Skip to content

Karol/llama int8 #229

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

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Karol/llama int8 #229

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
8f6435f
Support for int8 gptq models
kkontny Mar 1, 2024
9db9c3f
Add alpaca int8
kkontny Mar 1, 2024
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
9 changes: 6 additions & 3 deletions natural_language_processing/text_generation/alpaca/run.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@
from utils.benchmark import run_model


def run_pytorch(model_path, num_runs, timeout, dataset_path, use_torch_fp16=False):
def run_pytorch(model_path, batch_size, num_runs, timeout, dataset_path, use_torch_fp16=False, revision=None):
from transformers import AutoModelForCausalLM, AutoTokenizer

def run_single_pass(pytorch_runner, _dataset):
Expand All @@ -13,7 +13,7 @@ def run_single_pass(pytorch_runner, _dataset):
response = decode(outputs[:, inputs.input_ids.shape[1]:])
_dataset.submit_prediction(response)

model = AutoModelForCausalLM.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(model_path, revision=revision)
if use_torch_fp16:
model = model.half()
model.eval()
Expand All @@ -30,11 +30,14 @@ def run_single_pass(pytorch_runner, _dataset):


def run_pytorch_fp32(model_path, num_runs, timeout, dataset_path, **kwargs):
return run_pytorch(model_path, num_runs, timeout, dataset_path, use_torch_fp16=False)
return run_pytorch(model_path, num_runs, timeout, dataset_path)

def run_pytorch_fp16(model_path, num_runs, timeout, dataset_path, **kwargs):
return run_pytorch(model_path, num_runs, timeout, dataset_path, use_torch_fp16=True)

def run_pytorch_int8(model_name, batch_size, num_runs, timeout, dataset_path, revision, **kwargs):
return run_pytorch(model_name, batch_size, num_runs, timeout, dataset_path, revision=revision)


def run_pytorch_cuda(model_path, num_runs, timeout, dataset_path, **kwargs):
from transformers import AutoModelForCausalLM, AutoTokenizer
Expand Down
8 changes: 6 additions & 2 deletions natural_language_processing/text_generation/llama2/run.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,7 @@
from transformers import LlamaForCausalLM, AutoTokenizer


def run_pytorch(model_name, batch_size, num_runs, timeout, dataset_path, use_torch_fp16=False):
def run_pytorch(model_name, batch_size, num_runs, timeout, dataset_path, use_torch_fp16=False, revision=None):
def run_single_pass(pytorch_runner, _dataset):
input_tensor = tokenizer.encode(_dataset.get_input_string(), return_tensors="pt")
input_tensor = torch.cat([input_tensor for _ in range(batch_size)], 0)
Expand All @@ -18,7 +18,7 @@ def run_single_pass(pytorch_runner, _dataset):
np.random.seed(44)
torch.manual_seed(44)

model = LlamaForCausalLM.from_pretrained(model_name, torchscript=True)
model = LlamaForCausalLM.from_pretrained(model_name, torchscript=True, revision=revision)
model.eval()
if use_torch_fp16:
model = model.half()
Expand All @@ -42,6 +42,10 @@ def run_pytorch_fp32(model_name, batch_size, num_runs, timeout, dataset_path, **
def run_pytorch_fp16(model_name, batch_size, num_runs, timeout, dataset_path, **kwargs):
return run_pytorch(model_name, batch_size, num_runs, timeout, dataset_path, use_torch_fp16=True)

def run_pytorch_int8(model_name, batch_size, num_runs, timeout, dataset_path, revision, **kwargs):
return run_pytorch(model_name, batch_size, num_runs, timeout, dataset_path, revision=revision)


def main():
from utils.helpers import DefaultArgParser
llama_variants = ["meta-llama/Llama-2-7b-chat-hf", "meta-llama/Llama-2-13b-chat-hf"]
Expand Down