Install MiniG Locally - Long Context Model for Novel and Story Writing with Images

 This video shows how to locally install MiniG model which is trained on a synthesis dataset of over 120 million entries and has 1M token context window. It deals with both text and images.

Code:

conda create -n lm python=3.11 -y && conda activate lm

pip install torch

pip install git+https://github.com/huggingface/transformers

pip install git+https://github.com/huggingface/accelerate

pip install --upgrade sentencepiece

conda install jupyter -y

pip uninstall charset_normalizer -y

pip install charset_normalizer

jupyter notebook

pip install tiktoken torchvision

import torch

from transformers import (

    AutoModelForCausalLM,

    AutoTokenizer,

)

device = "cuda"

tokenizer = AutoTokenizer.from_pretrained("CausalLM/miniG",trust_remote_code=True)

query = "What is Happiness?"

inputs = tokenizer.apply_chat_template([{"role": "user", "content": query}],

                                       add_generation_prompt=True,

                                       tokenize=True,

                                       return_tensors="pt",

                                       return_dict=True

                                       )

inputs = inputs.to(device)

model = AutoModelForCausalLM.from_pretrained(

    "CausalLM/miniG",

    torch_dtype=torch.bfloat16,

    low_cpu_mem_usage=True,

    trust_remote_code=True

).to(device).eval()

gen_kwargs = {"max_length": 2500, "do_sample": True, "top_k": 1}

with torch.no_grad():

    outputs = model.generate(**inputs, **gen_kwargs)

    outputs = outputs[:, inputs['input_ids'].shape[1]:]

    print(tokenizer.decode(outputs[0], skip_special_tokens=True))

#===================

#For Images:

#==================

import torch

from PIL import Image

from transformers import AutoModelForCausalLM, AutoTokenizer

device = "cuda"

tokenizer = AutoTokenizer.from_pretrained("CausalLM/miniG", trust_remote_code=True)

query = 'Which lane should I drive in this image?'

image = Image.open("/home/Ubuntu/images/lane.png").convert('RGB')

inputs = tokenizer.apply_chat_template([{"role": "user", "image": image, "content": query}],

                                       add_generation_prompt=True, tokenize=True, return_tensors="pt",

                                       return_dict=True)  # chat mode

inputs = inputs.to(device)

model = AutoModelForCausalLM.from_pretrained(

    "CausalLM/miniG",

    torch_dtype=torch.bfloat16,

    low_cpu_mem_usage=True,

    trust_remote_code=True

).to(device).eval()

gen_kwargs = {"max_length": 2500, "do_sample": True, "top_k": 1}

with torch.no_grad():

    outputs = model.generate(**inputs, **gen_kwargs)

    outputs = outputs[:, inputs['input_ids'].shape[1]:]

    print(tokenizer.decode(outputs[0]))