File size: 11,439 Bytes
96c53cc | 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 | import os
import re
import json
import argparse
import random
from tqdm import tqdm
from datasets import load_from_disk
import torch
from src.mola_peft_model_hacked import PeftModel
from transformers import GenerationConfig, LlamaTokenizer, AutoConfig
import sys
from typing import Union
from src.mola_modeling_llama_hacked import LlamaForCausalLM_d
from transformers import AutoTokenizer
from src.mola_modeling_mistral_hacked import MistralForCausalLM_d
from src.mistralconfig import MistralConfig
from src.gemma_config import GemmaConfig
from src.mola_modeling_gemma import GemmaForCausalLM_d
if torch.cuda.is_available():
device = "cuda"
else:
device = "cpu"
try:
if torch.backends.mps.is_available():
device = "mps"
except: # noqa: E722
pass
seed = 10
random.seed(seed)
torch.manual_seed(0)
class Prompter(object):
__slots__ = ("template", "_verbose")
def __init__(self, template_name: str = "", verbose: bool = False):
self._verbose = verbose
self.template = {
"description": "Template used by Alpaca-LoRA.",
"prompt_input": "Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.\n\n### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n",
"prompt_no_input": "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n{instruction}\n\n### Response:\n",
"response_split": "### Response:"
}
if self._verbose:
print(
f"Using prompt template {template_name}: {self.template['description']}"
)
def generate_prompt(
self,
instruction: str,
input: Union[None, str] = None,
label: Union[None, str] = None,
) -> str:
# returns the full prompt from instruction and optional input
# if a label (=response, =output) is provided, it's also appended.
if input:
res = self.template["prompt_input"].format(
instruction=instruction, input=input
)
else:
res = self.template["prompt_no_input"].format(
instruction=instruction
)
if label:
res = f"{res}{label}"
if self._verbose:
print(res)
return res
def get_response(self, output: str) -> str:
return output.split(self.template["response_split"])[1].strip()
def main():
parser = argparse.ArgumentParser(description='Evaluation')
# Defining arguments
parser.add_argument('--test_dataset', type=str, default="./scienceqa/scienceq_test.json",
help='test_dataset')
parser.add_argument('--base_model', type=str, default="NousResearch/Llama-2-7b-hf", help='base_model')
parser.add_argument('--mola_weights', type=str, default="./scienceqa_mola",
help='mola_model')
parser.add_argument('--number_experts', type=str,
default="2,2,2,2,2,2,2,2,4,4,4,4,4,4,4,4,6,6,6,6,6,6,6,6,8,8,8,8,8,8,8,8",
help='experts number')
parser.add_argument('--top_k', type=str,
default="2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2",
help='lora_model')
parser.add_argument('--save_path', type=str,
default="./results/mola_test_sciqa_seed_10.json",
help='lora_model')
parser.add_argument('--lora_target_modules', type=str,
default="q_proj,v_proj,k_proj,o_proj,gate_proj,down_proj,up_proj", help='lora_target_modules')
parser.add_argument('--batch_size', type=int, default=8, help='base_model')
# Parsing arguments
args = parser.parse_args()
if args.test_dataset.endswith(".json") or args.test_dataset.endswith(".jsonl"):
data_a = json.load(open(args.test_dataset))
else:
data_aa = load_from_disk(args.test_dataset)["test"]
data_a = [d for d in data_aa]
base_model = args.base_model
mola_weights = args.mola_weights
max_batch_size = args.batch_size
lora_target_modules = args.lora_target_modules.split(",")
lora_target_modules = [str(lr) for lr in lora_target_modules]
print(lora_target_modules)
number_experts = args.number_experts.split(",")
number_experts = [int(lr) for lr in number_experts]
print(number_experts)
top_k = args.top_k.split(",")
top_k = [int(lr) for lr in top_k]
print(top_k)
print(args.test_dataset)
print(args.base_model)
print(args.mola_weights)
load_8bit = False
if "Llama" in base_model or "llama" in base_model:
tokenizer = LlamaTokenizer.from_pretrained(base_model, padding_side='left')
config = AutoConfig.from_pretrained(base_model)
config.lora_target_modules = lora_target_modules
if device == "cuda":
model = LlamaForCausalLM_d.from_pretrained(
base_model,
config=config,
