ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

Combine two versions of run_wikitext.py (#11597)

Browse source 
* Combine two versions of run_wikitext.py

* Update run_wikitext.py

* Update run_wikitext.py

* aligned the format

* update error display

* simplified argument parser

---------

Co-authored-by: jenniew <jenniewang123@gmail.com>
This commit is contained in:
RyuKosei 2024-07-29 00:56:16 -07:00 committed by GitHub
parent 6f999e6e90
commit 1da1f1dd0e
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 25 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

31
python/llm/dev/benchmark/perplexity/run_wikitext.py
View file

@ -20,16 +20,20 @@
import argparse
import torch
from tqdm import tqdm
from datasets import concatenate_datasets, load_dataset
from ipex_llm.utils.common import invalidInputError
parser = argparse.ArgumentParser()
parser.add_argument("--model_path", required=True, type=str)
parser.add_argument("--data_path", type=str, default='wikitext-2-raw-v1/wikitext-2-raw/wiki.test.raw')
parser.add_argument("--dataset", type=str, default=None)
parser.add_argument("--data_path", type=str, default=None)
parser.add_argument("--chunk_size", type=int, default=512)
parser.add_argument("--stride", type=int, default=0)
parser.add_argument("--device", type=str, default="xpu")
parser.add_argument("--precision", type=str, default="sym_int4")
parser.add_argument("--use-cache", action="store_true")
parser.add_argument("--max_length", type=int, default=None)
args = parser.parse_args()
if args.precision == "fp16": # ipex fp16
@ -46,14 +50,29 @@ else: # ipex-llm
model = model.to(args.device)
model = model.eval()
with open(args.data_path, "rb") as f:
data = f.read()
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained(args.model_path, trust_remote_code=True)
encodings = tokenizer(data.decode("utf-8").strip("\n"), return_tensors="pt")
max_length = model.config.max_position_embeddings
if args.dataset:
def parse_kwargs(kwstr):
kvpair = [item.split('=') for item in kwstr.split(',') if item != ""]
return {k:v for k, v in kvpair}
test = load_dataset(**parse_kwargs(args.dataset), split="test")["text"]
encodings = tokenizer("\n\n".join(test), return_tensors="pt")
elif args.data_path:
with open(args.data_path, "rb") as f:
data = f.read()
encodings = tokenizer(data.decode("utf-8").strip("\n"), return_tensors="pt")
else:
raise invalidInputError(False, "Must specify either dataset or datapath.")
if not args.max_length:
try:
max_length = model.config.max_position_embeddings
except:
max_length = model.config.seq_length # max_length in config of chatglm is 'seq_length'
else:
max_length = args.max_length
stride = args.chunk_size if args.stride <= 0 else args.stride
seq_len = encodings.input_ids.size(1)
num_chunks = seq_len // stride