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

Jump to bottom

feat: add mps support #22

Open
wants to merge 5 commits into
base: main
Choose a base branch
from
Open

feat: add mps support #22

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
c99e9f1
automatically detect the Python version, if above 3.10 patch collecti…
Fodark Mar 13, 2024
5db8156
use transformers 4.38.1 instead of 4.38.2 as autocast fails on mps on…
Fodark Mar 13, 2024
494e622
add an util function to detect platflorm and suitable dtype
Fodark Mar 13, 2024
40ec649
load model on detected device and use correct dtype
Fodark Mar 13, 2024
4eac5b5
modify example in README to use the new device and dtype function
Fodark Mar 13, 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
Prev Previous commit
Next Next commit
load model on detected device and use correct dtype
  • Loading branch information
@Fodark
Fodark committed Mar 13, 2024
commit 40ec6491d4b3c0667e427622fb185878d3e98127
6 changes: 3 additions & 3 deletions cli_chat.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,7 +8,7 @@
import torch
from transformers import TextIteratorStreamer

from deepseek_vl.utils.io import load_pretrained_model
from deepseek_vl.utils.io import load_pretrained_model, get_device_and_dtype


def load_image(image_file):
Expand All @@ -34,13 +34,13 @@ def get_help_message(image_token):

@torch.inference_mode()
def response(args, conv, pil_images, tokenizer, vl_chat_processor, vl_gpt, generation_config):

_, dtype = get_device_and_dtype()
prompt = conv.get_prompt()
prepare_inputs = vl_chat_processor.__call__(
prompt=prompt,
images=pil_images,
force_batchify=True
).to(vl_gpt.device)
).to(vl_gpt.device, dtype=dtype)

# run image encoder to get the image embeddings
inputs_embeds = vl_gpt.prepare_inputs_embeds(**prepare_inputs)
Expand Down
4 changes: 3 additions & 1 deletion deepseek_vl/utils/io.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,10 +52,12 @@ def load_pretrained_model(model_path: str):
vl_chat_processor: VLChatProcessor = VLChatProcessor.from_pretrained(model_path)
tokenizer = vl_chat_processor.tokenizer

device, dtype = get_device_and_dtype()

vl_gpt: MultiModalityCausalLM = AutoModelForCausalLM.from_pretrained(
model_path, trust_remote_code=True
)
vl_gpt = vl_gpt.to(torch.bfloat16).cuda().eval()
vl_gpt = vl_gpt.to(device, dtype=dtype).eval()

return tokenizer, vl_chat_processor, vl_gpt

Expand Down
8 changes: 5 additions & 3 deletions inference.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@
from transformers import AutoModelForCausalLM

from deepseek_vl.models import VLChatProcessor, MultiModalityCausalLM
from deepseek_vl.utils.io import load_pil_images
from deepseek_vl.utils.io import load_pil_images, get_device_and_dtype


# specify the path to the model
Expand All @@ -11,7 +11,9 @@
tokenizer = vl_chat_processor.tokenizer

vl_gpt: MultiModalityCausalLM = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True)
vl_gpt = vl_gpt.to(torch.bfloat16).cuda().eval()

device, dtype = get_device_and_dtype()
vl_gpt = vl_gpt.to(dtype).to(device).eval()

conversation = [
{
Expand All @@ -32,7 +34,7 @@
conversations=conversation,
images=pil_images,
force_batchify=True
).to(vl_gpt.device)
).to(vl_gpt.device, dtype=dtype)

# run image encoder to get the image embeddings
inputs_embeds = vl_gpt.prepare_inputs_embeds(**prepare_inputs)
Expand Down