It is a recommendation model for recommend mangas by using query(user requirement).
First, you have to pip install the necessary library.
!pip install sentence-transformers==2.2.2
!pip install torch==2.0.1
!pip install bing-image-urls==0.1.5
!pip install streamlit==1.10.0
!pip install numpy==1.21.6
or you can create requirements.txt and copy my requirements.txt in your file.
Then run this command in terminal or command prompt.
pip install -r requirements.txt
I just generate query by using Alpaca-LoRA and finetune it.
This is a code for generate querys.
I provided 2 scripts of finetuning code.
--fine_tune_eng.ipynb
--fine_tune_multi.ipynb
The first script used "all-MiniLM-L6-v" for being a pretrain model.
The second script used "paraphrase-multilingual-mpnet-base-v2" for being a pretrain model.
There are 2 finetuned models for encode the word to vector.
- Madnesss/fine-tune-all-MiniLM-L6-v2 : It is for English only.
- Madnesss/fine-tune-paraphrase-multilingual-mpnet-base-v2 : It is multilingual model.
from sentence_transformers import SentenceTransformer, util
path = "Madnesss/fine-tune-all-MiniLM-L6-v2" # or path = "Madnesss/fine-tune-paraphrase-multilingual-mpnet-base-v2"
model = SentenceTransformer(path)
#encode
embedding1 = model.encode(["example sentence1", "example sentence2", "example sentence3"], convert_to_tensor=True)
embedding2 = model.encode(["example sentence4", "example sentence5", "example sentence6"], convert_to_tensor=True)
#compute cosine sim scores
cosine_scores = util.cos_sim(embeddings1, embeddings2)
#Output the scores
print(cosine_scores) #tensor([0.1, 0,2, 0.3, ....])
Just clone my respository and use this command in terminal.
streamlit run Welcome.py
It will open in your browser.
I have written a blog for giving more details. You can find out here.