load_in_8bit=load_8bit,
torch_dtype=torch.float16,
device_map="auto",
)
model = PeftModel.from_pretrained(
model,
mola_weights,
torch_dtype=torch.float16,
number_experts=number_experts,
top_k=top_k,
)
else:
model = LlamaForCausalLM_d.from_pretrained(
base_model, config=config, device_map={"": device}, low_cpu_mem_usage=True
)
model = PeftModel.from_pretrained(
model,
mola_weights,
device_map={"": device},
)
elif "Mistral" in base_model:
tokenizer = AutoTokenizer.from_pretrained(base_model,padding_side='left')
config = MistralConfig.from_pretrained(base_model)
config.lora_target_modules = lora_target_modules
model = MistralForCausalLM_d.from_pretrained(
base_model,
config=config,
load_in_8bit=False,
torch_dtype=torch.float16,
device_map="auto"
)
model = PeftModel.from_pretrained(
model,
mola_weights,
torch_dtype=torch.float16,
number_experts=number_experts,
top_k=top_k,
)
else:
print("Not support")
obalance = False
model.get_new_parameters(number_experts, top_k, obalance)
print(model.config.pad_token_id, tokenizer.pad_token_id)
print(model.config.bos_token_id, tokenizer.bos_token_id)
print(model.config.eos_token_id, tokenizer.eos_token_id)
# 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)
prompter = Prompter(template_name="alpaca")
max_new_tokens = 128
save_every = 200
correct = 0
results = []
outputs = []
gt = []
for start_idx in tqdm(range(0, len(data_a), max_batch_size)):
end_idx = min(start_idx + max_batch_size, len(data_a))
batch = data_a[start_idx:end_idx]
answers = [str(example["answer"]) for example in batch]
# generate prompt
prompts = [prompter.generate_prompt(example['instruction'], example['input']) for example in batch]
inputs = tokenizer(prompts, padding=True, return_tensors="pt")
input_ids = inputs["input_ids"].to(device)
with torch.no_grad():
generation_output = model.generate(
input_ids=input_ids,
return_dict_in_generate=True,
output_scores=True,
max_new_tokens=max_new_tokens,
)
s = generation_output.sequences
output = tokenizer.batch_decode(s)
output = [prompter.get_response(otp) for otp in output]
# extract the answer
print(output)
# TODO: Here we use different pattern to extract the answer for different datasets
# pattern = re.compile(r'The anwser to the question is (\d+):*')
if "cola" in args.test_dataset or "mrpc" in args.test_dataset or "rte" in args.test_dataset:
pattern = re.compile(r'Answer: ([\w\s]+)')
res = [pattern.search(otp).group(1) for otp in output]
else:
pattern = re.compile(r'The answer is ([A-Z]).')
res = [pattern.findall(otp) for otp in output]
#-------------------------------------------------------------------------------------------------------
print(res)
pred = []
for r_i in range(len(res)):
#TODO: for diiferent datasets, we need to use different pattern to match the answer
if "rte" in args.test_dataset:
if res[r_i] == "entailment" or res[r_i] == "not entailment":
answer = res[r_i] # 'A', 'B', ...
else:
answer = "FAILED"
print(res[r_i])
elif "mrpc" in args.test_dataset:
if res[r_i] == "equivalent" or res[r_i] == "not equivalent":
answer = res[r_i] # 'A', 'B', ...
else:
answer = "FAILED"
print(res[r_i])
elif "cola" in args.test_dataset:
if res[r_i] == "acceptable" or res[r_i] == "unacceptable":
answer = res[r_i] # 'A', 'B', ...
else:
answer = "FAILED"
print(res[r_i])
else:
if len(res[r_i]) == 1:
answer = res[r_i][0]
#-------------------------------------------------------------------------------------------------------
pred.append(answer)
results.append(res[r_i])
outputs.append(output[r_i])
gt.append(answers[r_i])
if str(answer) == str(answers[r_i]):
correct += 1
print('correct:', str(answer), str(answers[r_i]))
else:
print('gt-ans:', str(answer), str(answers[r_i]))
acc = correct / len(results) * 100
if end_idx % save_every == 0 or end_idx == len(data_a):
result_file = args.save_path
os.makedirs(result_file, exist_ok=True)
print(f"{len(results)}/{len(data_a)}, correct: {correct}, acc: {round(acc, 2)}%, saving to {result_file}")
data = {}
data['acc'] = acc
data['correct'] = correct
data['len'] = len(results)
data['results'] = results
data['outputs'] = outputs
with open(result_file, 'w') as f:
json.dump(data, f, indent=2, separators=(',', ': '))
if __name__ == "__main__":
main() |