Skip to content

Enable AWQ example on CPU #1043

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Oct 10, 2024
Merged

Enable AWQ example on CPU #1043

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 7 additions & 6 deletions torchao/prototype/awq/example.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@ def get_calib_dataset(tokenizer=None, n_samples=100, block_size=512):
return [cat_samples[:, i * block_size : (i + 1) * block_size] for i in range(n_samples)]

# from https://github.com/mobiusml/hqq/blob/master/examples/llama2_benchmark/eval_model.py
def wiki2_eval(model, tokenizer, sequence_length, stride=512, verbose=True):
def wiki2_eval(model, tokenizer, sequence_length, stride=512, verbose=True, device="cuda"):
model.eval()
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = "right"
Expand All @@ -41,7 +41,7 @@ def wiki2_eval(model, tokenizer, sequence_length, stride=512, verbose=True):
dataset = load_dataset('wikitext', 'wikitext-2-raw-v1', split='test')
encodings = tokenizer('\n\n'.join(dataset['text']), return_tensors='pt')

encodings['input_ids'] = encodings['input_ids'].to('cuda')
encodings['input_ids'] = encodings['input_ids'].to(device)

lls, t = [], []
for i in tqdm(range(0, encodings['input_ids'].size(1), stride), disable=not verbose):
Expand All @@ -55,7 +55,8 @@ def wiki2_eval(model, tokenizer, sequence_length, stride=512, verbose=True):
t1 = time.time()
with torch.no_grad():
log_likelihood = model(input_ids, labels=target_ids).loss * trg_len
torch.cuda.synchronize()
if device.startswith("cuda"):
torch.cuda.synchronize()
t2 = time.time()
t.append((t2-t1))
lls.append(log_likelihood)
Expand All @@ -71,7 +72,7 @@ def wiki2_eval(model, tokenizer, sequence_length, stride=512, verbose=True):
return {'perplexity':ppl, 'prediction_time':pred_time}

# adapted from Hicham Badri (@mobicham)
def benchmark(model, tokenizer, max_length, tasks=None):
def benchmark(model, tokenizer, max_length, tasks=None, device="cuda"):
import numpy as np
import copy
import lm_eval
Expand All @@ -87,7 +88,7 @@ def benchmark(model, tokenizer, max_length, tasks=None):
tasks = ["PPL","truthfulqa_mc2", "winogrande", "arc_challenge", "hellaswag", "gsm8k", "mmlu"]
results = {}
if "PPL" in tasks:
results["perplexity"] = wiki2_eval(model, tokenizer, 512, verbose=True)
results["perplexity"] = wiki2_eval(model, tokenizer, 512, verbose=True, device=device)
############################################
if "truthfulqa_mc2" in tasks:
for task in [("truthfulqa_mc2", 0)]:
Expand Down Expand Up @@ -192,7 +193,7 @@ def wikitext2_ppl(
if compile:
model = torch.compile(model)

results = benchmark(model, tokenizer, sequence_length, tasks=tasks)
return benchmark(model, tokenizer, sequence_length, tasks=tasks, device=device)

if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Evaluate a model with the specified parameters.")
Expand Down
Loading