Skip to content
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

Improving hf_eval.py #342

Merged
merged 1 commit into from
Jun 11, 2024
Merged

Improving hf_eval.py #342

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
24 changes: 15 additions & 9 deletions scripts/hf_eval.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,10 +16,10 @@
torch._inductor.config.force_fuse_int_mm_with_mul = True
torch._inductor.config.fx_graph_cache = True

def run_evaluation(repo_id, task_list, limit, device, precision, quantization, compile):
def run_evaluation(repo_id, task_list, limit, device, precision, quantization, compile, batch_size, max_length):

tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForCausalLM.from_pretrained(repo_id).to(device="cuda", dtype=precision)
model = AutoModelForCausalLM.from_pretrained(repo_id).to(device="cpu", dtype=precision)

if compile:
torch.compile(model, mode="max-autotune", fullgraph=True)
Expand All @@ -29,21 +29,25 @@ def run_evaluation(repo_id, task_list, limit, device, precision, quantization, c
elif quantization == "int8wo":
change_linear_weights_to_int8_woqtensors(model)
elif quantization == "int4wo":
change_linear_weights_to_int4_woqtensors(model)
# note cannot quantize this model on cpu and run it on cuda at this time
change_linear_weights_to_int4_woqtensors(model.to(device=device))
elif quantization == "autoquant":
model = autoquant(model)
model = autoquant(model.to(device=device))

with torch.no_grad():
result = evaluate(
HFLM(pretrained=model, tokenizer=tokenizer),
HFLM(
pretrained=model.to(device),
tokenizer=tokenizer,
batch_size=batch_size,
max_length=max_length),
get_task_dict(task_list),
limit = limit
limit = limit,
)
for task, res in result["results"].items():
print(f"{task}: {res}")



if __name__ == '__main__':
import argparse
parser = argparse.ArgumentParser(description='Run HF Model Evaluation')
Expand All @@ -52,8 +56,10 @@ def run_evaluation(repo_id, task_list, limit, device, precision, quantization, c
parser.add_argument('--limit', type=int, default=None, help='Number of eval samples to evaluate')
parser.add_argument('--precision', type=lambda x: getattr(torch, x.split(".")[-1]), default=torch.bfloat16, help='dtype precision to use')
parser.add_argument('--device', type=str, default="cuda", help='Device to use for evaluation')
parser.add_argument('--quantization', default = "None", choices=["int8dq", "int8wo", "int4wo","autoquant", "None"], help='Which quantization technique to apply')
parser.add_argument('-q', '--quantization', default = "None", choices=["int8dq", "int8wo", "int4wo","autoquant", "None"], help='Which quantization technique to apply')
parser.add_argument('--compile', action='store_true', help='Whether to compile the model.')
parser.add_argument('--batch_size', type=int, default=1, help='Batch size to use for evaluation, note int8wo and int4wo work best with small batchsizes, int8dq works better with large batchsizes')
parser.add_argument('--max_length', type=int, default=None, help='Length of text to process at one time')

args = parser.parse_args()
run_evaluation(args.repo_id, args.task_list, args.limit, args.device, args.precision, args.quantization, args.compile)
run_evaluation(args.repo_id, args.task_list, args.limit, args.device, args.precision, args.quantization, args.compile, args.batch_size, args.max_length)
Loading