English
LlavaOLMoBitnet1B / llava /llava_olmo_explore.py
naveensp's picture
all the files required for inference
2ce0406
import torch
import sys
from OLMo_Bitnet_1B.model import OLMo
from OLMo_Bitnet_1B.config import ModelConfig
from OLMo_Bitnet_1B.configuration_olmo import OLMoConfig
from OLMo_Bitnet_1B.modeling_olmo import OLMoForCausalLM
import json
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, TextStreamer
import llava.model.language_model.llava_olmo1p58b as llava_olmo
import PIL
import torchvision
device = torch.device('cuda:5' if torch.cuda.is_available() else 'cpu')
torch.manual_seed(42)
with open('llava/config.json') as json_file:
data = json.load(json_file)
config_class = llava_olmo.LlavaOLMoBitnet1BConfig(**data)
# config_class = OLMoConfig(**data)
model = OLMoForCausalLM(config_class).to(device)
model.load_state_dict(torch.load('OLMo_Bitnet_1B/pytorch_model.bin'))
model.eval()
# tokenizer = AutoTokenizer.from_pretrained("NousResearch/OLMo-Bitnet-1B")
tokenizer = AutoTokenizer.from_pretrained(
"NousResearch/OLMo-Bitnet-1B",
cache_dir="./cache/",
model_max_length=1024,
padding_side="right",
pad_token_id=1,
unk_token='<|padding|>',
)
text = "Paris is a historic city with architectural marvels. It is also "
inputs = tokenizer(text, return_tensors='pt', return_token_type_ids=False).to(device)
# response = model.generate(**inputs, max_new_tokens=100, do_sample=True, top_k=50, top_p=0.95)
# llava olmo setup
image_tensor = torchvision.io.read_image('playground/data/LLaVA-Pretrain/images/00316/003163402.jpg')
lolmo = llava_olmo.LlavaOLMoBitnet1BForCausalLM(config_class).to(device)
lolmo.load_state_dict(torch.load('OLMo_Bitnet_1B/pytorch_model.bin'), strict=False)
response = lolmo.generate(inputs=inputs['input_ids'], max_new_tokens=100, do_sample=True, top_k=50, top_p=0.95)
print(tokenizer.batch_decode(response, skip_special_tokens=True)[0])