text
stringlengths 0
83
|
---|
@Switch01 |
A_Rog |
Aakanksha Agrawal |
Abhinav Sagar |
ABHYUDAY PRATAP SINGH |
abs51295 |
AceGentile |
Adam Chainz |
Adam Tse |
Adam Wentz |
admin |
Adrien Morison |
ahayrapetyan |
Ahilya |
AinsworthK |
Akash Srivastava |
Alan Yee |
Albert Tugushev |
Albert-Guan |
albertg |
Alberto Sottile |
Aleks Bunin |
Ales Erjavec |
Alethea Flowers |
Alex Gaynor |
Alex Grönholm |
Alex Hedges |
Alex Loosley |
Alex Morega |
Alex Stachowiak |
Alexander Shtyrov |
Alexandre Conrad |
Alexey Popravka |
Aleš Erjavec |
Alli |
Ami Fischman |
Ananya Maiti |
Anatoly Techtonik |
Anders Kaseorg |
Andre Aguiar |
Andreas Lutro |
Andrei Geacar |
Andrew Gaul |
Andrew Shymanel |
Andrey Bienkowski |
Andrey Bulgakov |
Andrés Delfino |
Andy Freeland |
Andy Kluger |
Ani Hayrapetyan |
Aniruddha Basak |
Anish Tambe |
Anrs Hu |
Anthony Sottile |
Antoine Musso |
Anton Ovchinnikov |
Anton Patrushev |
Antonio Alvarado Hernandez |
Antony Lee |
Antti Kaihola |
Anubhav Patel |
Anudit Nagar |
Anuj Godase |
AQNOUCH Mohammed |
AraHaan |
Arindam Choudhury |
Armin Ronacher |
Artem |
Arun Babu Neelicattu |
Ashley Manton |
Ashwin Ramaswami |
atse |
Atsushi Odagiri |
Avinash Karhana |
Avner Cohen |
Awit (Ah-Wit) Ghirmai |
Baptiste Mispelon |
Barney Gale |
barneygale |
Bartek Ogryczak |
Bastian Venthur |
Ben Bodenmiller |
Ben Darnell |
Ben Hoyt |
Ben Mares |
Ben Rosser |
Bence Nagy |
Benjamin Peterson |
Benjamin VanEvery |
Benoit Pierre |
Berker Peksag |
Bernard |
Bernard Tyers |
Bernardo B. Marques |
Bernhard M. Wiedemann |
Bertil Hatt |
Bhavam Vidyarthi |
Blazej Michalik |
Bogdan Opanchuk |
BorisZZZ |
End of preview. Expand
in Dataset Viewer.
nanoLLaVA - Sub 1B Vision-Language Model
Description
nanoLLaVA is a "small but mighty" 1B vision-language model designed to run efficiently on edge devices.
- Base LLM: Quyen-SE-v0.1 (Qwen1.5-0.5B)
- Vision Encoder: google/siglip-so400m-patch14-384
Model | VQA v2 | TextVQA | ScienceQA | POPE | MMMU (Test) | MMMU (Eval) | GQA | MM-VET |
---|---|---|---|---|---|---|---|---|
Score | 70.84 | 46.71 | 58.97 | 84.1 | 28.6 | 30.4 | 54.79 | 23.9 |
Training Data
Training Data will be released later as I am still writing a paper on this. Expect the final final to be much more powerful than the current one.
Finetuning Code
Coming Soon!!!
Usage
You can use with transformers
with the following script:
pip install -U transformers accelerate flash_attn
import torch
import transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
from PIL import Image
import warnings
# disable some warnings
transformers.logging.set_verbosity_error()
transformers.logging.disable_progress_bar()
warnings.filterwarnings('ignore')
# set device
torch.set_default_device('cuda') # or 'cpu'
# create model
model = AutoModelForCausalLM.from_pretrained(
'qnguyen3/nanoLLaVA',
torch_dtype=torch.float16,
device_map='auto',
trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(
'qnguyen3/nanoLLaVA',
trust_remote_code=True)
# text prompt
prompt = 'Describe this image in detail'
messages = [
{"role": "user", "content": f'<image>\n{prompt}'}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
print(text)
text_chunks = [tokenizer(chunk).input_ids for chunk in text.split('<image>')]
input_ids = torch.tensor(text_chunks[0] + [-200] + text_chunks[1], dtype=torch.long).unsqueeze(0)
# image, sample images can be found in images folder
image = Image.open('/path/to/image.png')
image_tensor = model.process_images([image], model.config).to(dtype=model.dtype)
# generate
output_ids = model.generate(
input_ids,
images=image_tensor,
max_new_tokens=2048,
use_cache=True)[0]
print(tokenizer.decode(output_ids[input_ids.shape[1]:], skip_special_tokens=True).strip())
Prompt Format
The model follow the ChatML standard, however, without \n
at the end of <|im_end|>
:
<|im_start|>system
Answer the question<|im_end|><|im_start|>user
<image>
What is the picture about?<|im_end|><|im_start|>assistant
- Downloads last month
- 136