretrained(
model,
lora_weights,
device_map={"": device},
torch_dtype=torch.float16,
)
else:
model = AutoModelForCausalLM.from_pretrained(
base_model, device_map={"": device}, low_cpu_mem_usage=True
)
model = PeftModel.from_pretrained(
model,
lora_weights,
device_map={"": device},
)
# unwind broken decapoda-research config
model.config.pad_token_id = tokenizer.pad_token_id = 0 # unk
model.config.bos_token_id = 1
model.config.eos_token_id = 2
if not load_8bit:
model.half() # seems to fix bugs for some users.
model.eval()
if torch.__version__ >= "2" and sys.platform != "win32":
model = torch.compile(model)
return tokenizer, model
def load_instruction(args) -> str:
instruction = ''
if not instruction:
raise ValueError('instruct not initialized')
return instruction
def extract_answer(args, sentence: str) -> float:
dataset = args.dataset
if dataset == 'boolq':
sentence_ = sentence.strip()
pred_answers = re.findall(r'true|false', sentence_)
if not pred_answers:
return ""
return pred_answers[0]
elif dataset == 'piqa':
sentence_ = sentence.strip()
pred_answers = re.findall(r'solution1|solution2', sentence_)
if not pred_answers:
return ""
return pred_answers[0]
elif dataset in ['social_i_qa', 'ARC-Challenge', 'ARC-Easy', 'openbookqa']:
sentence_ = sentence.strip()
pred_answers = re.findall(r'answer1|answer2|answer3|answer4|answer5', sentence_)
if not pred_answers:
return ""
return pred_answers[0]
elif dataset == 'hellaswag':
sentence_ = sentence.strip()
pred_answers = re.findall(r'ending1|ending2|ending3|ending4', sentence_)
if not pred_answers:
return ""
return pred_answers[0]
elif dataset == 'winogrande':
sentence_ = sentence.strip()
pred_answers = re.findall(r'option1|option2', sentence_)
if not pred_answers:
return ""
return pred_answers[0]
if __name__ == "__main__":
main()
|